View: session overviewtalk overview
Paper # 1, 12, 38, 77, 92, 139
09:00 | An Approach of Handling Verbal Inflections of Bengali Text: Conversion of Sadhu to Cholito Form of Language PRESENTER: Mohammad Masudur Rahman ABSTRACT. Bengali language follows Subject-Object-Verb (SOV) structure for constructing sentences where verb plays a crucial role. Machine translation for Bengali becomes easier if the verb is identified correctly in a semantically valid sentence. However, Bengali verbs are affected by various complex inflections while using in a sentence. Those inflections are determined according to tense (time of occurring event), person (nature of the subject), Sadhu and Cholito forms (historic evolution of writing styles of Bengali) of language. In this study, we propose an algorithm to identify different inflections of Bengali verbs to retrieve verb roots and present an approach to convert Sadhu to Cholito form of writing Bengali sentences. The methodology shows satisfying accuracy and fluency in the context of language evaluation. The source code along with relevant corpus and system output can be found at https://github.com/themasudur/Sadhu2Cholito. |
09:15 | A Complete Bangla Optical Character Recognition System: An Effective Approach PRESENTER: Tasnim Ahmed ABSTRACT. Bangla character recognition is a significant field of research because Bangla is the most widely-used language in the Indian subcontinent and 8th most used language for the written document. Research on Bangla character recognition has been started since mid 1980s. Different types of techniques have been already applied with performance review. This report describes a complete Optical Character Recognition (OCR) system for scripted Bangla characters. This paper proposes a new recognition system for scripted Bangla characters. We used convolutional Neural Network to build a model which identifies individual Bangla characters. From an image, the lines, words and characters are segmented based on empty space. We also used binarization, noise reduction and other techniques to improve accuracy. Finally, we have been able to achieve a very impressive accuracy rate in image to text conversion and by far the highest accuracy rate in scripted character detection. |
09:30 | Sentiment Analysis on Bengali Text using Lexicon Based Approach PRESENTER: Rajib Chandra Dey ABSTRACT. In this modern era, we daily involve in the internet strongly. We express our opinion about products, services, books, movies, songs, politics, sports, organizations, etc. through the internet in social media, blogs, micro-blogging websites or any media. Public opinion with Bengali text in internet media is increasing very rapidly. Due to a few works in Bengali text sentiment analysis, it has become an important issue of extracting opinions, emotions from Bengali textual data through Sentiment Analysis (SA) for better knowledge extraction. Sentiment Analysis (SA) is effectively used for classifying the opinion expressed in a text according to its polarity (e.g., positive, negative or neutral). This paper represents a lexicon dictionary-based approach for polarity detection of Bengali text data. We compared our proposed model with machine learning classifiers such as Decision Tree (DT), Naive Bayes (NB) and Support Vector Machine (SVM) classifiers and it works as a much better accurate model for Bengali text polarity detection. |
09:45 | A Continuous Space Neural Language Model for Bengali Language PRESENTER: Anisur Rahman ABSTRACT. Language models are generally employed to estimate the probability distribution of various linguistic units, making them one of the fundamental parts of natural language processing. Applications of language models include a wide spectrum of tasks such as text summarization, translation and classification. For a low resource language like Bengali, the research in this area so far can be considered to be narrow at the very least, with some traditional count based models being proposed. This paper attempts to address the issue and proposes a continuous-space neural language model, or more specifically an ASGD weightdropped LSTM language model, along with techniques to efficiently train it for Bengali Language. The performance analysis with some currently existing count based models illustrated in this paper also shows that the proposed architecture outperforms its counterparts by achieving an inference perplexity as low as 51.2 on the held out data set for Bengali. |
10:00 | Offline optical character recognition (OCR) method: An effective method for scanned documents PRESENTER: Md. Mujibur Rahman Majumder ABSTRACT. Optical Character Recognition (OCR) is a major computer vision task by which characters of image are detected and recognized by comparing to training set images. Process of detecting character is one of the perplexing tasks in computer vision. This is because of input image often not correctly aligned or because of noise. This paper presents a complete Optical Character Recognition (OCR) system which is worked for English character mostly for Calibri font. This system first corrects skew of image if input image is not correctly aligned followed by noise reduction from input image. This process is passed through line and character segmentation that are passed into the recognition module and recognize characters. By experimenting with a set of 50 images, average achievement is 92%, 98% is for Calibri font. Moreover, the developed technique is computationally efficient and requires less time than other Optical character recognition system. |
10:15 | Bangla License Plate Recognition Using Convolutional Neural Networks (CNN) PRESENTER: M M Shaifur Rahman ABSTRACT. In the last few years, the deep learning technique in particular Convolutional Neural Networks (CNNs) is using massively in the field of computer vision and machine learning. This deep learning technique provides state-of-the-art accuracy in different classification, segmentation, and detection tasks on different benchmarks such as MNIST, CIFAR-10, CIFAR-100, Microsoft COCO, and ImageNet [1, 2]. However, there are a lot of research has been conducted for Bangla License plate recognition with traditional machine learning approaches in last decade. None of them are used to deploy a physical system for Bangla License Plate Recognition System (BLPRS) due to their poor recognition accuracy. In this paper, we have implemented CNNs based Bangla license plate recognition system with better accuracy that can be applied for different purposes including roadside assistance, automatic parking lot management system, vehicle license status detection and so on. Along with that, we have also created and released a very first and standard database for BLPRS. |
Paper # 275, 277, 278, 312, 317, 324
09:00 | A High Figure of Merit Low Power LC VCO for D Band Applications PRESENTER: Nargis Akhter ABSTRACT. This paper presents a power efficient LC VCO model for D band wireless applications in 90 nm CMOS technology. An area efficient design of the proposed model is ensured by avoiding the requirement of additional capacitor for LC tank and varactor utilizing the parasitic capacitance of MOSs. The performance of the model is evaluated and its compatibility is analyzed using the process corner analysis, temperature swept analysis, Monte Carlo analysis and stability analysis in multifarious environments. An oscillation frequency from 163.25 GHz to 172.77 GHz is obtained with the applied tuning voltage ranged from 0.5 V to 1.1 V. Consuming a low DC power (2.82~2.96) mW, the proposed VCO provides relatively high output power on its differential terminal having the value of 15.53 dBm~6.43 dBm. The phase noise varies from -92.99 dBc/Hz to -75.81 dBc/Hz and the figure of merit has a value of -188.67 dBc/Hz at 164.39 GHz. Finally, comparing with the state-of-art VOC models, the presented model has demonstrated a better performance in terms of low DC power consumption, high output power and better figure of merit. |
09:15 | Effect of Human Blockage on an Outdoor mmWave Channel for 5G Communication Networks PRESENTER: Sumaiya Haq Aftabi Momo ABSTRACT. Millimeter-wave (mmWave)technology is going to be a fundamental part of the next-generation wireless communication for its potential to meet the ever-rising traffic demand and the bandwidth scarcity with its raw unused spectrum. However, channel modeling for mmWave is quite challenging owing to its susceptibility to blockage and weather elements and its requirement for line-of-sight (LOS) links for communication. In this paper, we proposed an extension of the channel simulator NYUSIM by introducing a new input parameter and observe the channel characteristics for several mmWave frequency bands (28, 38, 60, and 73 GHz) in a rural macrocell environment. A simple knife-edge diffraction model was used to simulate the human blockers in the channel model. The results include the effect of blocker density on various power delay statistics including received power, path loss, RMS delay spread, and path loss exponent (PLE). |
09:30 | Analysis of FBMC-OQAM over OFDM in Wireless Communication PRESENTER: Anamika Saha ABSTRACT. OFDM (Orthogonal Frequency Division Multiplexing) system has prevailed as one of the most eminent multicarrier technique in 4G wireless communication. The purpose of this paper is to overthrow the drawbacks of OFDM and investigate new candidate waveform Filter Bank Multicarrier (FBMC) which is more compatible for 5G wireless communication through the performance results of the comparison between OFDM and FBMC system. Because of achieving sharper PSD, better BER performance curve and lower PAPR need to discard Cyclic Prefix (CP) with the help of Filter Banks. To alleviate ISI (Inter-Symbol-Interference) and ICI (Inter-Carrier-Interference), this paper focuses the working procedure about OQAM processing, spreading frequency and extended IFFT/FFT and it reveals how the filter banks are responsible to improve the system performance of FBMC. Consequently, simulation based results of FBMC have exhibited that the desired solution evidently outperforms OFDM schemes in terms of Spectral Efficiency, FBMC Out Of Band (OOB) emissions declines from -31dB to -58dB. Next, in terms of Bit Error Rate (BER), 16 QAM FBMCperformance decays 27dB to approximately 20dB and 4 QAM FBMC have exhibited nearly 5dB lower than 4 QAM OFDM. Whereas, for Peak to Average Power Ratio (PAPR), OFDM CCDF performance curve drops from FBMC(16dB to 12dB). |
09:45 | Indoor Positioning Techniques using RSSI from Wireless Devices PRESENTER: Asif Ahmed Sohan ABSTRACT. The whole world is familiar with the Global Positioning System or GPS which can determine the exact location of any object with the help of satellite. But GPS signals are not available in indoors. To overcome this, Indoor Positioning System(IPS) is used which enables us to locate objects inside an indoor environment. Our goal is to build an Indoor Positioning System by estimating the location using Received Signal Strength Indication (RSSI) through wireless networks. The proposed model will determine the position of wireless devices in a room. We took the RSSI values as coordinates and specific reference points at every two meters making the room into a grid. The RSSI values on the reference point is measured. The position of the wireless devices will be estimated from the reference points using trilateration method and ITU indoor path loss model. With the aforementioned process we calculated the position of the wireless device using ITU indoor path loss model and trilateration. Using ITU indoor path loss model our mean error was 1.01166m, while using trilateration it was 1.22m. |
10:00 | An Efficient Spectrum Sharing Approach for Cognitive radio enabled Vehicular Cloud PRESENTER: Mohammad Mamun Elahi ABSTRACT. Vehicular ad hoc networks (VANETs) is an active area of research that enables road safety and infotainment applications for vehicular users. Although a significant amount of efforts is given in the literature, many non-trivial problems and challenges are still to be solved due to the inherent characteristics of vehicular networks like high mobility and dynamic topology change. Cognitive radio (CR) and cloud computing (CC) are two of the latest technologies used in vehicular networks to address those challenges. CR enables VANETs to deal with opportunistic spectrum management for utilizing the unused spectrum of licensed users by secondary users to solve the scarcity of spectrum. Vehicular cloud (VC) uses cloud computing to enable vehicles with excess storage or bandwidth to share the resources with other vehicles, who are willing to use that extra resource at some predetermined rate of price. One of the main challenges of cognitive radio-enabled vehicular ad hoc networks (CR-VANETs) is efficiently assigning wireless channels to the secondary users (vehicles) on demand so that maximum vehicles can communicate with the available spectrum without interference. Many approaches have already been proposed in the literature. Centralized algorithms suffer from lack of scalability and utilization of resources. On the other hand, distributed approaches need coordination and coverage issue. In this paper, we have developed a Cognitive Radio-enabled Vehicular Cloud Networking (CR-VCN) framework to assign channels in a distributed way to the vehicular cloud members to maximize efficiency in terms of overall network throughput and minimize interference among the users. Extensive simulation and analysis shows the effectiveness of the approach in terms of some performance measures. |
10:15 | Enhancing Security in Wireless Multicasting over κ − µ Fading Channel PRESENTER: Milton Kumar Kundu ABSTRACT. In this paper, a secure wireless multicast network over κ−µ fading channel is considered, where a source transmits secret information to a group of receivers. A group of eavesdroppers are also present in the network which are responsible for wiretapping the transmitted information. A perfect secrecy can be obtained if the eavesdroppers are not able to decode any information from the source. In order to investigate the secrecy capacity of the multicast network, at first a novel expression for Ergodic secrecy multicast capacity is derived in closed-form. The outage performance of the proposed network is analyzed obtaining the closed-form expressions of secure outage probability for multicasting. The effects of fading is determined and a beneficent insight is provided into investigating the secrecy performance of some other common fading scenarios such as Rician, Rayleigh, Nakagami-m, Gaussian etc. using the derived novel expressions of this paper. Monte-Carlo simulations are performed to validate the derived analytical expressions. |
Paper # 23, 80, 81, 91, 125, 177
09:00 | Performance evaluation of Proportional Fair and Round Robin Algorithm for Downlink Resource Schedulers at different User Mobilty PRESENTER: A.K.M. Fazlul Haque ABSTRACT. This paper presents an analytical study about two Resource Scheduling Techniques, Round Robin and Proportional Fair, for downlink LTE based on mobility. A Heterogeneous Network scenario has been considered to simulate the system model using Vienna LTE-A Downlink System Level Simulator v2.0 Q3 2018. In order to evaluate the performance, both stationary and moving users are considered. In terms of average UE (User End) throughput and average UE spectral efficiency, it is observed that, RR performs better in lower velocities while PF outperforms in higher velocities. A new switching technique suitable for the system has been proposed at the end of the paper. |
09:15 | Comparative Performance Analysis of K-Means and DBSCAN Clustering algorithms on various platforms PRESENTER: Nafi Shahriar ABSTRACT. While many data scientists are working hard just to improve a very fractional amount of performance, we wonder if there are any difference in performance of clustering among the platform we normally use. So we selected three datasets and perform K-means and DBSCAN clustering algorithm on the selected datasets in the four most popular platforms- Python, Matlab, R and Wolfram Mathematica. Then we summarized the results and compare the performances of different platforms. For performance metrics we used 2 criteria, first one is how much time it takes to execute clustering function and second one is the overall accuracy of clustering result. In our study we found that algorithm takes different execution time in different platform. Also variation was observed in terms of accuracy on various platform. Sometimes execution time was correlated with dataset size. Finally we suggest the platforms that are to be used for speed and for accuracy. |
09:30 | An Enriched Kasumi encryption algorithm by modifying F-functions for 4G LTE PRESENTER: Syeda Ishrat Mahjabin ABSTRACT. Long Term Evolution (LTE) has become most widespread Fourth Generation (4G) cellular networks for its high speed data rates. This new generation network provides more security and liability than previous generation mobile communication system. But this popular and secure telecommunication system is also affected by some cryptanalysis under some cases. So, it requires to use some security algorithms for users and network security purposes. Kasumi is one of the most popular security algorithms for LTE 4G. KASUMI is the block cipher that used F-function composed of FI, FL, and FO functions. The core of the block ciphers is function F. It creates the relationship between ciphertext and encryption key more complex by offering confusion property. In this paper, we have investigated the components of F-function of KASUMI to determine its cryptographic strength. We have proposed an enhanced Kasumi algorithm by modifying its functionalities. Finally, this paper shows the comparison in performances between original and improved Kasumi algorithms. |
09:45 | Applicability of Space Colonization Algorithm for Real Time Tree Generation PRESENTER: Rafsan Ratul ABSTRACT. Realistic tree modeling has been a subject of several studies but their application remains a computationally expensive task for real-time tree model generation. Space colonization algorithm is one of such method which utilizes competition of space between tree branches to generate visually realistic trees and bushes. In this paper, we study the applicability of space colonization algorithm for real-time procedural tree model generation for video games. We present optimal parameters for space colonization algorithm, several approaches to introduce variety in the generated tree models such as procedural generation of unique crown shapes, several point sampling techniques and procedural generation of branch segments while maintaining the visual appeal and acceptable generation time. |
10:00 | An efficient algorithm for reduce order modeling of discrete-time index-2 descriptor control systems PRESENTER: Mohammad Sahadet Hossain ABSTRACT. We have presented an efficient approach for reduce order modeling of discrete-time linear time-invariant (DT-LTI) index-2 descriptor systems which arise in the study of control system. This paper is a follow up and advancement of our previous work [1]. The contribution of this paper lies in finding the projection free model order reduction (MOR) strategy for the index-2 linear descriptor systems. Instead of reformulating the descriptor system into a generalized system as of [1], we propose a simpler and more efficient MOR strategy where balanced truncation can be applied only in the finite parts of the matrices. We derive an unique method analogous to the Smith iteration method in order to solve the generalized discrete-time algebraic Lyapunov equations (DALEs) without explicitly computing the projection matrices. Finally, we confirm the accuracy and better efficiency of our proposed algorithm with the help of results acquired from numerical simulations. |
10:15 | Parallelized RSA Algorithm: An Analysis with Performance Evaluation using OpenMP Library in High Performance Computing Environment PRESENTER: Md. Ahsan Ayub ABSTRACT. RSA algorithm is an asymmetric encryption algorithm used to maintain confidentiality and integrity of data as it is transported across networks. As time has gone on, security and confidentiality has grown in importance leading to more data requiring encryption. Parallelization has become an important aspect in improving the speed and efficiency of processing for encryption algorithms. Improvements in parallel implementations of the RSA algorithm lead to better security and efficiency for parallel systems utilizing the algorithm. In this study, we present a comprehensive survey of methods proposed by researchers for parallelization of the RSA algorithm from 1978 till date. This survey aims to provide a deeper understanding of the possible avenues that can be considered to obtain better performance of the algorithm in parallel environments. To demonstrate the improvements, this paper presents a parallel CPU-based implementation of the RSA algorithm using the OpenMP library. This implementation focuses on parallelizing the exponentiation operation of the algorithm. To provide a robust analysis, the study makes use of a High Performance Computing Environment to illustrate results for different scenarios in terms of parallel processing units. Through experimental analysis, the implementation is shown to have greatly improved execution times when compared against serial implementation. |
Paper # 89, 150, 183, 332, 340, 346
09:00 | Deep Learning based Ensemble Method for Household Energy Demand Forecasting of Smart Home PRESENTER: Saidur Rahman ABSTRACT. Electricity/Energy demand forecasting enables efficient electricity distribution through the use of smart grid. For the construction of such devices, we need to equip our homes with smart electricity probing devices that can record the electricity usage of our homes. When multiple homes are equipped with such devices, the aggregate electricity usage data can be analyzed and consequently future electricity demand of a city can be predicted through the use of machine learning models on the data. As a result, electricity distribution can be adjusted according to consumer needs through the use of smart grid technology. In this paper, household electricity consumption data of a single household has been analyzed. Exploratory Data Analysis (EDA) is carried out on the data, time-series analysis is performed and time-series forecasting models such as Autoregressive Integrated Moving Average (ARIMA) model is used to make electricity demand predictions. A Recurrent Neural Network (RNN) model with Long Short Term Memory (LSTM) units has also been trained using the data. Multi-variate and uni-variate linear regression models have been developed using the dataset. Finally, to obtain a more reliable and accurate household energy consumption prediction model, a Mahalanobis distance based ensemble is created out of the 4 aforementioned models. |
09:15 | A Machine Learning Approach to Predict Vulnerability to Drug Addiction PRESENTER: Arif Shahriar ABSTRACT. There are significant amount of differences between an addicted and non-addicted person on their social and familial behavior. In this paper we tried to find out the characteristics of a person related to his social and familial life and also health issues that can prove his vulnerability to drug addiction. The research was held on the context of the people of Dhaka, Bangladesh and on an age group of 15 to 40 years. We have collected and analyzed data associated with addicted and non-addicted people from different areas of Dhaka. For addicted person’s data we reached some rehabilitation center of Dhaka and for non-addicted person’s data we communicated different aged group people of different colleges and universities. A machine learning approach then helped us to find out some features that differentiate between these two groups of people. This will help people to understand if a person is going to be addicted or not based on their health issues and social and familial behavior. Our questionnaire was constructed on the basis of Addiction Severity Index with the help of psychologists and specialists on drug addiction. |
09:30 | Prognosis and treatment prediction of Type-2 Diabetes using Deep Neural Network and Machine Learning Classifiers PRESENTER: Md. Kowsher ABSTRACT. Type 2 Diabetes is a fast-growing, chronic metabolic disorder due to imbalanced insulin activity. As lots of people are suffering from it, access to proper treatment is necessary to control the problem. Most patients are unaware of health complexity, symptoms and risk factors before diabetes. The motion of this research is a comparative study of seven machine learning classifiers and an artificial neural network method to prognosticate the detection and treatment of diabetes with a high accuracy, in order to identify and treat diabetes patients at an early age. Our training and test dataset is an accumulation of 9483 diabetes patients’ information. The training dataset is large enough to negate overfitting and provide for highly accurate test performance. We use performance measures such as accuracy and precision to find out the best algorithm deep ANN which outperforms with 95.14% accuracy among all other tested machine learning classifiers. We hope our high performing model can be used by hospitals to predict diabetes and drive research into more accurate prediction models. |
09:45 | A Modified Support Vector Machine with Hybrid Kernel Function for Diagnosis Disease PRESENTER: Farjana Akter Tania ABSTRACT. Support Vector Machine is one of the most popular, promising and well-known machine learning tools for classification, regression and novelty detection. The performance of SVM totally depends on its core component of kernel functions like linear, polynomial and radial basis function (RBF). Conventional kernel functions have advantages and drawbacks, so it is very difficult to give the excellent performance while using only one kernel function for the large nonlinear dataset to nonlinear mapping from original input space to a high dimensional space. For this reason, in this paper, we proposed a hybrid kernel function combining linear and RBF kernel function and data preprocessing algorithm to reduce the noisy data for increasing learning ability and generalization performance. These two factors jointly improve the performance of SVM. In order to validate our proposed model, we used two different nonlinear breast cancer datasets and heart disease datasets with different characteristics. By comparing the simulation results, the results obtained from the proposed model are better than other SVMs constructed by ordinary kernel function and existing methods. |
10:00 | Design and Development of a Benchmark for Dynamic Multi-objective Optimisation Problem in the Context of Deep Reinforcement Learning PRESENTER: Md Mahmudul Hasan ABSTRACT. Different benchmarks have played an important role in analysing algorithms for dynamic multi-objective optimisation problems. According to the literature, there are several benchmarks to deal with the dynamic multi-objective optimisation problems, especially in the evolutionary approaches. In this study, a comprehensive review has been done regarding the existing benchmarks in the single objective and multi-objective reinforcement learning (MORL) settings. To the best of our knowledge, there is no benchmark in the context of dynamic multi-objective reinforcement learning (DMORL). Therefore, this study has addressed this gap by applying the existing knowledge to propose a benchmark which may help to investigate the performance of different algorithms. It can also support to understand the dynamics while objectives are conflicting with each other and deal with the constraints and problem parameters that change over time. The proposed benchmark is the modified version of the deep-sea treasure hunt problem where several features such as changing parameters and objectives have been integrated to support the dynamics in a multi-objective environment. This paper highlights the methodology part of designing and developing a benchmark. |
10:15 | A Belief Rule Based Expert System to Predict Student Performance under Uncertainty PRESENTER: Sohrab Hossain ABSTRACT. Student performance analysis is necessary and important for evaluating the skill and efficiency of students in different levels. To build a prospective institution, it requires continuous evaluation of the student performance as well as other stack holders. Due to uncertainty and other information gaps, most of the times evaluation process biased the outcomes. This paper represents an approach that can easily handle the uncertainty. In this research, applied Belief Rule Base a knowledge and Evidential Reasoning approach is applied for our evaluation. Our expert systems output is more reliable and accurate for evaluating student performance with considering personal and institutional parameters. |
Paper # 30, 64, 116, 266, 269, 296
09:00 | Authorship Identification of Source Code Segments Written by Multiple Authors Using Stacking Ensemble Method PRESENTER: Parvez Mahbub ABSTRACT. Source code segment authorship identification is the task of identifying the author of a source code segment through supervised learning. It has vast importance in plagiarism detection, digital forensics, and several other law enforcement issues. However, when a source code segment is written by multiple authors, typical author identification methods no longer work. Here, an author identification technique, capable of predicting the authorship of source code segments, even in case of multiple authors, has been proposed which uses stacking ensemble classifier. This proposed technique is built upon several deep neural networks, random forests and support vector machine classifiers. It has been shown that for identifying the author-group, a single classification technique is no longer sufficient and using a deep neural network based stacking ensemble method can enhance the accuracy significantly. Performance of the proposed technique has been compared with some existing methods which only deal with the source code segments written exactly by a single author. Despite the harder task of authorship identification for source code segments written by multiple authors, our proposed technique has achieved promising results evident by the identification accuracy, compared to the related works which only deal with code segments written by a single author. |
09:15 | An Integrated CNN-RNN Framework to Assess Road Crack PRESENTER: Tawsin Uddin Ahmed ABSTRACT. Road crack detection and road damage assessment are necessary to support driving safety in a route network. Several unexpected incidents (e.g. road accidents) take place all over the world due to unhealthy road infrastructure. This paper proposes a deep learning approach for road crack detection and road damage assessment which will contribute to the transport sector of a country like Bangladesh where a plethora of roads undergo the crack problem. The proposed model consists of two phases. In the first phase, the model is trained using transfer learning (VGG16) to detect the existence of crack on the road surface. In the second phase, an integrated framework, combining CNN(VGG16) and RNN(LSTM), is trained to classify the crack in one of the two categories-severe and slight. After experiments, the validation accuracies obtained by the prosposed models (VGG16 and VGG16-LSTM) are respectively 99.67% and 97.66%. |
09:30 | Malicious Nodes Detection based on Artificial Neural Network in IoT Environments PRESENTER: Mirza Akhi Khatun ABSTRACT. The central promise of the Internet of Things (IoT) is to accelerate the interaction with surroundings. Smartwatch, smart bulbs, thermostats, fitness tracker, smart cars are now being controlled with the help of various embedded devices that operate with little to no human interaction. These embedded devices, however, bring forth several security changes as most manufacturers still prioritize performance and production over security. This inherent flaw has manifested itself in the form of various attacks including the Distributed Denial of Service (DDoS) where malware prepared with the aid of malicious nodes in IoT devices are used to carry out the attack. It is, therefore, essential to comprehend the nature of these attacks and swiftly identify infected devices to combat this situation. Machine Learning, or more particularly a branch of this technique commonly known as Deep Learning, has already demonstrated its outstanding potentials while administering with the heterogeneous data of diverse sizes. Using the Artificial Neural Network (ANN), this paper suggests a method of detecting malicious nodes in IoT environments. The contribution of the paper is in two-fold. First, it classifies the usual and malicious patterns of IoT devices in the network, and second, describes a scheme for successfully detecting malicious nodes with 77.51\% accuracy. |
09:45 | Agglomerative Clustering of Handwritten Numerals to Determine Similarity of Different Languages PRESENTER: Md. Rahat-Uz- Zaman ABSTRACT. Handwritten numerals of different languages have various characteristics. Similarities and dissimilarities of the languages can be measured by analyzing the extracted features of the numerals. Handwritten numeral datasets are available and accessible for many renowned languages of different regions. In this paper, several handwritten numeral datasets of different languages are collected. Then they are used to find the similarity among those written languages through determining and comparing the similitude of each handwritten numerals. This will help to find which languages have the same or adjacent parent language. Firstly, a similarity measure of two numeral images is constructed with a Siamese network. Secondly, the similarity of the numeral datasets is determined with the help of the Siamese network and a new random sample with replacement similarity averaging technique. Finally, an agglomerative clustering is done based on the similarities of each dataset. This clustering technique shows some very interesting properties of the datasets. The property focused in this paper is the regional resemblance of the datasets. By analyzing the clusters, it becomes easy to identify which languages are originated from similar regions. |
10:00 | 3DIGARS-PSP: A Novel Statistical Energy Function and Effective Conformational Search Strategy based ab initio Protein Structure Prediction PRESENTER: Avdesh Mishra ABSTRACT. To solve protein structure prediction (PSP) problems computationally, a plethora of template-based methods exist. However, there are very few ab initio models for PSP. Template-based modeling relies on the existing structures and therefore is not effective for non-homologous sequence-based structure prediction. Thus, ab initio modeling is indispensable in such cases, even though it is a challenging optimization problem. To cope, we utilize an effective energy function (called 3DIGARS) and an advanced search algorithm (called KGA) based ab initio PSP, called 3DIGARS-PSP. To address critical search, the proposed genetic algorithm deploys two effective operators: angle rotation and segment translation. Further, propensities of torsion angle and secondary structure distribution have been utilized to guide the conformation search. Crucial features, such as sequence-specific accessibility, hydrophobic-hydrophilic properties and torsion angles of protein residues are mined to formulate an optimized energy function, which is then combined with the advanced sampling algorithm to explore critical conformational space. Consequently, 3DIGARS-PSP performed well compared to the state-of-the-art method for a set of low TMscore models from CASP data. |
10:15 | A Machine Learning Approach to Extract Keyphrases from Bengali Document using CNN-Bidirectional LSTM PRESENTER: Nishat Tasnim Ahmed Meem ABSTRACT. Keyphrases are single or multiple word phrases of a document which describe the principal topics of that document. These keyphrases help readers to get an overview of the document. In this paper, we proposed a system that uses the combination of Convolutional Neural Network and Bidirectional Long Short-Term Memory (BiLSTM) Recurrent Neural Network (RNN) to automatically detect keyphrases from a document. We also used some preprocessing steps to clean and generate candidates keyphrases to train the model. Convolutional Neural Network can analyze semantic meanings of sentences. Bidirectional LSTM can learn the relations among words in the sentences. A Bengali pre-trained word embedding is used in this work. |
Light Snacks and Tea/Coffee will be distributed
Prof. Dr. Celia Shahnaz, Chair, IEEE Bangladesh Section and other members of the IEEE Bangladesh Section Executive Committee will conduct a session on activities of IEEE as well as prospects and benefits of joining at IEEE
Invited Talks 2-4
Prof. Dr. Yasushi Makihara
Osaka University, Japan
Talk Title: On Gait Attributes: Age, Gender, and Attractiveness by Video-based Gait Analysis
Prof. Dr. Md. Atiqur Rahman Ahad
University of Dhaka, Bangladesh
Osaka University, Japan
Talk Title: Vision- and Sensor-based Activity Analysis: Future Scopes
Prof. Dr. Atsushi Inoue
Eastern Washington University, USA
Talk Title: Computer Vision for Artificial Intelligence~ Pivotal Roles of Computing with Words
Paper # 13, 66, 109, 154, 293, 353
11:00 | Automatic Individual Information Aggregation using Publicly Available Social Media Data PRESENTER: Mir Tafseer Nayeem ABSTRACT. In this paper, we make use of freely available data on public social media sites in two ways. First, we develop a search application capable of aggregating information about an individual, using only their name as input. Second, we investigate the feasibility of mining public data and linking this information across multiple social media sites in an attempt to produce an information profile for an individual. The inspiration of our work is to allow a person to see firsthand how much information about them exists online, and how this information could potentially compromise their privacy whereas the objective of our research is to analyze the feasibility of building a profile by gathering and linking information of an individual across different social media sites. This is done in the hope of perhaps inspiring an individual pay additional heed to the privacy settings on social media accounts, and to be more vigilant about what information they choose to share online. |
11:15 | Classification of Depression, Internet addiction and Prediction of Self-esteem among University Students PRESENTER: Anika Islam Apsara ABSTRACT. Machine learning is massively used in the prediction of cognitive and psychological features in recent times. This research aims to find the predictability between leading disorders like Internet addiction, depression and low self-esteem. For this purpose, 461 undergraduate students have been selected arbitrarily from several educational institutions of Dhaka city and voluntarily completed a standard questionnaire which was prepared based on the self-reported measures concerning the aforementioned disorders. Different standard psychometric scales such as Internet Addiction Test (IAT) by Dr. Kimberly Young, Self-esteem Scale by M. Rosenberg, PROMIS Emotional Distress Depression short scale by PROMIS Health Organization have been used in the correlational survey. The internal consistency of the data has been proven by Cronbach alpha. Subsequently, Shapiro-Wilk Normality test revealed the data to be non-parametric. To reduce redundancy from the data, important features has been extracted using minimum-redundancy-maximum-reduction (mRMR) and Chi-square test and prediction model has been created from them using Logistic Regression, Naive Bayes, Random Forest, C4.5 Decision Tree and k-Nearest Neighbors. After applying these methods, the result shows that Internet addiction and depression are interconnected with self-esteem and thereby prediction models can be built to reduce the severity of these disorders. |
11:30 | Performance Analysis of NB Tree, REP Tree and Random Tree Classifiers for Credit Card Fraud Data PRESENTER: Nayan Kumar Rudro ABSTRACT. Credit card fraud has been growing tremendously mainly in recent years. For this inconceivable occurrence, along with the financial losses of banks, companies, NGOs and personal accounts, the reputation of the organizations have gone through at a stake. In this respect, detecting fraud, accuracy has become very important to avoid harassing innocent customers. In this paper, we elicit three tree-based classifiers (NB Tree, REP Tree, and Random Tree) regarding the German credit card data set. We have combined this tree classifier with ensemble techniques to get better accuracy. Using the Resampling, Random Committee and Logit Boost, we were able to increase the performance of those classifiers. |
11:45 | Analysis of Different Predicting Model for Online Shoppers’ Purchase Intention from Empirical Data PRESENTER: Md. Rayhan Kabir ABSTRACT. The recent outburst of online shopping, commonly known as e-shopping, has added a new dimension in the business sector. In these days, people tend to explore online for finding the items they need and buy through online transaction. It has made their life more easy and comfortable. At the same time, it has become a great need for sellers to know the patterns and intentions of different types of online customers. The customer’s purchase intention can be predicted by analyzing the history of the customers. In this study, we have analyzed the empirical data of online shoppers for building a better prediction model to predict their purchase intention. We have analyzed different classification algorithm such as Decision Tree, Random Forest, Naive Bayes, SVM to predict whether a customer, visiting the webpages of an online shop, will end up with a purchase or not. We have also performed some ensemble methods to boost up the performance of these algorithms. Our study has shown that Random Forest is most suited to predict the customer’s purchase intention. Moreover, if we choose to do gradient boosting using this algorithm, it can predict with the highest accuracy, which is 90.34%. Considering the application of ensemble methods to this dataset makes this study unique. |
12:00 | Comparison of the efficiency of Machine Learning algorithms on Twitter Sentiment Analysis of Pathao PRESENTER: Mahmudul Islam Sajib ABSTRACT. Now a days, we are virtually living on social media, constantly sharing views and opinions regarding things that associate with us. As a result, social media sites are getting flooded with data in the form of opinions, especially on Twitter. However, all these tweets or opinions can be mined and used in various aspects of life and business for the betterment of human beings as a whole. Opinions on social media are unstructured and scattered and needs to be organized properly so that it contains valuable information regarding a specific subject. Sentiment analysis plays a big role here. In this paper, we have presented a way to automatically analyze the sentiment of tweets posted by Pathao (popular ride sharing service in South-East Asia) users. We have merged data mining with text mining and computational intelligence to label the tweets as negative or positive. We have used Twitter API to collect dataset from Twitter and followed the supervised learning approach for the development of training corpus. For text classification, we have trained the dataset and feed into two different machine learning algorithms to get the polarity of those tweets which is either negative or positive. We also made a comparison of the efficiency two algorithms. Our proposed method is showing effective result and outcomes are satisfactory. |
12:15 | Embedding with Pixel Repetition for Reversible Data Hiding PRESENTER: Hussain Nyeem ABSTRACT. This paper presents a new Pixel-Repetition (PR) based embedding scheme to better utilize the redundancy of image data for a high-capacity Reversible Data Hiding (RDH) scheme. Each pixel of an input image is repeated to create an image-block of size 2x2 in the up-sampled image. The proposed embedding scheme maps the up-sampled pixels using a set of unique conditions to embed 4-bit of data in each 2x2 block. The mapping conditions also ensure the regeneration of the first-pixel of a block from the other pixels in the block, which allows embedding of additional data-bits in every first-pixels of a block. Unlike the existing schemes, the proposed scheme with the modified mapping conditions and embedding in the first pixel in each block ensures better utilization of the redundancy. Thus, the proposed scheme demonstrates a significant improvement in the embedding rate-distortion performance over the relevant RDH schemes. |
Paper # 17, 46, 53, 98, 228, 314
11:00 | Control Room Software and Hardware Development for a Radio Telescope PRESENTER: Kala Meah ABSTRACT. Control room software communicates with the back-end and front-end software to maintain the appointments and database. Control room hardware consists of motor driver, PLC, power supplies, and sensor interfaces. Control room software controls the PLC operations to move the radio telescope at the appropriate locations using azimuth and elevation motors and receives astronomical data from the SpectraCyber. This paper describes the control room software and hardware development for a radio telescope as well as the laboratory testing of the complete system. |
11:15 | Improved Embedded Vehicle Safety System With STM32 PRESENTER: Tushar Rahman Khan ABSTRACT. According to the present situation, road accident is a dangerous threat of our day to day life. We don’t feel safe while travelling because of this dangerous problem. To confront this issue, we have come up with a real-time vehicle safety system which can reduce the rate of the accident using an emergency braking system and also can be used to reduce the problem of traffic jam. The concept of embedded system is used to build the module. An STM32 IC is used along with different sensors like IR Sensor, Sonar Sensor. The sensors detect the presence of obstacles in front of the vehicle and send the sensed value to the STM32 IC. According to that, the STM32 IC activates the braking system. A GSM module is also connected to the STM32 IC to find the alternative route when the road is highly crowded. We also used Arduino instead of STM32 to check the difference in the performance of the module. We observed that the accuracy of braking and object detection system is 100 percent for the device with STM32 whereas that is 60 percent for the device having Arduino.There is a noticeable difference in terms of load-carrying capacity and the device with STM32 leads in that case also. By doing so we come up with a conclusion that STM32 is better than Arduino. A Motor Driver Shield (l293d) is used for both STM32 and Arduino to control the motors of the car. For controlling the car, an application named Bluetooth RC controller is used for both the module with STM32 and Arduino. |
11:30 | Design a Real Time Temperature Monitoring System to Protect DC Motor PRESENTER: Naeemul Islam ABSTRACT. This paper illustrates a solution to protect the motor from overheating. There are many explanations for motor failure and one of them is overheating. Motor overheating causes mechanical damage and reduces lifespan. A real-time monitoring and data logging system has developed to reduce the percentage of damage and the rate of motor failure, improving the reliability of motor operation. The system has been developed using Arduino UNO and Lab VIEW software. Using a computer, this scheme allows for real-time surveillance and information logging opportunities. It also indicates the condition of the motor at the time of running with an alarming system and trips the motor when overheated. The advanced system is superior in terms of lower power consumption, adjustability, the capability of detecting the heating condition and actuating system accessibility. Practical operation shows that the system is stable, reliable, and simple to run and has the function to satisfy the environmental requirements. |
11:45 | DiChokro: An Anti-Theft System for two wheelers PRESENTER: Md. Rabiul Ali Sarker ABSTRACT. Bicycle theft has increased in developing countries in recent years. Being light and easy to hide, a stolen bicycle is often difficult to search. It has, therefore, become a pressing need to develop a low cost, easy to use solution to track the bicycles. DiChokro, proposed in this work, is a solution to that widespread problem of Bicycle theft and has two major components a device and an android application-based tracking facility that can be availed through any smartphone. The device contains a GPS Module that sends the location of the cycle to the cloud, highly sensitive vibration sensor and a processor. Users can search for the secured parking locations, track their parked bicycle through the android application that is connected to the device through the cloud. The vibration sensor installed in the device helps users to get informed if someone attempts to steal the bicycle. The proposed solution is very cheap (<$30) and will be able to address the issue of bicycle theft. |
12:00 | Segmentation of Arabic Letters on Dot Matrix Play Board PRESENTER: Md Mohiuddin Soliman ABSTRACT. Because of the immense necessity of displaying short messages on electronic devices the segmented display system has become greatly popular. Numerous research activities are done to represent English and Bengali letters on segmented display. On the other hand, no research work is found till now for Arabic letters. This paper brings an accurate and systematic segmentation method for Arabic letters. The points covered by each segment, the segments needed for each letter, the truth table for the letter, simplified expressions and the logic circuits of the segments are given. In this paper, we have generated all the Arabic letters with understandable and proper shapes. Also, the segment-to-letter ratio is found to be less comparatively. |
12:15 | An Arduino Based Robot for Pipe Surveillance and Rescue Operation PRESENTER: Koushik Ahmed ABSTRACT. Pipe surveillance and rescue robot is a type of automation that is used for inspection and rescue. In this paper, an Arduino based robot is designed and implemented for mainly pipe inspection and rescue operation. It can be used for cleaning and welding, as well. A prototype is built which can move through a pipe with a diameter of 8-20 inches. The architecture of this robot is simple and effective. It has a cylindrical body with adjustable legs and changeable shaft. As a result, it can easily pass through both straight and curved pipes. According to the needs, the shaft of the robot can be changed. By using a motor, a mechanical shaft is also designed which can grab things during rescue operation. The robot can carry a weight up to 25 kg. The robot is user-controlled. It has an interface with the user through a camera. It is a wire-based device. Therefore, energy is supplied to the robot from outside. |
Lunch Packets will be distributed
Paper # 99, 103, 143, 179, 200, 245
14:00 | Time delay coordinate based Dynamic Mode Decomposition of a compressible signal PRESENTER: Mohammad Niaz Murshed ABSTRACT. Dynamic Mode Decomposition (DMD) is a data-driven modeling technique used to extract dynamic features in a complex physical system. We review the algorithm and establish its connection to the Koopman operator. DMD fails to work on problems where the data has a highly oscillatory flow. Time delay coordinates have been used as a modification in the DMD algorithm and is tested on data acquired from a compressible signal. This version of DMD perfectly captures the dynamics and results in a reliable model for future prediction. |
14:15 | A Cost-Effective Approach to Transmit Data Through Visible Light PRESENTER: Noushad Hossain ABSTRACT. This paper presents a study on how optical wireless communication (OWC) can be used as a new form of data transmission in small distances and a solution for current radiofrequency spectrum shortage. OWC has a major drawback of not having the proper light source that can be used for modulation. Persisting light sources do not have the flexibility for implementation of complex modulation processes. These are now replaced by more efficient and mobile source like light emitting diode (LED). LEDs can be used for data transmission in small distances which can be increased using the proper lens. In this study, LED is used to build a low cost communication system via visible light with simple modulation technique using Raspberry Pi which is known as visible light communication (VLC) system. And we believe that it is one of the first instances of VLC system using Raspberry Pi. This approach tends to find the simplest way to transmit data using a light source. Data transfer rate can be increased by using other modulation techniques and adding more levels to it. Hence, an initiative is taken to develop a better data transfer process with higher transmission speed (using Raspberry Pi). The transmission distance can be increased using more coherent light sources and efficient receiver. |
14:30 | Double Coated VGG16 Architecture: An Enhanced Approach for Genre Classification of Spectrographic Representation of Musical Pieces PRESENTER: Partha Protim Das ABSTRACT. In the field of Music Information Retrieval(MIR), one of the most demanding task would be music genre classification. The search for automated categorization of music on the basis of genre yielded some diverse techniques, prominent ones of them rely on machine learning, and recently deep learning. In this paper we approached this very problem with Convolutional Neural Network(CNN), which has already been proven very useful in categorizing images in the past few years. In this work, some of the popular CNN architectures including ResNet, VGG etc., are experimented with, of which a version of the latter-the VGG16 has outperformed the rest by a handsome degree. On top of the vanilla VGG16 architecture, we coupled that with two layers of additional dense layers tuned for this specific task. We used audio clips of each genre, in this case ten genres, and got the spectrogram images of them and fed them into the network. In this approach we were able to achieve 84% accuracy, thus providing a promising stance in the genre classification problem. |
14:45 | Energy Efficiency Investigation in Massive MIMO 5G System using Nakagami-m Fading Channel PRESENTER: Tamanna Hossain ABSTRACT. Massive multiple-input-multiple-output (MIMO) is considered as one of the major technologies for 5G communications, implementing huge elements of antenna array to boost spectral efficiency, robustness, and reliability. However, massive deployment of MIMO antenna elements also increases power consumption, which ultimately cannot fulfill another important design criterion of 5G communications, i.e., energy- efficiency. In addition, huge deployment of closely spaced antenna also increases co-channel interference. Nakagami-m fading channel has the $m$ diversity order to minimize interference in different fading distributions. Based on base station antennas distribution, massive MIMO can be classified as co-located, distributed, and hybrid. In this paper, we investigate the energy efficiency of massive MIMO systems for different distributions using MMSE (minimum mean square error) detector in Nakagami-m fading channel. Simulation results show high energy efficiency can be exploited in hybrid system compare to other two MIMO distribution. Moreover, the trade-off between spectral efficiency and energy efficiency with signal-to-noise ratio (SNR) is also portrayed for different distribution. |
15:00 | A New Adaptive Discontinuous Reception (DRX) Model for Power Saving in LTE System PRESENTER: Mahmud Matin ABSTRACT. LTE that is a standard for wireless broadband communication for cell phones and information terminals, has incorporated Discontinuous Reception (DRX) power-saving strategy to enhance energy-saving at the user equipment (UE). But for the networks that have different channel quality, it is tough at all the time for fixed DRX parameters to gain the desirable achievement. To save energy, DRX model parameters need to be optimized efficiently for minimizing power consumption based on CQI (Channel Quality Indicator) reporting by UE and MCS (Modulation and Coding Scheme) assignments by eNodeB as networks channel quality are not equal. This research develops a new model based on CQI reporting that will reduce energy consumption by eliminating unnecessary wake-ups of UE named as CQI DRX. The optimization is done by configuring DRX parameters depending on MCS assignments by eNodeB. The simulation process is carried out using ns3 a discrete event network simulator as energy consumption can be decreased by approximately 17% compared with the traditional static DRX that is used in stable networks. |
15:15 | Millimeter Wave Channel Modeling for Aviation Backhaul Networks in 5G Communications PRESENTER: Hussnin Binte Hamid Dutty ABSTRACT. The emergence of bandwidth-hungry mobile applications and services has led to a massive explosion of the mobile data traffic. In order to reduce this issue, millimeter Wave communications (mmWave) systems are a promising technology for future 5G systems avail to the large amount of bandwidth. The backhaul network (base station to core network) is generally supported by optical fiber technology. MmWave is nowadays used as an alternative backhaul technology removing the barrier of optical fiber, e.g., high deployment cost and difficult to install in hard to reach areas. In an airport scenario of fifth generation (5G) network, mmWave channel can be useful for backhauling huge mobile traffic. However, mmWave suffers from absorption loss, path loss, and atmospheric attenuation. In this research, we investigate channel modeling in aviation backhaul networks. Simulation results show the directional and omnidirectional power delay profiles, path loss for both V band and E band in case of light-of-sight (LOS) and Non-LOS scenarios. |
Paper # 28, 135, 203, 289, 337, 350
14:00 | ABYS(Always By Your Side): A Virtual Assistant for Visually Impaired Persons PRESENTER: Md Rakibuz Sultan ABSTRACT. Visually impaired persons do not have either a perfect or near-perfect vision. That’s why they face so many problems in their day-to-day tasks. In this work, we reviewed the current state-of-the-art in assistive technology using voice recognition and developed a voice-controlled desktop application model on Microsoft Windows platform especially for the visually impaired persons to reduce their hassle for using a computer and accessing home appliances. In addition, even non visually impaired people can also use the facilities provided by the application and use it as a virtual-assistant. The application is user-independent and human voice commands are given to the application in English Language. This application model enables the user to control installed applications, send and receive e-mails or perform internet browsing. The model is also capable of reading a news portal and also allows the users accessing their home appliances over voice commands |
14:15 | IoT Based Vehicle Monitoring with Accident Detection and Rescue System PRESENTER: Mohammed Abdul Kader ABSTRACT. Road accident has turned out to be one of the serious nuisances in Bangladesh. The main reason behind these accidents is the carelessness and reckless driving of the drivers. In this paper, an IoT based system is proposed to ensure the continuous monitoring of the parameters of a running vehicle like vehicle speed, number of hard brakes and rolling of the vehicle which can be a measure of the ‘quality of driving’. The parameters will be uploaded to a server at every instant so that the bus authority can directly observe the driver performance which will increase the responsibility of the driver to avoid reckless driving. Whenever the vehicle speed will exceed the safe calculated threshold speed, the system will alert the driver by playing sound on the buzzer. In addition, the system has also a button. In case of an accident if this button is pressed the system will send SMS instantly to proper authority with the location information. By this way, this system can speed up the rescue operation, minimize the casualties after the accident and numerous important human lives can be spared. |
14:30 | Numerical Simulation of Thin Film Solar Cell Using SCAPS-1D: ZnSe as Window Layer PRESENTER: Md. Abu Sayeed ABSTRACT. Investigation on properties of Copper-Indium-Gallium-Selenide (CIGS) based solar cells has been numerically performed by Scaps-1D. Designing ZnSe/CdS/OVC/CIGS structure, we have extracted current-voltage relation, quantum efficiency, cells efficiency etc. and also evaluated the effect of thickness for both absorber (CIGS) and window (ZnSe) layer on cells performance like short circuit current (Isc), open circuit voltage (Voc), fill factor, quantum and cells efficiency. We varied thickness of absorber layer from 0.2 μm to 1.0 μm and window layer from 30 nm to 90 nm and our simulation results showed that cells parameters are significantly affected by both absorber and window layer thickness. Like as the thickness of absorber layer increases, current density as well as cells efficiency enhances which is just reverse for ZnSe window layer thickness. We got maximum 21.47% cells efficiency for 30 nm ZnSe window layer and 22.86% for 1 μm CIGS layer and minimum 18.52% and 17.03% for 90 nm ZnSe and 0.2 μm CIGS layer respectively. Effect of these two layers on quantum efficiency is also significant with thickness variation. A thin layer of MgF2 having thickness of 10 nm is used as antireflection coating over window layer to accelerate internal absorption and Mo and Al are utilized as back and front contact respectively. |
14:45 | Solar Based Energy Saving Smart Industrial Exhaust Fan PRESENTER: Md. Moine Mu Shabbir ABSTRACT. Energy consumption rate around the world is increasing due to industrialization. Industries are consumed about 27% of produced energy throughout the world. A large amount of energy is accounted for ventilating purpose in these industry. It has been estimated that about 40% of total energy consumption is consumed for ventilating in an industry. So, development is needed in the ventilating system to save energy. An energy-saving smart ventilating system for industrial building is presented by this paper. This is a user-friendly system based on Arduino Nano and GSM module. Bluetooth module provides easy controlling for the system. Here, SIM900D works as GSM module. It is applicable to send information by voice data to user through an application. Experimental results show that energy can be saved because it consumes less power than the conventional system. The use of solar panel as an energy source rather than grid energy is provided more accuracy. The produced energy from burning coal, gas is cause global warming. So, the use of solar panel as an energy source can be environment friendly. A flame sensor is ensured the fire safety of an industry to prevent from an unwanted explosion. We have used PIR sensor to detect the employee in working place for start up the exhaust fan. A temperature sensor is included in the feature. The proposed system has both automatic and personal control system. |
15:00 | Investigating The Performance of Nanocrystalline Silicon HIT Solar Cell by Silvaco ATLAS PRESENTER: M. M. Haque ABSTRACT. In this work, distinct parameters of nanocrystalline silicon HIT (heterojunction with intrinsic thin layer) solar cells were investigated. The modeling and simulations were performed by Silvaco ATLAS in accordance with the global standard spectrum (AM1.5g). PV characteristics of the solar cells, particularly fill factor, open-circuit voltage, efficiency, and short circuit current were evaluated. The device possesses two sensitive parameters which are shown by the outcome from modifying the work function of front contact, emitter thickness and doping concentration of emitter. The work function of the ITO layer was observed as the most reactive parameter and the emitter doping concentration was detected as the second most sensitive in that regard. The efficiency meliorates as the work function rises. The maximum efficiency was observed at 5.3eV of work function in front contact. Raising the doping concentration of emitter as well as optimizing the thickness of emitter also improves cell efficiency. Energy band diagrams which are shown explore the causes of the increase and decline in the efficiency. Photovoltaic properties remain unchanged after varying the length of absorber. |
15:15 | Enhancement of Conversion Efficiency of CdS-CdTe Photovoltaic Cell Sandwiching Intrinsic CdTe Layer between Window and Absorber Layers PRESENTER: Nahid Akhter Jahan ABSTRACT. An ultra-thin Cadmium Sulphide (CdS) - Cadmium Telluride (CdTe) photovoltaic cell has been designed incorporating an intrinsic layer (i-CdTe) and p+-CdTe layer as a back surface field (BSF) to achieve high conversion efficiency. The optimized thickness of the proposed photovoltaic cell is only 1.9 micron. The conversion efficiency of our proposed cell increased noticeably with introducing a CdTe intrinsic layer. It is found that, with the increase of thickness of intrinsic layer, the conversion efficiency increased. At a solar irradiance of AM 1.5, we achieved open circuit voltage (Voc) of 0.93 V, short circuit current (Jsc) of 47.52 mA/cm2, the maximum power of 36.91 mW/cm2 and fill factor (FF) of 83.98%. The overall conversion efficiency of the proposed cell structure is achieved as 31%. |
Paper # 22, 158, 159, 201, 244, 249
14:00 | Classification of ECG signals by dot Residual LSTM Network with data augmentation for anomaly detection PRESENTER: Zabir Al Nazi ABSTRACT. Classification of ECG signals is of great importance for the detection of cardiac dysfunction. Recurrent Neural Network family has been greatly successful for time series related problems. In this paper, we compare different RNN variants and propose dot Residual LSTM network for ECG classification. Here, we use extracted features both from time and frequency domain with the network to improve the classification performance. A data generation scheme was developed with Conditional variational autoencoder (CVAE) and LSTM to increase training samples. A comparative analysis was studied to assess the performance of the model. The proposed dot Res LSTM achieved maximum accuracy of 80.00% and F1 score of 0.85. Furthermore, the model achieved maximum F1 score of 0.87 with augmented data. The study is expected to be useful in automatic cardiac diagnosis research. |
14:15 | One-Class SVM for Human Brain Segmentation PRESENTER: Sarwar Saif ABSTRACT. Tractography is a recently used technique in neuroscience to represent human brain neurons in 3D data. The represented 3D data is named as streamline and the set of streamlines from the whole brain is called tractography, which contains 1M to 3M streamlines. Consequently, the group of streamlines with anatomical meaning is known as a tract. In neuroscience, tract analysis used in human brain disease identification and accordingly segmentation of tract from the whole brain tractography plays an important role in that. In this paper, we propose a novel idea of supervised tract segmentation using a One-class Support Vector Machine (OCSVM). Here we have adapted the OCSVM to segment a single tract from the whole brain tractography. Furthermore, we also proposed a new embedding technique for tractography data to fit the OCSVM model. We have shown that OCSVM works better than traditional supervised KNN (K nearest neighbour) approach for tract segmentation through various experiments. |
14:30 | Estimating spectral heart rate variability (HRV) features with missing RR-interval data PRESENTER: Rawshon Ara Rupa ABSTRACT. Physiological signals, ECG signal, have been widely used for diagnosis, disease identification and nowadays for selfmonitoring. Missing data represents the problem in spectral analysis. This study focuses on the HRV power spectral analysis in frequency-domain using three methods with simulated missing data in real RR-interval tachograms. Actual missing ECG data are collected from the unconstrained measurement. Parametric, Non-parametric and uneven sampling approach were used for calculating the power spectral density (PSD), and cubic spline interpolation method was used for the non-parametric method. Based on this studies outcome, the effect of missing RR-interval data and optimal method was observed through the simulated real RR-interval tachograms for missing data. About 0 to 6 percentage data were removed according to the exponential Poisson distribution from the real RR-interval data for normal sinus rhythm, atrial fibrillation, tachycardia and bradycardia patient which data obtained from MIT-BIH Arrhythmia database to simulate real-world packet loss. For this analysis, 5 min duration data were used in all and 1000 Monte Carlo runs is performed for certain percentage missing data. PSD corresponding each frequency component was estimated as the frequency-domain parameters in each run and error power percentage based on each element difference between with and without the missing data duration were calculated. |
14:45 | 3D U-Net: Fully Convolutional Neural Network for Automatic Brain Tumor Segmentation PRESENTER: Fairuz Shezuti Rahman ABSTRACT. A principal problem in brain tumor-related diagnosis, monitoring and treatment is the assessment of the tumor area and location. Automatic segmentation is attractive in this context, as it grants us a faster approach and potentially more accurate description of tumor parameters. In this paper, we have proposed a semantic segmentation technique to 3D MRI images based on popular U-net architecture which is a Fully Convolutional Network (FCN) and robust algorithm. Our presented network is evaluated on the Multimodal Brain Tumor Image Segmentation (BRATS 2015) dataset, which contains 274 people's brain MRI images. We have used soft dice loss function to subsist with class disparities and augmented the data to prevent overfitting. Cross-validation of tumor segmentation has also displayed that our model architecture can get promising segmentation efficiently than other popular architectures with a dice score of 0.79 for the whole tumor. |
15:00 | A Modified Method for Brain MRI Segmentation Using Dempster-Shafer Theory PRESENTER: Salma Akter Lima ABSTRACT. In any kind of medical imaging, segmentation plays a crucial part. To detect brain tissues or tumor, brain MRI segmentation is the safest way. But brain MRI segmentation is not an easy task because of various kind of artifacts such as intensity non-uniformity, partial volume effects, and noise. Brain MRI artifacts lead to uncertainty in pixel values of the different parts of MRI modalities, which includes T1- and T2-weighted, PD (proton density) and Flair images. In this paper, the proposed method has followed two main steps. Firstly, Brain MRI is divided into some clusters using Fuzzy c-mean (FCM) then Fuzzy clustering and specific mapping were used to form the Dempster–Shafer belief structure. The purpose of this research is to combine Dempster–Shafer theory and fuzzy clustering which gives a technique based on data fusion of different modalities by considering spatial neighborhood information. The results of the proposed method have been evaluated by using Dice and Jaccard coefficients to compare the results with other states of the art methods. The proposed method has shown better results than other benchmark methods. |
15:15 | Comorbidity Effects of Mitochondrial Dysfunction to the Progression of Neurological Disorders: Insights from a Systems Biomedicine Perspective PRESENTER: Md. Shahriare Satu ABSTRACT. Mitochondrial-dysfunction is linked to various neurological diseases. To understand these complications we developed a quantitative framework to explore how mitochondrial-dysfunction influences progression of Alzheimer’s, Parkinson’s, Huntington’s, Amyotrophic Lateral Sclerosis and Cerebral Palsy. We sought insights from the gene profiles of mitochondrial and associated neurological disorders by constructing gene-disease networks. We also employed KEGG pathways and Gene Ontology to explore functional enrichment, and protein-protein interaction networks to identify the protein groups shared between these diseases. These identified potential biomarkers were verified using gold-benchmark databases. Our identified signature genes and pathways are useful to identify co-morbidity outcomes for the mitochondrial dysfunction. |
Paper # 9, 52, 74, 126, 271, 363
14:00 | Grayscale Portrait Colorization using CNNs and Pretrained VGG-Face Descriptor PRESENTER: Naimul Haque ABSTRACT. Colorization problem is a process of adding colors to a grayscale image. Traditionally it requires human-labeled color scribbles on the grayscale image and the image is colored by propagating the scribbled colors throughout the image using optimization techniques. It’s not very long ago that colorization without the intervention of human was almost impossible. Recently colorization using Neural Networks has achieved big success due to the availability of huge datasets. This paper attempts to color only portrait images using Convolutional Neural Networks along with a pretrained VGGFace descriptor as a global feature extractor. It has been shown that using separate networks, one to extract 3x3 patch sized features or local features and another to extract global features, result in a higher quality of colorization outputs. In portrait colorization, we used VGGFace descriptor as the global feature extractor that encodes a grayscale portrait into a vector representation which later can be used to train the system to have overall coloring effects on the grayscale portrait using fusion technique. To measure the colorfulness of the output image we introduce the use of colorfulness metric that has the potential to capture the perceptual colorfulness essence from the output images. To evaluate the model performance we used Mean Squared Errors (MSE) and Mean Absolute Errors (MAE). |
14:15 | An Improved Method for Facial Expression Recognition Based on Extended Local Binary Pattern and Differential Features PRESENTER: Sanjana Farial ABSTRACT. In our study, we made an effort to combine the Extended Local Binary Pattern(ELBP) and PCA (Principal Component Analysis) to weigh out the robust facial features of a person from an image. We set our goal as to develop and perform a low-computational approach on two most widely used databases - JAFFE and Cohn-Kanade based on the differences in the feature vectors to classify the human expressions shown in the images into seven basic emotions-anger, disgust, fear, happiness, sadness, surprise, and neutral. Moreover, we have also proposed a completely unique idea of feature vector generation based on differential method. Our approach yielded test accuracies of 88.29% and 81.04% on the mentioned databases respectively. Furthermore, we have also trained and tested our model on a custom database by mixing up both of these databases and it held an accuracy of 86.89%. |
14:30 | A Fast Interactive Geometric Shape Recognition Method PRESENTER: Marjuka Ferdousi Lazin ABSTRACT. This paper focuses on a simple, fast and highly accurate procedure for recognizing 2D geometric shapes such as rectangle, trapezium, acute triangle, etc. thereupon providing an interactive platform for the children. Proposed method for recognition is based on fundamental digital image processing and basic metrics of the shapes. On the basis of the pixels position of the shape’s image on a predefined bounding-box and geometric logic, broadly 7 features have been addressed for recognizing 10 shapes. Defined novel features and use of decision tree not only simplifies the process of recognition but also lowers the response time. For experimental evaluation, two different types of data set consisting a total of 4,725 shapes of 10 different 2D geometric shapes together are used and the overall accuracy of the recognition process for the data sets are 98.79% and 99.39% respectively. |
14:45 | A Comparative Study of License Plate Detection and Recognition Techniques PRESENTER: Md. Farhan Sadique ABSTRACT. This paper reviews several works on license plate detection and recognition (LPDR) techniques. LPDR system detects license plates from images and recognizes what is written in the license plate. The importance of LPDR system is increasing day by day. Starting from imposing traffic rules to security measures, everywhere in traffic management system, automation is one of the most important tasks. Some most recent approaches of LPDR system have been discussed and their relative performances have been analyzed in this paper. The applications, advantages and limitations are also discussed based on the approaches. A comparative study is shown based on their suitability on several aspects such as performance, detecting capability of multiple license plates, detecting capability of oblique license plate, scale adaptiveness and sensitivity to noise. |
15:00 | Spectral-Spatial Feature Extraction Using PCA and Multi-Scale Deep Convolutional Neural Network for Hyperspectral Image Classification PRESENTER: Md Rakibul Haque ABSTRACT. Hyperspectral imaging (HSI) is one of the emerging research fields in remote sensing technology as it contains huge information about a scene which can be further analyzed to detect various kind of objects. HSI contains several narrow and contiguous spectral bands which provide different research challenges for dealing with high dimensional data for better classification. Most of the traditional method takes only the spectral feature into account for classification. Principal Component Analysis (PCA) is widely used for this task. Recently with the introduction of the convolutional neural network, spatial features are being fused with the spectral features for better classification. However, traditional convolutional neural network (CNN) based methods take only a single scale spatial kernel in order to explore the spatial information. In this paper, we have proposed a hybrid approach of a dimensionality reduction technique based on PCA and a novel multi-scale convolutional neural network named PCA-MS-CNN in order to extract spectral and spatial feature for hyperspectral image classification. We have achieved an average accuracy of 99.10% on Indian Pines dataset using this method. This result shows that it outperforms other state-of-the-art deep learning based methods used for HSI classification. |
15:15 | Water Bodies Identification in Landsat 8 OLI Image Using Machine Learning PRESENTER: Farzana Yasmin ABSTRACT. Water is one of the important natural resources on the earth which is used for various purposes. But human activities and climate change causes reduction of surface water. So water bodies monitoring and management are essential. In recent years, machine learning methods have given better performance than other methods to classify image. In this study, two machine learning methods- Classification And Regression Trees(CART) and Support Vector Machine(SVM) are used to identify water bodies in Landsat 8 OLI images. The performances of the models are compared with water index methods and evaluated based on overall accuracy, F1 score and kappa coefficient. For machine learning models two test sets are used. One is the samples of own dataset and another is completely new unseen samples to check how the models perform when there are unknown samples. Evaluation result shows that, though CART can achieve higher accuracy among all the methods when samples of own dataset are used, its performance decreased when unseen samples are used. But SVM has given similar performance for both test sets and its accuracy is more than 98%. |
Light Snacks and Tea/Coffee will be distributed
Paper # 111, 145, 256, 287, 341, 366
16:00 | A Wireless Electronic Stethoscope to Classify Children Heart Sound Abnormalities PRESENTER: Md. Riadul Islam ABSTRACT. In this research paper, a wireless stethoscope has introduced that can communicate with a smartphone to receive children’s heart sound. Along with an automated method that recognizes children heart sound abnormalities. That isolation of heart sound is based on time-frequency characteristics. Where it is preceded using Mel-frequency Cepstral Coefficients (MFCCs) signal processing method. The processed sounds are extracted using five feature extraction algorithms. Then it is classified using four support vector machines (SVM) kernel. Total 60 heart sounds were collected, where 30 sounds having abnormalities and rest 30 sounds containing normal heart sound. Though massive measures of action have already been done in this area, still necessity of more bearable cost device and accurate method is present. Here, the submitted apparatus cost is approximately 18 USD, which is the cheapest than most other device used in previous work. Simultaneously it is lightweight and bearable to use in rural and underprivileged area. With RBF kernel of SVM, the proposed method shows 94.12% accuracy which is the highest. |
16:15 | Time Series Parameter Prediction for ICU Patient PRESENTER: Sharmin Sharwardy ABSTRACT. Now a days many scientist are working on prediction of happenings of Intensive Care Unit (ICU). But prediction of the future state of patients in ICU is a complex process. Huge amount of quality data and past experience to the analysis of this data is required to predict such scenario. In ICU, patient data is usually considered as Time series data. Forecasting time series data to predict patient status is an important research area in ICU data mining research. Time series data forecasting has been shown to be efficient in appropriate decision-making in health sector. This study aims to develop Simple Exponential Smoothing model and ARIMA model to forecast the hourly pediatric heart disease data in ICU. |
16:30 | Disha: An Implementation of Machine Learning Based Bangla Healthcare Chatbot PRESENTER: Md. Moshiur Rahman ABSTRACT. While humans as a whole do live longer nowadays than ever before, we now suffer certain illnesses to a degree never seen in the past - including rising rates of Diabetes, Hypertension, Hypotension, Cholesterol imbalance, obesity and, ailments such as fever. People around the world are preemptively seeking medical advices on how to live a healthy lifestyle. They are looking to lower their risk of various diseases. A healthcare chatbot can play a significant role to monitor a person’s health status. A healthcare chatbot is a computer program designed to simulate conversation with human users as a virtual medical assistant. Our study shows, there are some healthcare chatbots available in English and other languages but not in Bangla. In this paper, we have demonstrated a machine learning-based closed domain Bangla healthcare chatbot ‘Disha (Direction)’ which can converse in Bangla with the user with the help of its knowledge base and through learning from interactions with the user. It helps a user to diagnose potential diseases based on inputted symptoms, keep track of a user’s health status, and alert a user from potential health hazards. This paper explores the use of six supervised machine learning approaches and showed significant efficiency. |
16:45 | Toddlers Working Memory Development via Visual Attention and Visual Sequential-Memory PRESENTER: Md. Ezaz Raihan ABSTRACT. Human psychology is a very interesting thing. How people think, learn and interact are the main branches of psychological development. The reasoning style of individuals as in how the general population think and learn is exceptionally equivocal. The impact of visual attention and working memory on the mental learning arrangement of humans made this research extremely inquisitive. Visual inclining is one of the primary drivers of mental advancement in a toddler. Through the eyes, the human brain always looks for important visual information. And for most tasks, people rely on working memory a lot. Working memory involves storing, focusing attention on, and manipulating information for a relatively short period of time. Children mostly learn things by watching the adults and this is one of the major examples of how visual attention helps anyone in learning very quickly. Researchers are attempting to understand visual attention in various aspects such as kid improvement tracking, conduct acknowledgment, field of intrigue and so on. This research is focused on learning the development of intellectual ability of toddlers in various age groups through assigning some tasks to them. |
17:00 | Design and implementation of a head motion controlled semi-autonomous wheelchair for quadriplegic patients based on 3-axis accelerometer PRESENTER: Nusrat Jahan ABSTRACT. Quadriplegic patients are affected by paralysis of all four limbs and therefore they cannot move without the assistance of other people. Fortunately, they can move their head and it could be a way to control the navigation of a wheelchair. In order to give assistance to paraplegic patients, a head motion-controlled semi-autonomous wheelchair is proposed in this paper. This wheelchair can help any paraplegic patients as well as disable persons to move indoor or outdoor places without the assistance of any other people. Here a 3-axis accelerometer sensor is used to detect the head movement and based on the movement of head two DC motors are controlled to navigate the wheelchair. The system also has two sonar sensors to detect the obstacle in the front or back direction of the wheelchair to avoid clash or accident. In case of any accident, it has ability to detect the accident and to inform the family member by sending an SMS through GSM modem with location information. The detail methodology to develop a prototype of the proposed wheel chair, the performances of the prototype and the advantages of this wheelchair compared to the existing system are presented in this paper. |
17:15 | Symptomatic Activity Recognition Using Smartphone Sensors PRESENTER: Dr. Abdul Kadar Muhammad Masum ABSTRACT. Being concerned about the rising rate of the usage level of smartphone for last few years, researchers are striving to let out the disguised assets of smartphones. Regarding this phrase, the embedding of a variety of sensors such as accelerometer sensor, gyroscope sensor, humidity sensor etc. has attained the considerable attention of researchers for facilitating the human activity recognition task employing the sensor’s applications. In this paper, we practiced the Long Short-Term Memory a.k.a. LSTM deep learning model with an aim to recognize thirteen human activities stated as walking, walking upstairs, walking downstairs, sitting, standing, jogging, squatting in the toilet, fallen down, lying, cycling, drinking, eating and genital itching. We opted for these activities with an intention of early diagnosing of diabetes in near future. We amassed data from four sensors stated accelerometer sensor, gyroscope sensor, humidity sensor and temperature sensor subjecting ten volunteers implying a frequency of 10Hz. The data was attained using an android application which was developed by us for the purpose of accumulation of sensor data from the smartphone. |
Paper # 149, 160, 161, 166, 176, 339
16:00 | Neural Machine Translation for Bangla-English Language Pair PRESENTER: Md. Arid Hasan ABSTRACT. Due to the rapid advancement of the different neural network architectures, the task of automated translation from one language to another is now in a new era of MT research. In the last few years, Neural Machine Translation (NMT) architectures have proven to be successful for resource-rich languages, trained on a large dataset of translated sentences, with variations of NMT algorithms used to train the model. In this study, we explore different NMT algorithms – Bidirectional LSTM and Transformer based NMT, to translate the Bangla to English language pair. For the experiments, we used different datasets and our experimental results outperform existing performance by a large margin on different datasets. Besides, we also investigated some factors affecting the data quality and how it influences the performance of the models. It also shows a promising research avenue to enhance NMT for the Bangla-English language pair. |
16:15 | Unsupervised Context-Sensitive Bangla Spelling Correction with Character N-gram PRESENTER: Shuvendu Roy ABSTRACT. We propose an unsupervised context-aware spelling error detection and correction method for traditional Bangla written script. The context-aware capability is introduced by generating a representation of information from the context words and using the concept of cosine similarity of find the best candidate for the misspelled words. We used fast-text embedding concept of using character n-gram embedding. The character n-gram embedding is capable of generating embedding for unknown words, which is a capability missing in any conventional methods. We performed our experiments on Bangla language collected from popular newspapers, blogs, Bangla Wikipedia and Banglapedia. Our proposed method outperforms conventional methods. |
16:30 | Employing Machine Learning techniques on Sentiment Analysis of Google Play Store Bangla Reviews PRESENTER: Md. Muhtasim Jawad Soumik ABSTRACT. This article offers an in-depth insight on a number of existing methodologies to perform sentiment analysis using text classification on Bangla dataset. Although the rapidly developing machine learning algorithms are showing promising results, the viability of those methods for non-English languages such as Bangla is yet to be fully explored. This research aims to fill in some of those existing research gaps through proper implementation of machine learning techniques where words are converted into feature vectors via implementation of TF-IDF algorithm on data crawled from Google play store, the largest Android application market. Many significant algorithms staring linear algorithms like Naive Bayes, Linear Support Vector Machine (SVM) are implemented. An in-depth comparison is also made among the results of various existing algorithms. The experimental results indicate that even the base-line algorithms, after proper pre-processing, can show promising results on our Bangla dataset. Naive Bayes, Support Vector Machine and Logistic Regression has shown very promising results (accuracy score of 0.75 on average) even with the data limitation. An Ensemble method is also proposed with Adaptive Boosting technique showing an accuracy score of 0.7639 with five-fold applied. SVM has the best accuracy score of 0.7648 among all the algorithms when five-fold is applied and Gradient Boosting has the best accuracy score of 0.7695 when five-fold is not applied. |
16:45 | Authorship Attribution in Bangla literature using Character-level CNN PRESENTER: Aisha Khatun ABSTRACT. Characters are the smallest unit of text that can extract stylometric signals to determine the author of a text. In this paper, we investigate the effectiveness of character-level signals in Authorship Attribution of Bangla Literature and show that the results are promising but improvable. The accuracy is 2-5% less than the best performing word-level models but the time and memory efficiency is much higher. Comparison of various word-based models is performed. Moreover, CNN model performs better with larger datasets. We also analyze the effect of pre-training character embedding of diverse Bangla character set in authorship attribution. It is seen that the performance is improved by up to 10% on pre-training. We used 2 datasets from 6 to 14 authors, balancing them before training and compared the results. |
17:00 | A Novel Approach to Classify Bangla Sign Digits using Capsule Network PRESENTER: Tonmoy Hossain ABSTRACT. Communication between hearing impaired and general people is one of the most immense problems nowadays. The main medium of conversation with the deaf people is through sign language. But it is an arduous task to learn and communicate with the sign language for any class of people. Over the years, researchers from diverse background tried to establish a model to automate the process of the detection of sign language. Traditional machine learning technique—k Nearest Neighbor, Support Vector Machine etc. and Neural Network architecture Back-propagation method, Convolutional Neural Network, Ensemble Neural Network etc. are such existing methods by which the researchers proposed their work. In this article, we worked on the detection of Bangla sign digits. We adopted Capsule Network and modified by the means of detecting Bangla sign digits. The capsule is a group of neurons constitutes of the feature vector and spatial properties of the object. A seven-layer of capsule network architecture along with some pre-processing is proposed for the detection. The IsharaLipi dataset is used for the training of the model and we get 98.84% accuracy for the classification. |
17:15 | Bangla Intelligence Question Answering System based on Mathematics and Statistics ABSTRACT. The Question Answering System (QAS) is one of the significant Machine Learning processes that assist a user to find out the relevant information by Natural Language Processing (NLP). In this research paper, we have described the design and implementation of three Bangla intelligence question answering systems (BIQAS) based on mathematical and statistical procedures using question answering data. These procedures are cosine similarity, Jaccard similarity, and Naive Bayes algorithm. The cosine similarity has interacted with dimension reduction technique SVD on user questions and questions answering so that space and time complexity can be reduced. The methodology of this research is separated into three parts: data collection, pre-processing data, and the establishment of a relationship between users’ questions and information. The system’s primary source of knowledge is a collection of the informative data of Noakhali Science and Technology University. We have gotten 93.22% accurate answer by using cosine similarity, 82.67% by Jaccard similarity and 89.32% by Naïve Bayes algorithm. |
Paper # 58, 129, 223, 262, 265, 284
16:00 | Explore Voice and Foot-based Interaction Techniques to Navigate 2D Radiological Images in the Virtual Reality Operation Theatre PRESENTER: Nusrat Jahan Farin ABSTRACT. During surgery, the surgeons often need to interact with the 2D radiological images. Due to sterility, surgeons are unable to interact with the system and depends on duty assistant. However, communicate with the substitute might be complicated and error-prone if the operator and the surgeon do not have an equal level of communication skills and might interrupt the workflow. To get rid of the dependency on substitute operator, two hands-free modalities i.e. voice and foot-based interactions have been analyzed and investigated for the surgeon to interact with 2D radiological images. To feel like a real operation theatre (OT) we design the environment in Virtual Reality. The hands-free modalities are deployed in the virtual reality environment. Around 16 participants had evaluated the interaction systems. To analysis, the system usability, the qualitative and quantitative studies have been considered. Our finding indicates that both systems reached general usability rating. As well as the observation indicated that the voice command system is comfortable to use whereas foot-based interaction is more efficient. Nevertheless, for the short and longer interaction foot-based interaction is preferable than a voice command, however, for short interaction voice command is more acceptable. |
16:15 | GNSS Position Accuracy Considering GDOP and UERE for Different Constellation over Bangladesh. PRESENTER: A.A.M Shah Sadman Sakib ABSTRACT. GNSS position accuracy is very important as small error can upset the true position determination. Accuracy depends on both accurate measurements and good geometric relationship between the receiver device and the measurement units required for calculation. Dilution of Precision (DOP) and User Equivalent Range Error (UERE) are considered the two parameters that determine the position accuracy. DOP is a unit less parameter, used to describe the satellite geometry and UERE is the total sum of errors involved in satellite communication expressed as the unit of distance. In this paper, an analysis is done by calculating the number of in view satellites and DOP for Bangladesh over a certain period of time considering GPS and Galileo constellation as well as calculating GNSS position accuracy by using the typical UERE value. |
16:30 | Socio-network Analysis of RTL Designs for Hardware Trojan Localization PRESENTER: S M Asaduzzaman ABSTRACT. The recent surge in hardware security is significant due to offshoring the proprietary Intellectual property (IP). One distinct dimension of the disruptive threat is malicious logic insertion, also known as Hardware Trojan (HT). HT subverts the normal operations of a device stealthily. The diversity in HTs activation mechanisms and their location in design brings no catch-all detection techniques. In this paper, we propose to leverage principle features of social network analysis to security analysis of Register Transfer Level (RTL) designs against HT. The approach is based on investigating design properties, and it extends the current detection techniques. In particular, we perform both node- and graph-level analysis to determine the direct and indirect interactions between nets in a design. This technique helps not only in finding vulnerable nets that can act as HT triggering signals but also their interactions to influence a particular net to act as HT payload signal. We experiment the technique on 420 combinational HT instances, and on average, we can detect both triggering and payload signals with accuracy up to 97.37%. |
16:45 | Statistical mmWave Channel Modeling and Characterization in Indoor Airport Environments PRESENTER: Muhosina Aktar Mou ABSTRACT. With the extensive demand of higher data rate with advancement of mobile applications currently facing the need of unbounded bandwidth for everlasting mobile traffic. Recently, researchers showing great interest in improving wireless communication inside airports, as airports are very important spots that needs inexhaustible communication and tackles huge indoor mobile traffic. This has become a quite challenging issue in this regard. Despite of limitations, mmWave with latest technology 5G, along with its special features is expected to satisfy this huge demand of users. This article provides investigation of power delay profiles for statistical channel model for both line-of-sight (LOS) and non-line-of-sight (NLOS) conditions inside airport environments at mmWave bands 28, 60, and 73 GHz. The simulation outputs exhibit the channel response statistics in different frequencies and conditions. The results presented here may facilitate improvements in the future works related to inside airport or overall aviation communications. |
17:00 | Weather Impact analysis of mmWave Channel Modeling for Aviation Backhaul Networks in 5G Communications PRESENTER: Hussnin Binte Hamid Dutty ABSTRACT. Millimeter wave (mmWave) spectrum (30 GHz to 300 GHz) is about to serve the 5G communications as the frequency spectrum of 4G and 3G is overcrowded by the users worldwide. However, mmWave experiences attenuation during its propagation from transmitter to receiver on account of various weather conditions like rain, fog, precipitation etc. In this paper we present the impact of weather on the characteristics of channel model of mmWave band frequencies considering rainy season and winter season of Bangladesh. Two carrier frequencies, one from unlicensed V-band i.e 60 GHz and other from licensed E-band i.e 73 GHz are selected and a study is made to observe the behavior of mmWave due to the changes in rain rate, temperature, humidity and pressure. Finally, we draw a comparative survey on different climatic conditions is made to give a general concept on the propagation losses of mmWave link in backhaul network at 60 GHz and 73 GHz. |
17:15 | A Smart Approach Of Parking Services And Minute Coin Concept PRESENTER: Rowshni Tasneem ABSTRACT. With the increase in vehicle production, more and more parking spaces and facilities are required. The current transportation infrastructure and car park facilities which are being developed, are unable to cope with the influx of vehicles on the road. In this paper, a new car parking system called Smart Car Parking System is proposed to initially assist drivers to find vacant spaces for car parking in a shorter time using an android based application system. The software also holds information of both driver and parking provider in a database communicated over a wireless network. Moreover, the system is integrated with a hardware solution utilizing sensors, LCD controlled with a micro-controller for the prototype. To ensure security there are involvements of QR scanner, VLP scanner. The system is built considering the advancement of technology and smart living aiming to improve the nature of human life. This paper introduces the idea of a Minute transaction system incorporated with the parking framework and android app. The system will help in the proper management of the vehicle, considering the availability of space, which will further assist in optimized usage of land, and provide smart transaction idea, build upon the basic knowledge of cryptocurrency. |
Paper # 94, 140, 213, 294, 315, 359
16:00 | Dynamic Obstructed Group Trip Planning Queries in Spatial Databases PRESENTER: Sweety Lima ABSTRACT. In this thesis, we introduce dynamic obstructed group trip planning (DOGTP) query which allows a group of members to plan a trip dynamically with a minimum total trip distance in the obstructed space. Here, the group of members in a trip can join or leave or rejoin in the middle of a trip at any point of interests (POIs) i.e. restaurant, shopping mall, library or theater. Researchers developed solution for dynamic group trip planning queries in the Euclidean space. They did not consider obstacles in their problem space. In the obstructed space, members can face different types of obstacles i.e. building, private property, main road, lake etc. Therefore, the key challenges of our research is to minimize the number of retrieved obstacles from the database, avoid retrieval of same obstacle multiple time and to reuse the already computed obstructed distances. Thus in this thesis, we extend the research work of dynamic group trip planing query in the Euclidean space into the Obstructed space. We develop solution for processing DOGTP query that returns the optimal trip for all group members with reduced I/O and query processing overhead. |
16:15 | SMIFD: Novel Social Media Image Forgery Detection Database PRESENTER: Md. Mehedi Rahman ABSTRACT. Image forgery or manipulation changes the contents of a set of original images to create a new image. Unfortunately, manipulated images become a growing concern with respect to spreading misinformation via image sharing in the social media. Despite the availability of a large number of automatic Image Forgery Detection (IFD) methods, their evaluation in real-world benchmarks seems to be limited due to the lack of diverse datasets. Moreover, the motifs behind the manipulation remains unclear. This research aims to address these issuees by proposing a novel social media IFD database, called SMIFD-500, to evaluate the efficiency and generalizability of the IFD methods. The unique property of this dataset is the availability of the technical and social attributes in its ground truth annotations. These will benefit the scientific community to develop efficient methods by exploiting such annotations. Moreover, it provides interesting statistics, which highlights the motifs of image manipulation from social science perspective. |
16:30 | Selectively Oversampling Difficult Positive Samples from Imbalanced Data for Preprocessing PRESENTER: Md Mahin ABSTRACT. Oversampling is a procedure traditionally has been applied to train machine learning classifiers for a better per- formance in presence of class imbalance. This work suggests a new insight for oversampling imbalanced data. In literature Borderline samples are mainly focused for oversampling. How- ever, because of low number of samples within the positive class a huge percentage of samples can be labeled as Rare and Outliers. These samples are often overlooked by the traditional oversampling methods or the nearest negative samples are often removed to increase positive prediction rate- while sacrificing the negative prediction rate. This work demonstrates that by only oversampling the Borderline, Rare and Outlier samples at different rate, better performance can be achieved than all other pre-processing methods. The proposed method is applied on four datasets- Abalone, CMC, Solar Flare and Seismic Bump, collected from the UCL digital library and compared with four traditional pre-processing methods ADYSYN, SMOTE, Border- line SMOTE 1 and 2 from imbalanced learn toolkit python. The result analysis shows that with fine tuning better performance can be achieved for all known performance measurements: Accuracy, True Positive Rate, True Negative Rate, Geometric Mean, Area Under the Curve measure and F-measur. |
16:45 | A hybrid genetic algorithm with chemical reaction optimization for multiple sequence alignment PRESENTER: Promal Barua ABSTRACT. Multiple Sequence alignment is the ultimate challenging tasks of biological science. It is used for comparison or difference or similarities in these sequences of data. Here, we applied a pragmatic Genetic Algorithm (GA) & Chemical Reaction Optimization (CRO) apparently the most suitable and familiar expansion technique and influenced by the natural genetic structure. Inquiring the magnificent alignment of a biological sequence set is classified as an NPhard optimization problem for that, GA-CRO algorithms are capable to drive this complication. To find good results, we are going to shown the benchmark dataset, the suggested approach is compared with those of the current tools like the SB-PIMA, SAGA, RBT-GA and GAPAM, HMMT. The simulation results recommend that our method be a viable solution with the other methods in terms of efficiency with the appropriate selection of parameters. |
17:00 | SRCS: A New Proposed Counting Sort Algorithm based on Square Root Method PRESENTER: Hridoy Roy ABSTRACT. Counting sort is one of the basic sorting algorithms in computer science which has a time complexity of O(N + K) where N is the number of elements and K is the maximum value among those N elements. It is superior when it comes to sort countable objects that come from a discrete set of values, such as bounded integers. But, it fails to provide efficiency if the range of K is significantly greater than N . For this reason, it is less used in practical fields of computer science though it has an extensive use as a sub-routine in other sorting algorithms. There were several approaches in the literature which have proposed several extensions of counting sort but none of those aims to increase the limit of K. This paper proposed an extension of counting sort algorithm that is named Square Root Counting Sort (SRCS) which has an increased limit of K. Specifically, the proposed approach can handle the maximum value of K^2 where the classic one is only able to solve K. The first phase of the approach is to prepare some blocks of elements using square root technique and then, sort each of the blocks simultaneously to get final sorted array. The proposed extension of counting sort outperforms than classic counting sort as well as bubble, selection and insertion sort. |
17:15 | Performance Evaluation of IT Professionals Based on Fuzzy-AHP and DEMATEL Models: Bangladesh Perspective PRESENTER: Abdullah Al Musarraf Tawhid ABSTRACT. The number of people engaging in IT services is increasing as technology has a significant impact in day to day life that needs more expert, dedicated, disciplined employee, which lead to the term of maintaining standard employee performance. In our work, we proposed a framework by which IT company management can evaluate its employee performance. In our work, we have built an employee performance model by selecting a set of criteria and sub-criteria such as subjective knowledge & professional achievement, research aptitude, personal skill, management skill. We took opinion from managers of renowned IT Company. After that Fuzzy AHP method is used to rank the factors and subfactors which has an impact on the employee to evaluate his performance. Dematel also used to watch the relationships of criteria's. Finally, an evaluation framework is built to determine the performance of an IT employee. |
Workshop Title: International Workshop on Computer Vision and Application (IWCVA)
Resource Persons:
Prof. Dr. Yasushi Makihara
Osaka University, Japan
Prof. Dr. Atsushi Inoue
Eastern Washington University, USA
Prof. Dr. Md. Atiqur Rahman Ahad
University of Dhaka, Bangladesh
Osaka University, Japan
Invited Talk 5
Talk Title: Fourth Industrial Revolution for Healthcare Innovation
Prof. Dr. Khondaker Abdullah Al Mamun
United International University, Dhaka, Bangladesh
18:00 | Fourth Industrial Revolution for Healthcare Innovation |
Conference Grand Gala Buffet Dinner
Sky View Restaurant
270 Tejgaon I/A, Dhaka 1208
(Opposite of Southeast University Campus)