ABSTRACT. Positron emission tomography (PET), x-ray computed tomography (CT) and magnetic resonance imaging (MRI) and their combinations (PET/CT and PET/MRI) provide powerful multimodality techniques for in vivo imaging. This talk presents the fundamental principles of multimodality imaging and reviews the major applications of artificial intelligence (AI), in particular deep learning approaches, in multimodality medical imaging. It will inform the audience about a series of advanced development recently carried out at the PET instrumentation & Neuroimaging Lab of Geneva University Hospital and other active research groups. To this end, the applications of deep learning in five generic fields of multimodality medical imaging, including imaging instrumentation design, image denoising (low-dose imaging), image reconstruction quantification and segmentation, radiation dosimetry and computer-aided diagnosis and outcome prediction are discussed. Deep learning algorithms have been widely utilized in various medical image analysis problems owing to the promising results achieved in image reconstruction, segmentation, regression, denoising (low-dose scanning) and radiomics analysis. This talk reflects the tremendous increase in interest in quantitative molecular imaging using deep learning techniques in the past decade to improve image quality and to obtain quantitatively accurate data from dedicated standalone (CT, MRI, SPECT, PET) and combined PET/CT and PET/MRI imaging systems. The deployment of AI-based methods when exposed to a different test dataset requires ensuring that the developed model has sufficient generalizability. This is an important part of quality control measures prior to implementation in the clinic. Novel deep learning techniques are revolutionizing clinical practice and are now offering unique capabilities to the clinical medical imaging community. Future opportunities and the challenges facing the adoption of deep learning approaches and their role in molecular imaging research are also addressed.

ABSTRACT. Chest diseases are among the most common diseases in the world. This implies an extensive utilization of Xray screening radiation, creating an increased demand for medical professionals. In our research paper, we focus on the disease of pneumonia. We use a pneumonia database for classification in order to recognize pneumonia's disease based on Chest X-ray images, employing various methods of deep learning. This study involves a comparative analysis of existing deep learning models.

ABSTRACT. Industry 4.0 is defined as the fourth industrial revolution, integrating digital technologies into manufacturing processes to enhance production efficiency, flexibility, and customization. Welding is a critical process in the industry that may lead to defects and deformations in welded structures. To avoid such issues, non-destructive testing through radiography is used. However, manual analysis of images is time consuming and complex due to the diverse shapes, sizes, and positions of defects. In this research, we propose a promising approach utilizing deep learning (DL), specifically segmentation techniques, to automate the detection of welding defects in radiographic images. The image dataset was curated from three datasets, the most comprehensive of which was obtained through collaboration with the Algerian company Eurl TESTIAL, with input from domain experts. After constructing a new patch-based dataset, we sequentially utilized two models: one for binary classification and the other for multi classification, aiming to localize defects in welds. The system is built on deep convolutional neural network architectures, namely VGG, ResNet, DenseNet, and Inception. Evaluation of the proposed models reveals that the combination of Inception V3 and DenseNet121 achieved promising results, reaching an accuracy rate of 99%. This automated detection system has the potential to revolutionize non-destructive testing by reducing time and cost, while also enabling the detection of previously invisible defects, thereby enhancing the safety and quality of welded products. Consequently, the application of deep learning in this field proves to be a promising approach with the strong potential to transform the non-destructive testing process.

ABSTRACT. This paper addresses the issue of supervised classification in the context of hyperspectral satellite images. It deals with two fundamental aspects: dimensionality reduction of data and the selection of appropriate supervised classification techniques. Firstly, we delve into dimensionality reduction, a critical step in simplifying the management of hyperspectral data. The reduction aims to decrease complexity in terms of memory and computing time. We examine two commonly used methods: Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Subsequently, we explore the selection of the most suitable supervised classification algorithms for hyperspectral images. We compare the performance of three methods: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest (RF) using real hyperspectral data. The results highlight that the combination of PCA and RF yields the highest overall accuracy and Kappa coefficient.

ABSTRACT. The analysis and identification of structural changes within the brain from magnetic resonance images are increasingly used for the study of Alzheimer’s disease (AD). Although structural MRI (sMRI) is capable of detecting cortical shrinkage, it is incapable of detecting damage to the integrity of the white matter (WM), which occurs before cortical atrophy. Diffusion tensor imaging (DTI), a novel MRI technique, enables the early detection of these brain abnormalities. In this work, we constitute an efficacy DTI diagnosis tool to classify AD vs. mild cognitive impairment (MCI), AD vs. cognitively normal (CN) subjects, and CN vs. MCI. We propose a new 2D Convolutional Neural Networks (CNN) architecture and a 2DCNN-Support Vector Machine (SVM) technique using DTI-scalar metrics (Fractional Anisotropy (FA) and Mean Diffusivity (MD)). We performed a binary classification. We have also proposed a fusion approach, which consists of concatenated features extracted by 2D CNN, to classify them using SVM. The proposed method demonstrates superior performance with an accuracy 99.89% and 98.33 %, and 94.40% to classify AD vs. CN, AD vs. MCI and CN vs. MCI respectively.

ABSTRACT. The accurate identification of tomato plant diseases is essential for improving crop yield and reducing economic losses in agriculture. Conventionally, the Convolutional Neural Network (CNN) is used as a model for feature extraction and data classification. However, the heavy computational requirements of most state of the art deep learning architectures based on this approach make them unsuitable for low-powered devices. In order to design a lighweight identification system for tomato disease, this paper proposes an alternative implementation by combining the Convolutional Autoencoder (CAE) for generating features with the CNN for classification. The CAE is used for encoding and decoding the data to determine the encoder’s effectiveness in preserving relevant features. Then, the encoding layers are transferred to the CNN having fully connected layers for classification. The proposed framework is evaluated using the 10 classes of tomato leaf diseases available in the public Plant Village dataset. The experimental evaluation shows an outperformance comparatively to several heavy architectures with an accuracy of 99.13%. Additionally, It has low memory usage and fast inference time, making it suitable for deployment on resource-limited devices.

ABSTRACT. Distortions like blur and smoke in real-time laparoscopic videos often result from lens contamination. Detecting these distortions automatically and “in real time” is a step preceding automatic lens cleaning and leads to a clear vision for surgeons, hence reducing surgical time and minimizing risks for patient. Our approach leveraged transfer learning, transposing knowledge from natural images to the laparoscopic video domain. We utilized a pre-trained ResNet50 convolutional neural network (CNN) to extract image features, subsequently processed by a cascade of support vector machine (SVM) classifiers to categorize various distortions. This last strategy amalgamates outputs from two binary classifiers. The first classifier distinguishes videos as good or distorted. The second classifier focuses on smoke and blur detection. The first classifier attains 99 % accuracy. The second classifier achieves 100% accuracy in detecting smoke and blur. These values prove the effectiveness of our approach, which combines ResNet50-based transfer learning and cascaded SVM classification for automatically detecting smoke and blur distortions in laparoscopic videos. Such results are promising for the detection of the remaining distortions.

ABSTRACT. In real world, data understanding involves finding relationship between dependent and independents events, and usually this data is embedded in the form of images, audio, videos, speech, text and much more and to extract useful and representative information, feature extraction and representation is a front-end step. For machine learning algorithms, feature extraction is a preprocessing technique that helps in finding the relationship between different variables. In biomedical signal analysis such as cerebral emboli signal detection, feature extraction from samples constitutes an important step. In this paper, we performed two step feature extraction procedure. First we extracted Mel Frequency Cepstral Coefficients (MFCC) and biologically inspired Gamma-tone Cepstral Coefficients from doppler signals. Second we extracted kurtosis and skewness parameters from all the Mel-frequency cepstral coefficients and Gamma-tone Cepstral Coefficients coefficients. To gain the understanding of extracted features from doppler signals, we trained some machine learning algorithms such as k-nearest neighbors, support vector machines and logistic regression. GTCC based kurtosis and skewness features show better classification between emboli, artifact signals and Doppler speckle signals. We present evaluation results using confusion matrix for classification between emboli signals (ES), Doppler speckle (DS) and artifact signals (AS).

ABSTRACT. Automatic Speech Recognition (ASR) performance in mobile communications degrades significantly if the environment includes many sources of variability, such as when the test environment differs from the training environment and when the acoustical environment includes disturbances like noise, channel distortion, speaker differences, and mobile codecs. In this work, we have used two architectures for speech recognition in mobile networks. The first one is Distributed Speech Recognition based on DSR coding, and the second architecture is based on Adaptive Multi-Rate Narrow-Band (AMR-NB). We propose a novel robust feature extraction (Front-End) technique to improve speech recognition performance in noisy mobile communications. This technique utilizes special parameters such as Gabor features, Power Normalized Spectrum Gabor filter (PNS-Gabor), and Power Standardized Cepstral Coefficients (PNCC). These features take into account psychoacoustic effects like the temporal masking effect and have different distributions of filter banks and filter forms to better model human perception. In the back end, we investigated speech classification systems using Continuous Hidden Markov Models (CHMM) and Deep Neural Networks (DNN). Based on the results obtained in noisy mobile communications, the proposed features PNS-Gabor and PNCC show significant improvements over conventional acoustic features such as Mel frequency cepstral coefficients (MFCC).

ABSTRACT. To diagnose epilepsy at the first stage, clinicians can use scalp EEG(sEEG) as a safe option. However, the literature has no sEEG-based automatic diagnosis of epileptic (ES) and non-Epileptic (NES) Seizures. On the other hand, the existing benchmark datasets are unsuitable for an sEEG-based model. Thus, we created an sEEG-based dataset for both classes by merging the Bonn and CHB-MIT datasets. We performed a critical channel selection, comparing the stability of four classifiers across different channels. The results demonstrate that when utilizing Channel 1, the NB classifier achieves 98.97% accuracy, 99.71% specificity, and 97.73% sensitivity.

ABSTRACT. Controlling smart forearm and hand myoelectric prostheses requires reliable recognition of one type of arm movement among several other types. In this study, using real EMG signals of 4 types of arm movements, we searched the best multiclass classifier using SVM. Based on our previous results, features were first extracted from the relevant intrinsic mode functions (IMFs), namely the first two of each signal. Then, relevant feature vectors of short lengths (≤ 8) were selected. Two approaches of multiclass classification were tested: one vs one (OVO) and one vs rest (OVR) using different kernel functions. By varying the SVM cost C and the kernels parameters γ and n and using 100 cross-validation iterations, we found that the best classifier is OVO-RBF associated to the values (C, γ) = (1, 1.1) which gave mean values of accuracy =94.4%, sensitivity=94.44%, specificity=98.27% and AUC=0.96. Moreover, the use of SMOTE to increase the size of the database improved classifications results and all the kernels manage to discriminate between the four movements.

ABSTRACT. The electrocardiogram (ECG) is widely used for diagnosing heart diseases, including arrhythmia, due to its non-invasive nature and simplicity. Accurate detection and classification of arrhythmic types are crucial in preventing heart disease and reducing mortality. To enhance ECG heartbeat classification, we propose an automated system based on a fine-tuned deep convolutional neural network consisting of 13 residual blocks with 28 layers in total. This model was trained using a dataset of 103247 single-lead ECG recordings. Our deep convolutional neural network achieved remarkable results by validating the independent test dataset from the MIT-BIH Arrhythmia Database. It demonstrated an Accuracy of 99%, a Macro avg Precision of 96%, a Macro avg Recall of 93%, a Weighted avg of 99%, and an impressive Macro avg F1-score of 94%. Notably, our approach has demonstrated superior performance to the original architecture, with a 3% increase in F1-score. These findings showcase the potential of our improved deep learning approach to effectively classify a wide range of arrhythmias from single-lead ECGs, yielding high diagnostic performance comparable to that of cardiologists.

ABSTRACT. Physiological signals have gained significant attention in the field of biometrics due to their unique characteristics and potential for secure and reliable personal identification. However, despite the extensive exploration of various physiological signals, such as the Electrocardiogram (ECG), electroencephalogram (EEG), and facial recognition, the potential benefits offered by blood pressure (BP) signals in the context of biometric identification systems have remained largely unexplored. Therefore, the primary objective of this paper is to investigate the feasibility and usefulness of incorporating the unique characteristics of blood pressure signals, in combination with a support vector machine (SVM) classifier, for multi-person identification purposes. Through rigorous experimental analysis, we achieved a promising accuracy rate of 87.3% and an AUC value ranging from 94.8% to 99.66%. These promising results shed light on the untapped usefulness of blood pressure signals as a valuable biometric trait, thereby opening up new avenues for their application in biometric identification systems.

ABSTRACT. In this paper the Pachycondyla Apicalis Algorithm is used to optimize and identify the optimal parameters of the Ericsson propagation model. The API is one among many metaheuristic algorithms, which are natureinspired algorithms, the API is mainly based on the collaborative hunting concept of the Pachycondyla Apicalis ants. For this reason, the API algorithm is adopted to inspect the large space of solutions for locating the most appropriate parameters values of Ericsson propagation model to be more adapted to the real data and the environment of the eNB under study. The obtained results show that the API gives a good result and the proposed model offers near-real model for the problem of Path Loss estimation.

ABSTRACT. This article discusses the effect of laser emission power and PON network range when GPON and XGSPON systems coexist. The analysis aims to understand how these two factors influence the overall performance and stability of the coexistence between the two optical network technologies. In order to also meet the increasing demand for data rates and new services such as videoconferencing and very high quality television (4K, 6K, …etc). the coexistence system is preferred, allowing effective collaboration between GPON and XGSPON technologies. The work deals with the performance evaluation of GPON and XGSPON systems, focusing on reconciliations and adjustments related to BER error rates and Q factor. The study explores different scenarios by adjusting the laser power from -12 dBm to 12 dBm and varying the array range from 10 km to 120 km. The main objective is to determine the best combination of laser power and network range to ensure optimal performance for coexistence systems.

ABSTRACT. Free space optics (FSO) communication is an innovative and promising technology that effectively addresses the demand for high data rates and provides efficient last-mile connectivity solutions. The FSO communication links operate in an open environment, which is susceptible to various weather conditions such as fog, rain and haze. Among these weather factors, fog stands out as a significant contributor to severe attenuation in FSO links. This paper focuses on the estimation of optical attenuation for the transmission of color images in various fog events. The Kim and Kruse models are employed to analyze the impact of different fog conditions on the performance of the optical transmission system. The findings demonstrate that image quality is negatively affected as the transmission range increases. The maximum distances for acceptable image quality are 700 m, 500 m and 100 m in light, moderate and heavy fog conditions, respectively, according to the Kim model. However, the Kruse model predicts higher maximum distances of 900 m, 600 m and 200 m in light, moderate and heavy fog conditions, respectively.   

ABSTRACT. Cognitive radio (CR) is a game-changing technology that allows devices to adjust transmission and/or reception parameters in real time, improving wireless communication efficiency. Digital modulation identification (DMI) plays a crucial role in in CR, wireless communications, and spectrum management, providing adaptive and efficient signal processing. In this paper, we introduce a DMI approach designed for spatial modulation (SM) within the context of wireless multiple input multiple output (MIMO) technologies. In this study, we use the higher order statistics (HOS) as feature-based digital modulation identification (FB-DMI) to assess the performance of frequently utilized classifiers namely, k-nearest neighbor (KNN), and decision tree (DT). To train these classifiers, several configurations of the SM-MIMO network were evaluated, using the zero forcing (ZF) equalizer. Experiments show that KNN outperforms DT in identifying modulation schemes across all SNR levels.

ABSTRACT. The rapid advancement of networking and broadband communications has created a demand for efficient relaying methods, particularly in the context of high-fidelity and low latency last-mile communications. This paper explores the utilization of cellular-enabled-drones as Aerial-BS (Aerial-BS) in cellular networks next-generation, specifically focusing on their potential as flying base stations in Beyond 5G/6G networks. The study summarizes various aspects, including energy availability, mobility, optimal positioning, and node localization, handover efficiency. Additionally, multi-objective optimizations for bandwidth allocation, accommodating highly mobile User Equipment (UE), mitigating interference, and ensuring quality of service (QoS) are discussed. The paper highlights the use of drone-base-stations (DBS) as extending terrestrial network capacity and coverage, assisting mobile ad-hoc-networks, and enabling beam-forming applications. The challenges associated with optimal positioning, energy availability, path planning mobility, security, and QoS are examined, providing insights into future research directions. The findings contribute to the enhancement of cellular connectivity and offer valuable insights into the utilization of drones as flying aerial base stations in B5G / 6G network.

ABSTRACT. My work represents a novel proposed structure of AND/NAND logic gate based on photonic crystal ring resonator with nonlinear Kerr effect, Sicarbide is the nonlinear material applicated in this structure. The article went through two principal steps: the first step, we design and simulate a new simple filter resonator ring shaped has characterized by a high transmission 100% at wavelength of ʎ=1.5692 µm, the switching threshold power more than 0.7kw/µm2.After that, using this filter to design the logic gate. We found that it is characterized by the following: a high contrast ratio, short delay time and a good bit rate.

ABSTRACT. This study introduces a fault estimation observer for nonlinear systems, employing a proportional-integral approach, convex optimization techniques, and quadratic Lyapunov functions. The operational conditions of this system are expressed as linear matrix inequalities. It also incorporates an H$\infty$ performance criterion to enhance its robustness against sensor noise and external disturbances. This observer is adept at accurately estimating both the system's state and actuator faults, even in the presence of disturbances. The practicality and effectiveness of the proposed methodology are demonstrated through its implementation in a three-tank hydraulic system.

ABSTRACT. Terrains Recognition (TR) using Inertial Measurement Units has received special attention in recent years due to the widespread availability of wearable devices. In this study, we proposed a hardware implementation of a Terrains Recognition system on a Xilinx FPGA platform. The system uses a single IMU sensor placed on the shank and an MLP neural network classifier to classify five terrains (Walking, Stair descent/ascent, Ramp descent/ascent). The hardware design of the proposed system was optimized using parallelism optimization techniques such as Unroll, Pipeline and Memory partition. This approach led to a significant improvement in performance, with faster processing speeds.

ABSTRACT. Digital holography is becoming more and more popular as a 3D visualization and measurement technique in various fields. Particularly in the field of fluid mechanics where it is widely used in the detection and tracking of tracer particles commonly called DHPIV (Digital Holographic Particle Image Velocimetry). This is achieved by digitally reconstructing holo- grams, in which an important parameter is the reconstruction distance. Several autofocusing criteria have been proposed to obtain the best reconstruction distance in digital holography. In this work, we are interested in the evaluation of many families of autofocus operators: edge detection, sparsity measure, statistical texture image feature, spectral decomposition (Fourie, cosine, Welsh-Hadamard) and others in the presence of increased noise, i.e., by simulating on a real hologram the various noises which intervene at the time of their recording by the CCD sensors of the off-axis digital holograms. The analysis and the comparison of the criteria of auto-focus will be given on ideal holograms (without the addition of noise) and noisy holograms. Finally, we will robustness and confidence criterion for the best autofocusing criteria was also proposed. The experimental results show that some criteria are difficult to determine the location of these particles because of the effect of noise.

ABSTRACT. This paper introduces an enhanced approach for the control of hybrid sources, specifically fuel cells and batteries, within a hybrid electric vehicle. The proposed enhancement involves integrating the Gray Wolf Optimization algorithm with the passive-based control technique. The primary aim of this research is to enhance the operational characteristics of the studied system that is governed by the passivity-based control approach. This enhancement is achieved by optimizing the damping matrix parameter using the Gray Wolf Optimization algorithm. It is noteworthy that prior studies utilizing this technique did not address its impact, considering it a stochastic value ranging from 0 to 1. The central focus remains on proficiently and effectively managing the energy exchange between the two sources. Through Simulink-Matlab simulations, the outcomes underscore the influence of the damping matrix parameter on the overall dynamic performance of the system. Moreover, these results emphasize its pivotal role in achieving tracking precision and system robustness. This research constitutes a significant advancement in the realm of sophisticated control methodologies for hybrid power sources. Furthermore, it lays the groundwork for future investigations and the potential implementation of supplementary energy management strategies based on this innovative technique.

ABSTRACT. Real-time pedestrian detection is an expanding research topic that is crucial in various vision-based applications. Effective detection of pedestrians in various environments and situations will significantly improve human safety. However, the performance of current pedestrian detection algorithms faces challenges in scenarios with reduced visibility and blurred appearance, such as hazy weather conditions. This study addresses this issue by seeking an optimal balance between accuracy and speed for deploying deep learning-based pedestrian detection in real-time intelligent surveillance systems. For this aim, we used the transfer learning technique to fine-tune small versions of state-of-the-art YOLO models (YOLONAS-small, YOLOv8-small, YOLOv6-small, and YOLOv5-small) to perform pedestrian detection in hazy weather effectively. Experiment results demonstrate the effectiveness of our methods in accurately detecting pedestrians under hazy conditions by achieving high average precision and F-Score. Moreover, they all operate in real time, with impressive processing speeds ranging from 52.41 to 186.22 FPS.

ABSTRACT. The objective of this work is to operationalize and validate an acoustic platform designed for mini drones localization. This is achieved by deploying a microphone array ar- ranged in a circular configuration. The core methodology involves implementing the diagonal unloading algorithm for direction of arrival estimation, that allows accurate drone localization. Various simulations are conducted to evaluate the algorithm’s performance using Signal-to-Noise Ratio (SNR), the number of snapshots used for covariance matrix estimation and the reverberation time RT60. Comparative analysis is performed against considered algorithms such as MUSIC, MVDR, SRP and SRP-PHAT. Remarkably, the obtained results demonstrated that the diagonal unloading algorithm achieved comparable outcomes to the MUSIC algorithm while requiring less computational complexity. The validation process is further strengthened by experimental results obtained through practical implementation.

ABSTRACT. Deep neural network (DNN) models are often very large, which makes them computationally and memory intensive. However, many real-world problems require fast inference times. For example, machine vision applications demand real-time performance, with dozens of samples requiring inference every second. Additionally, many other applications rely on cloud inference computing, which can lead to overwhelming costs. The computational barrier during inference presents a significant challenge for DNNs to handle real-world use cases. In this talk, the focus will be on exploring different approaches to improve the computational performance and memory requirements of DNN models during the inference process.

ABSTRACT. The focus of this study concerns the design and modeling of power sources utilized in a hybrid vehicle system. Specifically, the principal source of energy is a Proton Exchange Membrane Fuel Cell (PEMFC), while an Ultra-Capacitor (UC) bank serves as an auxiliary source. This study is centered around the assessment of the mechanical and electrical performances of electric vehicles (EVs). In order to analyze the mechanical characteristics of a vehicle, the instantaneous velocity during its motion is measured using the Worldwide Harmonised Light Vehicles Test Procedure (WLTP), which is a recently adopted replacement to the New European Drive Cycle (NEDC). The mechanical and electrical performances that are necessary for the electric vehicle (EV) have a direct impact on the power demand, hence influencing the overall design of the system. During periods of high demand, the Proton Exchange Membrane Fuel Cell (PEMFC) is unable to meet the demand in isolation, necessitating the utilization of the UC bank to supply the excess power. The efficiencies of every power transformer, including the motor, gearbox, differential, and DC-DC converters, are considered in our work. The data analysis is conducted using the MATLAB software environment.

ABSTRACT. Plant diseases have a significant impact on agricultural productivity and global food security. Accurate and timely identification of plant diseases is crucial for effective disease management. In recent years, deep learning (DL) techniques, particularly convolutional neural networks (CNNs), have emerged as powerful tools for automated plant disease identification and classification. This comprehensive review aims to provide an in-depth analysis of the current state of deep learning-based approaches for plant disease diagnosis. We discuss the advantages of deep learning in learning discriminative features from raw image data, explore various model architectures, transfer learning strategies, data augmentation techniques, challenges related to labeled datasets, performance evaluation metrics, and limitations. Furthermore, we highlight the potential impact of deep learning in precision agriculture and its implications for crop health monitoring and disease management.

ABSTRACT. Passive RFID tag in UHF band using a reflector is proposed in this paper. The antenna consists of a metallic square loop, fed by two strips using a T-match structure in order to ensure matching in impedance between the chip and the antenna with a reflector. The distance between reflector and antenna is set to have a maximum gain. The proposed design shows an impedance bandwidth (IBW) ranges from 907 to 932 MHz (25 MHz). In addition, the tag has a long read range up to 22.56 meters with a high gain of 7.72 dBi at 915 MHz. Also, the proposed geometry shows a circularly polarized (CP) radiation. The proposed design is suitable for high gain, CP radiation RFID applications

ABSTRACT. Visual inspection of dual-energy X-ray radiographic images of cabin baggage requires high performance, but is hindered by various challenges such as low target prevalence, variability in target visibility, possible presence of multiple targets, and security personnel fatigue and inattention. Artificial intelligence (AI) techniques, particularly deep convolutional neural networks (CNNs), have shown promise in improving the automatic detection of explosives, even with low-resolution radiographic images, especially in high baggage throughput scenarios. In this paper, we focus on the detection of detonators as components of improvised explosive devices. The proposed approach involves comparing two experiments implemented in a deep CNN architecture using TensorFlow and Keras libraries. In the first experiment, raw dual-energy radiographic images without any enhancement were used. The second experiment includes three methods for contrast enhancement and feature extraction: the Contrast Limited Adaptive Histogram Equalization (CLAHE) method, the wavelet transform-based method, and the mixed CLAHE RGB-Wavelet method. In the latter two methods, Haar, Db2, Coif2, and Sym2 mother wavelet functions at two levels (HH and HL) were employed. The analysis of results focuses on a comparative study of performance measures such as accuracy, precision, recall, and F1 score. It was found that the pre-processing methods used in experiment 2, for the two evaluated classes (detonator and no detonator), achieved higher accuracy compared to the raw radiographic images used in experiment 1 (98.08%). The highest accuracies in experiment 2, with a value of 100%, were obtained with the CLAHE method (green channel in grayscale, blue channel in grayscale, and RGB channels) and the wavelet transform method with Haar mother wavelet at two levels HL.

ABSTRACT. Modern 5G telecommunications systems tend to use higher frequencies. The design of such devices is becoming increasingly miniaturized for more powerful applications and technologies. In this context, Substrate Integrated Waveguides (SIW) replace conventional waveguides. Their design is mainly based on mathematical models. However, the resulting model presents a frequency shift with respect to the working frequency. In this paper we present an optimization method for ensuring the good performance of waveguides, based on SIW technology. These waveguides are easily integrated into 5G telecommunication systems. We'll then look at the different results obtained after changing the variables (asiw and h).

ABSTRACT. Interferometric synthetic aperture radar (inSAR) phase image is a key for the digital elevation model of earth mapping. To achieve this three-dimensional reconstruction, phase unwrapping process must be performed whitch is used to calculate the accurate elevations from the wrapped phase map. For the noise-free images, this process is just a simple integration of the wrapped gradient. But in reality, there is no phase image without noise, therefore the phase unwrapping have to be adaptive with a strong immunity to noise. Several adaptive algorithms have been proposed in such area where Goldstein’s branch-cut and Flynn’s quality-guided are the widely known in the path-following category, they are the most used methods and all other propositions are just enhancements or hybridizations. In this paper, we analyze the performance of each one and provide the substantial difference between them. Both algorithms are implemented using simulated and real inSAR data of different patterns and are analyzed under several relevant criteria.

ABSTRACT. The electrocardiogram (ECG) is a physiological signal widely used in clinical settings to assess and interpret patients' cardiac health. Accurate QRS detection is an essential first step for every ECG signal analysis algorithm. However, QRS detection is a complicated task due to the high diversity of QRS complexes and overlap from different noise types and cardiac abnormalities. This work focuses on analyzing and segmenting electrocardiographic signals. First, ECG signals were filtered by applying a high-pass filter to remove baseline drift and a median filter to reduce high frequency Gaussian white noise (GWN). After this, we implemented a QRS complex detection algorithm based on the conjoint use of the empirical mode decomposition method (EMD) and Hilbert’s enhanced approximate envelope. To demonstrate the efficiency of the R- wave detection algorithm, we have applied it to some segments from the MIT-BIH database containing various disturbances and arrhythmic pathologies. Finally, all characteristic waves of ECG signal have been well extracted. Keywords— ECG signal, Segmen

ABSTRACT. The diagnosis and monitoring of industrial machinery faults is crucial to the Industrial Revolution but is frequently difficult and labor-intensive. Because of its quick computation, higher accuracy, and robustness in performance, the use of artificial intelligence techniques has become a crucial component of condition monitoring of mechanical and electrical machines, this paper aims to use the artificial intelligence (neural network technique) to be able to detect and identify air gap eccentricity (static and mixed) and its influence on the stator current signature, the aim is to demonstrate the reliability and accuracy of these techniques to help us make the decision, The achieved simulation results allow distinguishing the types of eccentricity based on the Motor Current Signature Analysis technique data.

ABSTRACT. Wheat is a significant staple crop worldwide, but its cultivation is challenged by various factors such as pests, diseases, climate change, lack of water, and yield enhancement. To prevent wheat infections, nondestructive and self-sufficient remote detection methods are necessary, which are sensitive and reliable. Hyperspectral imaging with deep learning is a powerful method for identifying diseases early, mapping them accurately, and evaluating their severity in plants in an unbiased way. This method involves data collection, preprocessing, and analysis of Hyperspectral images to identify wheat diseases. The article also presents several Hyperspectral imaging datasets and relevant research on using this technology for detecting wheat illnesses. The integration of spectroscopy and imaging in Hyperspectral imaging represents a novel approach to achieving an in-depth and comprehensive understanding of plant life. Moreover, deep learning, a subset of artificial intelligence, has gained considerable attention due to its ability to process vast and intricate data sets. By synergistically merging these two advanced technologies, researchers and scientists can significantly enhance the analysis of hyperspectral imagery, thereby enabling improved crop monitoring, disease identification, and yield projection in wheat farming This article summarizes the use of deep learning techniques in Hyperspectral imaging for wheat crops. Hyperspectral imaging is a non-destructive way to study crops but generates a lot of data that can be difficult to analyze. Deep learning automates the analysis process and improves accuracy. However, it requires large and diverse datasets, complex model development, and computational resources. This report provides information on the benefits and challenges of using deep learning in Hyperspectral imaging for wheat crops.

ABSTRACT. Improving patient outcomes requires early and precise lung cancer classification. Deep learning has been shown to be successful in the interpretation of medical pictures, particularly when convolutional neural networks (CNNs) are used. In this paper, we proposed a deep modified CNN network using three pre-trained models (Densenet169, MobileNetV2, and Resnet50V2) to improve lung cancer classification performance based on the IQ-OTH/NCCD Lung Cancer dataset. The experimental results show that our modified Densenet169 based strategy outperformed existing methods, earning the highest accuracy rates for lung cancer classification. Early identification of lung cancer can enhance patient prognosis and treatment options dramatically. This report outlines future research and advancement opportunities in this vital topic.

ABSTRACT. Person re-identification, a critical task in computer vision and surveillance, aims to identify and track individuals across different camera views or frames. Is it commonly used in video surveillance and security applications, the primary challenge in person re-identification is the significant variability in appearance due to changing lighting condition, pose variations, occlusions, and camera viewpoints. several methods and techniques have been developed for (person re-identification) this critical task. In this work we propose new framework by combining deep-learned feature of Convolutional Neural Network (CNN), with Local Maximal Occurrence (LOMO), and the hierarchal Gaussian algorithm (GOG) descriptors, are all used to get richer information to match people in across cameras. We evaluate the performance of our approach on four standard benchmark datasets: small scale (VIPeR), and large scale (GRID), (PRID450S) and (CUHK01), validates the effectiveness of our proposal. and experimental results show that our method is superior to the state-of-the-art approaches. Cross-view Quadratic Discriminant Analysis (XQDA) is successful metric learning model.

ABSTRACT. Antenna technology has undergone significant ad- vancements with the integration of innovative techniques such as MIMO (Multiple Input Multiple Output) and phased array systems. By utilizing multiple antennas for transmission and reception, MIMO enhances data rates, spectral efficiency, and overall system performance. Phased array antennas, on the other hand, introduce beamforming capabilities, allowing for dynamic control and steering of radio wave beams without physical movement. The goal of this work is to optimize the control of the antenna network (MIMO) from the perspective of directivity and secondary lobe reduction using the API metaheuristic algorithm.

ABSTRACT. In this study, we propose a brushless direct current (BLDC) motor speed controller for a quadrotor. These motors are commonly used in quadrotors due to their high power-to-weight ratio and precise control capabilities, which allow for independent speed adjustment of each motor to control the quadrotor's rotation direction, track trajectories, and execute necessary flight maneuvers. The main objective of this study is to regulate the speed of the BLDC motor to achieve optimal response performance using artificial neural networks (ANN). First, we tuned the gain values of the classical PID controller using the particle swarm optimization (PSO) algorithm within a closed loop to achieve optimal parameter values and reduce error for stability during the tracking of the desired speed signal. Subsequently, data regarding the input and output values of the PID controller were collected and stored within the MATLAB workspace. Next, we created a new controller trained with this data using an artificial neural network (ANN) and implemented it in the BLDC motor model in SIMULINK-MATLAB. Finally, the results demonstrate that the ANN-trained controller effectively controls the required motor speed.

ABSTRACT. Quantification of the Autonomic Nervous System (ANS) helps detect pathological patterns through the study of the normal and abnormal HRV signals, which can char- acterize Sympathetic and parasympathetic alterations in ANS, through the low–frequency (LF) band related to sympathetic activity and the high–frequency (HF) band which is a marker of parasympathetic activity. The analysis of these bandwidths was carried out on time–frequency distribution distributions for normal sinus rhythm (NSR) of men and women. Therefore, in this study, we calculated the features related to the energy of low–frequency (LF) and high–frequency (HF) components and the ratio between them denoted as LF/HF ratio in 23 normal subjects to know the variation of LF/HF ratio between men and women by studying their HRV signal within Quadratic time–frequency. Indeed, We defined Energy features, denoted as ELF and EHF, which have been calculated within low–frequency and high–frequency bandwidths within quadratic time–frequency representations calculated using the STFT–spectrogram (SP) of HRV signals. The HRV signals were detected by localizing R–peaks within ECG records of the MIT–BIH Arrhythmia Database of the PhysioNet research repository. As a result, we obtained that women have a high value of the mean of the energy ratio (LF/HF) calculated within LF and HF ranges ( LF/HF = 1.09 ± 1.00) compared to the men’s group (LF/HF = 0.56 ± 0.53) in the normal rhythm, which confirms that the sympathetic activity of women is higher than that of men.

ABSTRACT. In the rapidly advancing domain of wireless communication networks, the delivery of high-speed data services has entered a heightened phase of competition. As the demand for uninterrupted user connectivity experiences continual growth, the precise prediction of downlink user throughput emerges as a pivotal challenge within the field of network optimization. In this research endeavor, we harness the capabilities of cuttingedge machine learning techniques to discern the most pertinent and impactful Key Performance Indicators (KPIs) governing the evolution of downlink throughput. Our overarching objective is to construct a pioneering predictive model with the acumen to accurately estimate throughput, grounded in these indispensable KPIs.

ABSTRACT. This paper describes an approach for design a new observer based on the mean value theorem and analyzing the stability of a continuous-time Takagi-Sugeno fuzzy system with unmeasurable premise variables through the line-integral Lyapunov function approach. The stability conditions are more general and relaxed than those obtained through conventional quadratic Lyapunov functions, which can be expressed as linear matrix inequality forms, these contraints can be solved using linear programming approch via specialized software tools. A implementation on Flexible joint robot was performed, resulting in findings that clearly demonstrate the effectiveness and efficiency of the proposed method.

ABSTRACT. the field of biometrics research continues to advance, and although many identification and verification methods have been established, their effectiveness is still insufficient due to the increasing demands on security. For the purposes of biometric identification or authentication applications, hand biometric recognition has been effectively established. Multiple biometric methods are utilized since using only one biometric method often decreases the systems' confidence. In the present article, we suggest a multi-biometric fusion method for person recognition. In fact, we employ two different forms of biometrics: the hand's shape and palm print. The texture was extracted using discrete cosine transform (DCT) method. Subsequently, PSO algorithm is employed in the process of identification to integrate the geometric properties of the hands with the texture of the examined palm. Simulation results demonstrate the key benefits of this strategy with a realization rate of over 85% for a population of 230 people.

ABSTRACT. Autonomous landing system are indispensable for unlocking the complete potential of unmanned aerial vehicle (UAV) without the necessity of human intervention in various industrial and commercial applications. In this research, we present a quadrotor Landing system based on visual navigation. We investigated a method for safely landing a drone in scenarios where GPS signals are not available. This approach, enables an autonomous and efficient landing of UAV on a predetermined landing pad, which effectively minimizes the risks associated with the physical testing phase using a Gazebo-based simulation, our approach not only ensures the overall fulfilment of design requirements for the autonomous landing system but also provides a valuable tool for secure parameter adjustment. Finally, the simulation results conclusively demonstrate the robust and precise performance of our suggested landing system.

ABSTRACT. A multimodal biometric system combines information from multiple biometric sources to overcome the limitations of each individual biometric system. This paper presents an efficient algorithm for extracting details based on the image representation of fingerprints. Additionally, we conduct an in-depth investigation of T-Norm algorithms and offer a comprehensive score-level merging analysis. This analysis combines matching outcomes derived from the left and right fingerprints using the Hammacher, Schweizer-Sklar, Dombi, and Yager algorithms. The latter approaches are compared based on the receiver operating characteristic curve. The experimental evaluation conducted on the NIST fingerprint database validates the efficacy of fusion at the score level. Furthermore, it demonstrates the effectiveness of the proposed algorithm when compared to other T-Norms.

ABSTRACT. This paper presents an efficient method to optimize the energy absorbed by an induction motor in transient state via optimal control, under the field-oriented control technique. This approach consists of minimizing a cost function by gathering the sum of the copper losses inside the machine, and as a consequence, an optimal rotor flux used in the direct field oriented control is deduced. This set value is obtained by taking into account some dynamic constraints linking flux and speed through a reduced model of the machine. The effectiveness of the proposed strategy is verified by several simulation tests using MATLAB/Simulink.

ABSTRACT. During the last few years, remote sensing is considerably used for Earth observation for the environment and sustainable development. The temporal classes of satellite images provide better information for monitoring the earth's surface at different scales; therefore these images are becoming a very relevant opportunity of investigating. The numerical analyses of image processing are based on the artificial intelligence (AI) tool rather than the traditional methods. The environmental indicators from these images offer very important statistics, the carried out work of dynamic phenomena, the observation event and interpretation of evolving circumstances like: atmospheric conditions, crises and natural disasters taking the opportunity to be studied, several problems come out of deforestation and monitoring of water resources (water bodies). The purpose of this study is focused on drawing out details from the database used and then to give an offer for a texture analysis strategy. Firstly, using specific development and then developing the existing software in the second hand. This work is to establish an algorithm for the detection of region of interest (water body) which is based on the theory of fuzzy logic and the hypothesis of Fuzzy C-Means.

ABSTRACT. To identify depth intervals within reservoirs that have a high likelihood of containing hydrocarbons and possess adequate permeability for efficient hydrocarbon extraction, the examination of well cores is crucial. Given the heterogeneous nature of petrophysical properties across different locations, a single core per area is insufficient, leading to higher expenses due to the need for multiple cores. This research suggests the application of machine learning techniques, particularly weak learners, to predict lithofacies using well log data. The results obtained show the effectiveness of the hybrid weak learners approach, even when dealing with limited data.

ABSTRACT. As a pivotal resource, social media empowers the real-time identification of disasters through content created by users. This content is frequently abundant and open to interpretation, underscoring the pivotal role of artificial intelligence in categorizing disaster-related information to bolster immediate response systems. In our study, we leveraged about 5400 tweets from the CrisisLEX T26 dataset, conducting both binary and multiclass classification tests to enhance the research outcomes. We have compared the performance of various PLM-based models tailored to this dataset. In our pursuit of effective disaster detection, we propose two approaches. First, our PLM-BiLSTM-CNN model integrates PLMs, Bidirectional Recurrent Neural Networks (BiLSTM), and Convolutional Neural Networks (CNN), enabling disaster detection with commendable results: an 88.14% F1-Score in multiclass classification and a 90.83% F1-Score in binary classification. In the second approach, we fine-tune PLMs using our considered dataset and utilize their outputs as an initial input for our PLMBiLSTM-CNN model. This sequential process enhances disaster detection capabilities by harnessing the strengths of both finetuned PLMs and our PLM-BiLSTM-CNN architecture.

ABSTRACT. —this article is focused on desining a convenient controller that is created with an acceptable approach for PEMFC HEVs. Nonlinear controllers can be helpful for obtaining dynamic capabilities, but they require a lot of complexity. An alternate linear solution is the fractional-order proportionalintegral (FOPI) controller, which is a more competent variant of the commonly used linear PI controller. In this work, a FOPI controller was employed to obtain optimal performance for a three level interleaved boost converter (IBC) using a unique metaheuristic technique that incorporates both Atom Search Optimization and simulated Prairie Dogs Optimization algorithms (ASOPDO). The created algorithm was validated by conducting relevant analysis on a Simulink model of the system and nonparametric statistical tests revealed its higher capacity compared to other techniques. This allowed for superior closed-loop output voltage control performance of the IBC in terms of time and frequency domain responses as well as disturbance rejection. The suggested PDO-based FOPI controller suitable one for your simulation tests terms of efficiency, flexibility, reliability.

ABSTRACT. Diabetic retinopathy (DR) is an ocular disease caused by long-term diabetes. It is the leading cause of blindness among working-age people. DR is an asymptomatic pathology, that causes changes in the area, diameter, and tortuosity of thin blood vessels. Therefore, analyzing the morphological changes of the retinal vasculature is critical for helping ophthalmologists in the diagnosis of DR at an early stage and preventing vision loss. In this work, we propose an automatic approach for diabetic retinopathy screening from color retinal images based on semantic segmentation of retinal blood vessels. First, a preprocessing and data augmentation steps are integrated to extract the green channel and increase the number of images. Then, the segmentation step is carried out using a modified U-Net network. After that, a set of geometrical parameters are calculated from segmented images such as area, perimeter and tortuosity. Finally, Artificial Metaplasticity Multi Layer Perceptron ‘AMMLP’ classifier is employed to discriminate between healthy and DR images. The proposed method for vessel segmentation was evaluated on the DRIVE database, and it achieved an accuracy of 99.65%. For DR classification, AMMLP got a sensitivity of 98.45% on Messidor dataset. The experimental results show that our approach outperforms the existing methods, which can assist ophthalmologists in making an early and accurate DR diagnosis.

ABSTRACT. This article presents the optimization and simulation of three novel metamaterial unit cells utilizing CST Microwave Studio Suite 2019, our approach is rooted in comprehending the operational mechanisms of traditional Split Ring Resonator (SRR) and Complementary Split Ring Resonator (CSRR) cells, the design process involves adjusting the ring dimensions, inter- connecting metallic stubs,splitting gaps, and inter-ring distances to achieve the satellite applications. The designed ENG, MNG, and DNG cells are simulated within a waveguide environment with carefully selected boundary conditions. The Scattering parameters are converted using the Smith technique into Matrix Laboratory (MATLAB) code to extract the EM characteristic of MTM. A Rogers is the substrate, and 0.035-mm PEC is the conductive material for designing of the unit cells. the obtained results confirm that the suggested unit cells are promising candidates for satellite communications.

ABSTRACT. Conditional maintenance, grounded on industrial monitoring and diagnostic methodologies, has become indispensable. In this context, the predominant technique is vibration analysis, wherein the data analysis phase holds utmost significance. This crucial stage is conducted through temporal, frequency, and time-frequency methods. The objectives of this study are to examine the efficacy of temporal methods (RMS and Kurtosis) in early-stage detection of bearing faults, along with exploring influential factors such as rotation speed, sensor positions, load, and the types and dimensions of faults. To achieve this, various combinations of these factors have been explored to assess the effectiveness of these scalar parameters under diverse conditions. The case study demonstrates that kurtosis emerges as the most effective scalar in early-stage identification of faults, in contrast to RMS, which proves more effective for high-speed scenarios but serves as a late indicator. Moreover, the complementary of these methods is crucial for fault identification and facilitates informed decisions within the context of condition-based maintenance policies

ABSTRACT. Feature point based registration technique is widely used for aligning and registering images by identifying and matching distinctive key points features between them. The purpose is to establish a geometrical transformation between two images to be compared that spatially aligns each point from one image to its corresponding of the other image. Recently, local feature key-points such as scale invariant feature transform (SIFT), speeded-up robust features (SURF) and oriented FAST and rotated BRIEF (ORB) are widely used due to its inherent properties such as invariance, changes in illumination, noise. This study provides an automated feature points based registration technique applied to satellite images. The registration task is solved by an iterative heuristic optimization method namely genetic algorithm (GA) to find the best transformation parameters through minimizing a distance between the two feature point sets. An adaptive GAs based on fitness sharing and elitism techniques are performed to enhance the capability of GAs. A comparative study is conducted between different types of the extracted feature techniques. The results demonstrates the effectiveness of SIFT algorithm compared to the other competing techniques for registering satellite images.

ABSTRACT. Despite recent advances in detection and recognition models, the development of robust and accurate algorithms for describing and classifying object visual content in real-life scenarios remains a significant challenge. This paper introduces an object part appearance module, OPA, within the framework OPA-Yolo, specifically designed for occluded object detection. Our approach combines handcrafted features with deep ones across network layers to capture the visual content parts of occulted object classes. Mapped deep features identify different regions of interest, characterizing object part appearances with a significant amount of statistical information. To enrich this feature representation with invariant local features, we train the network to learn an optimal combination of object appearance, providing it with an additional attention mechanism. The model is built on a recent YOLO object detector architecture. Evaluated on the PASCAL-VOC dataset, OPA-Yolo achieved an average detection rate of approximately 24.76% on occulted objects, with some objects having an occlusion ratio of up to 60%.

ABSTRACT. This paper comprehensively overviews image and signal processing, including their fundamentals, advanced techniques, and applications. Image processing involves analyzing and manipulating digital images, while signal processing focuses on analyzing and interpreting signals in various domains. The fundamentals encompass digital signal representation, Fourier analysis, wavelet transforms, filtering, and noise removal. Advanced techniques, such as deep learning for image classification and object detection, are explored. Image and signal processing applications include computer vision, medical imaging, audio processing, and communications. This paper is a valuable resource for understanding image and signal processing principles and applications, fostering further research and development in these fields.

ABSTRACT. Breast cancer, characterized by uncontrolled cell division in breast tissue, affects individuals of all genders. Risk factors encompass age, genetics, hormones, and lifestyle. Symptoms comprise lumps, breast and nipple changes, and skin abnormalities. Early detection via self-exams, clinical exams, and mammograms is crucial. Artificial intelligence (AI) in breast cancer detection systems utilizes advanced algorithms and image analysis techniques to aid radiologists in identifying potential cancerous abnormalities, enhancing accuracy, and facilitating earlier diagnosis. In this paper, various feature fusion methods were proposed for binary classification of breast cancer. The first technique involves concatenating the flattened layers of VGG19 and Xception models. The second technique removes the fully connected layers of both VGG19 and Xception models, concatenates their outputs, and applies an Encoder to reduce spatial dimensions and extract hierarchical features. Both Flatten and GlobalAveragePooling2D are utilized to convert a 2D feature map into a 1D vector and choose the best ones. Our approach achieved the best results with an accuracy of 94.61% for classification and a dice score of 72% for segmentation task.

ABSTRACT. The quest for improving human lifestyle has driven the development of novel concepts and technologies aimed at enhancing the quality of life. Among these concepts is a Reinforcement Learning (RL) method, a division of Machine Learning (ML). RL’s current applications have demonstrated remarkable achievements in various domains, including Video Games, Finance, Medicine, Healthcare, and Robotics. This work gives a succinct contextual overview of RL and conducts a comprehensive review of its recent research and applications in the realm of medical imaging.

ABSTRACT. In recent years, the application of artificial intelligence (AI) techniques for fire detection has gained significant attention due to its potential for enhancing early fire detection systems. This study aims to compare the performance of deep learning convolutional neural networks (CNN) and support vector machine (SVM) machine learning algorithms in the context of fire detection.We present a comprehensive analysis and evaluation of the two approaches, highlighting their strengths and weaknesses, and discussing their potential for real-world fire detection applications.

ABSTRACT. Breast cancer is the main lead of women's cancer-related mortalities. Therefore, early detection is imperative for preventing breast cancer from developing to advanced stages. Moreover, the emergence of Computer-aided diagnosis systems combined with Deep Learning techniques improve breast cancer diagnosis at its early stages. This study proposes a system for classifying mammogram images scanned from different views based on Deep learning techniques. In this paper, we propose combining four pre-trained models in one system employing the Transfer Learning technique with different input images of the same lesion. This proposed system provides a binary classification of the lesion into malignant or benign classes. The Mini-DDSM dataset images were used in this study; the images were pre-processed and augmented to achieve the best results. Several well-known pre-trained models for image classification (ResNet, Xception, DenseNet, MobileNet and Inception) were experimented with for the proposed model construction. The study results showed that our proposed model achieved an accuracy of 94,18%, a precision of 94,26%, a recall of 94,16%, and an F1-Score of 94,21%, outperforming the pretrained model on classifying mammogram images of the same dataset.

ABSTRACT. Numerous modern Computer Assisted Diagnosis (CAD) systems often make use of a Convolutionanl Neural Network (CNN) architecture for early breast cancer diagno- sis and for cancer classification. However, several CNN-based architectures have been developed with various features and characteristics. This study aims to compare and suggest some CNN-based architectures for breast cancer classification on histopathological images. Eight models based on different CNN architectures are covered by this study, such as CNN, VGG 19, ResNet 152, MobileNet V2, Inception V3, DenseNet and AlexNet architectures, while the eighth model is a hybridization between three architectures (VGG, DenseNet and MobileNet). All of the models are trained on the same dataset (the BreaKHis dataset) and the comparison is performed using the Accuracy, Recall, Precision and F1 Score metrics. The results of this study are extensively discussed by pinpointing the distinctive characteristics of each model and their own inherent strengths and weaknesses. Finally, some suggestions regarding the best models are outlined at the end of this paper.

ABSTRACT. Free Space Optics (FSO) technologies are gaining recognition as promising solutions to handle the increasing bandwidth demand while ensuring a high level of security. FSO systems, which utilize directional lasers and operate within a Line-of-Sight (LOS), provide information security on par with optical fiber systems. Moreover, FSO systems offer a cost-effective alternative derived from mature fiber optic telecommunications technology, making them an appealing option for resolving last mile challenges in urban telecommunications. To enhance accessibility for a larger user base, Non-Orthogonal Multiple Access (NOMA) techniques can be considered for implementation on FSO systems. Unlike Orthogonal Multiple Access (OMA) techniques, NOMA techniques allow a less complex design of telecommunications systems while maintaining high transmission rates despite a lower Signal-to-Noise Ratio (SNR). However, this decrease is unsuitable for FSO systems since they already experience low SNR, mainly due to atmospheric scintillation. Hence, exploring the integration of both OMA and NOMA techniques becomes intriguing, as it holds the potential to create a simplified FSO telecommunications system while simultaneously increasing the user capacity and transmission rate. In this study, we investigate the effectiveness of applying PD-NOMA technique to a 1x2 FSO-OCDMA transmission link affected by scintillation. Monte-Carlo simulations were conducted to calculate the Bit Error Rate (BER) under various atmospheric conditions. The results indicate that a 1x2 FSO-PD-NOMA transmission link can maintain reliability up to moderate atmospheric scintillation levels which could potentially double the number of users of an FSO-OCDMA telecommunication system. However, this transmission link becomes unsustainable under strong atmospheric scintillation conditions unless a trade-off is made by decreasing the transmission rate.

ABSTRACT. Abstract—developing and employing methods to reduce the mutual coupling between elements of an antenna array has become a hot topic in the design of antenna arrays. The utilization of electromagnetic band-gap (EBG) structures is an attractive way to reduce surface wave excitation in printed antenna geometries so to mitigate the mutual coupling problem. In this article, the proposed antenna array operating at 2.4 GHz is comprised of two rectangular patches and verious EBG structures, EBG structures are printed on substrate between two rectangular patches. The FEM based EM simulator HFSS is used to characterize the EBG structure. Simulation results show that a significant value of mutual coupling reduction, more than 13 dB, obtained by using the proposed structures.

ABSTRACT. Human Activity Recognition (HAR) is the process of interpreting human actions from sensor data. This paper presents a hybrid approach for HAR utilizing Convolutional Neural Network (CNN) for feature extraction and Support Vector Machine (SVM) for classification. The model is end-to-end trainable, where the SVM classifier replaces the softmax layer of the CNN. Evaluation of the approach was conducted on two benchmark datasets, UCI HAR and UniMiB SHAR, achieving accuracies of 96.13% and 87.85%, respectively. These results surpass those reported in the state-of-the-art and demonstrate the effectiveness of the proposed approach in interpreting human activities.

ABSTRACT. In this work, a new compact single-device array antenna is proposed for Ka-band Internet applications. The proposed model has more than the ability to work without the need for routing with characteristics that allow to improve the performance of the antenna. The designed antenna was modeled according to the needs of the Algerian satellite currently operating ALCOMSAT, with all antenna charac-teristics (impedance bandwidth, gain and radiation pattern) extracted.

ABSTRACT. This paper aims to improve the detection performance of the conventional mean-level Constant False Alarm Rate (CFAR) detectors in homogeneous and nonhomogeneous Gaussian environments. This can be done by the design of the New Weighted Cell Averaging (NWCA-CFAR) detector. Assuming that the primary target follows a Swerling I (SW I) type, closed forms of the detection (Pd) and the false alarm (Pfa) probabilities of this detector are obtained. Optimum pairs of the weights that maximize the Pd and guarantee a constant Pfa are also obtained. Monte Carlo simulation results have shown that this detector has improved the performance of the detection and the regulation of the Pfa of the conventional ones.

ABSTRACT. This paper presents the development of various forms of photonic crystal fibers (PCFs), including hexagonal, square, and circular structural shapes. The hexagonal and square structures studied in this work are composed of five circular air holes, while the circular structure consists of six air holes around a solid core on a silica background. To analyze the optical properties of these proposed fibers, such as effective index, chromatic dispersion, and confinement losses, Finite-Difference Time-Domain (FDTD) method with perfectly matched layers was employed. A comparison among the different fiber structures was conducted. The results of this comparison revealed that, at a wavelength of 1.55 μm, the hexagonal PCF exhibited the lowest confinement loss value, measuring at -1.5498e-19 dB/m. On the other hand, the circular PCFs displayed the minimal value of chromatic dispersion at 7.0168e-5 ps/(km·nm).

ABSTRACT. This study presents an innovative ultra-miniaturized antenna system designed for 2.45 GHz ISM band operation, tailored to the distinct requirements of biomedical implants. The proposed antenna boasts an exceptional level of compactness, with a mere 3.3 mm³ in volume and physical dimensions of 3 mm × 3 mm × 0.37 mm. In a three-layer body model incorporating skin, fat, and muscle, the antenna system achieves an impressive peak gain of -25.4 dB at 2.45 GHz while maintaining a remarkable impedance bandwidth of -10 dB, spanning 33.8%. This accomplishment is particularly noteworthy, given the antenna's compact design. Additionally, it exhibits outstanding omnidirectional radiation patterns. Furthermore, an evaluation and analysis of the antenna's specific absorption rate (SAR) are conducted. These results emphasize the practicality and potential applications of this antenna system within biomedical devices.

ABSTRACT. Steel is one of the metals used in most industrial sectors. Any damage to steel will have serious consequences on the industry's economy. Therefore, defect detection has become one of the most important and competitive tasks in steel quality control, either through classification or defect detection, which combines classification and localization. This paper proposes a new architecture for steel surface defect classification based on the modified mobilenet v3 as a backbone to extract the semantic information along the horizontal and vertical spatial direction that is missing extracted with the local receptive field or the global receptive field, moreover, the latter is integrated into the path aggregation network which enhanced by an improved double bottom-up pathway to aggregate features with multiple resolutions to achieve strong classification capability, our experiments in neu-det show the classification improvement of the proposed architecture with other CNN’s architectures.

ABSTRACT. Kinship verification from face images is a novel and formidable challenge in the realms of pattern recognition and computer vision, offering numerous potential applications, including image annotation, forensic analysis, and social media research. This work makes notable contributions through the use a preprocessing technique known as Multiscale Retinex (MSR), which enhances image quality and improves contrast. Our approach harnesses the strength of complementary deep and shallow texture descriptors by combining them at the score level using Logistic Regression (LR) technique. Specifically, we employ the Local Phase Quantization (LPQ) descriptor for non-deep (shallow) texture features, while for deep features, we leverage the VGG16 model. We assess the effectiveness of our approach by conducting comprehensive experiments on three challenging kinship datasets: Cornell KinFace, UB KinFace and TSKinFace.

ABSTRACT. Over the past ten years, breast cancer (BC) has become a major health concern due to its high fatality rate. It remains the most common cancer in women to be diagnosed. The urgency to address this disease has prompted extensive research efforts toward developing effective classification models and diagnostic tools. With an emphasis on the study of classification outcomes utilizing the well-known WDBC dataset, we examine the efficacy of several feature selection (FS) methods and classifiers within the context of the majority voting ensemble technique in this work. We explore a diverse set of FS algorithms, including PCA, Relief, RF, LASSO, GA, and CFS, combined with various classifiers such as SVM, MLP, and RF. Our study reveals that the PCA FS algorithm consistently achieves high accuracy results across multiple classifiers, notably 99.12\% accuracy when paired with RF and LR. This consistency highlights its effectiveness in capturing informative features and its compatibility with the majority voting ensemble technique, which also achieved an accuracy of 99.12\%. These findings contribute to advancing our understanding of ensemble techniques and FS algorithms, particularly in the context of the WDBC dataset, providing valuable insights for future development of robust classification models for similar datasets.

ABSTRACT. With the rapid proliferation of internet technology, social media platforms, particularly microblogging sites like X (ex Twitter), have emerged as essential channels for news dissemination. However, this growth has been shadowed by the rampant spread of fake news and rumors, largely due to the lack of stringent oversight. While early research strategies employed traditional machine learning to address this challenge, the sheer volume of data in recent times has made deep learning methods increasingly relevant. In this study, we introduce a rumor detection model that harnesses the capabilities of fine-tuned pre-trained language models. Our primary model achieved an F1-score of 92.14%. For comparison, we also developed a model that combines a pre-trained language model with a logistic regression classifier, obtaining an F1-score of 91.94%. Another model we evaluated combined pre-trained language models for feature extraction with a Multi-Layer Perceptron (MLP) for classification and registered an F1-score of 92.28%. Intriguingly, our results underscore the potential of straightforward classifiers that leverage rich vectors from pre-trained language models, showing they can rival or even outperform complex neural architectures in rumor detection tasks. This revelation points towards a more efficient and less resource-intensive strategy to tackle online misinformation.

ABSTRACT. Human Activity Recognition using Inertial Measurement Units has received particular attention in recent years due to the widespread availability of wearable devices. In this article we present a Convolutional Neural Network-based technique for HAR using IMUs to recognize six human activities. Data from 30 subject was used to validate the system using UCI HAR Dataset. The time series signals were filtered, segmented then transformed into a two-dimensional matrix format, where each row represents a specific time step and each column corresponds to the IMU axes; these data were then fed into the convolutional neural network as input. The results of the experiments demonstrate that the suggested model can effectively capture the spatial and temporal features of IMU data, leading to superior recognition accuracy and relatively lower complexity in comparison to conventional 2-dimensional CNN approaches.

ABSTRACT. Automatic classification of different skin lesions from dermoscopic images is a challenging task due to the high similarity in visual features among various lesion types in terms of size, shape, texture, and colour. Other problems include the existence of artefacts in dermoscopic images, lack of data, and training of deep architectures requiring millions of parameters, which usually lead to overfitting and weak generalization. Our work aims to develop a computer-aided diagnosis (CAD) that can accurately classify different skin lesions using a public ISIC 2019 challenge dataset and a private dataset. In this research, three main tasks are proposed and implemented. Task 1 is data quality improvement by solving the imbalanced class problem, missing values, and dermoscopic multi resolutions. Task 2 is CAD development via a pretrained Noisy Student (EfficientNet-L2) architecture as a feature extractor using transfer learning. We incorporate additional metadata using a dense neural network concatenated with the CNN output. Then, the classifier follows with eight units representing skin lesion classes and the unknown class. Task 3 concerns the development of automatic hyperparameter selection (CNN-AHPS) using big data tools (MapReduce). This system can help to determine the best configuration of training hyperparameters for our combined architecture.The experimental results illustrate that the use of metadata in classification improves the balanced multiclass accuracy and F1-score from 46.9% and 67.0% to 55.3% and 69.0%, respectively. Additionally, CNN-AHPS improves on default configurations, yielding a balanced multiclass accuracy and F1-score of 69.4\% and 87.7%. The experimental results confirming the effectiveness of NS-EfficientNet-L2 to classify skin lesions outperforming the state-of-the-art methods including the winner of the ISIC 2019 challenge for skin lesion classification.

ABSTRACT. The application of a Long Short-Term Mem-ory (LSTM) neural networks have been explored in thisstudy to predict the active earth pressure coefficient(AEPC). The AEPC is a critical parameter in geotech-nical engineering for designing retaining structures. Tra-ditional methods rely on empirical formulas and manualcalculations, which may lack accuracy and efficiency. Byleveraging the capabilities of LSTM neural networks,which are well-suited for sequential data analysis andlong-term dependencies, this research aims to develop amore accurate and reliable prediction model. The perfor-mance of the LSTM model is evaluated based on metricssuch as Mean Squared Error (MSE) and determinationcoefficient (R2), and score function (S) to ensure itspredictive accuracy. The obtained experimental results(M SE = 0.0003, R2 = 0.97, and S = 0.0014), even fora limited number for training epochs, demonstrate thatthe LSTM neural networks can effectively outperformsand predict the active earth pressure coefficient with highprecision compared to conventional methods widely usedin this field.

ABSTRACT. This article addresses a critical problem in the field of target interception using a new foldable quadrotor with rotating arms. The primary challenge is maximizing linear acceleration to enhance the quadrotor’s ability to intercept targets effectively. To address this challenge, we introduce a novel optimization framework. Our study employs two distinct optimization algorithms, namely the genetic algorithms and the whales optimization algorithm, to ascertain the maximum attainable linear acceleration for the foldable quadrotor. The results obtained are confirmed through the use of our innovative foldable quadrotor.

ABSTRACT. Technology develops day by day, which is sometimes surprising. Our vision is a new generation of automated, intelligent, independent, flexible, compliant, and interconnected systems that work seamlessly with humans in agricultural greenhouses. This work consists of studying, designing, and building an automatically and manually controlled robot, which operates in greenhouses as an almost complete substitute for humans. The project consists of three parts: • The weather station part: collects information on the outside climate and sends it to the master unit. Master unit: collects climate information from the agricultural greenhouse and automates the agricultural greenhouse, as well as storing the data in a database. • The robot part: The robot part: Three modes of movement, moving automatically with the possibility of avoiding obstacles, on the other hand following the line, in the third manually via the web page. • The Artificial Intelligence part: It detects and identifies vegetables and fruit.

ABSTRACT. Virtual reality (VR) technology is increasingly used in a variety of applications, including healthcare, where it has shown promise as a functional rehabilitation tool after stroke to provide an immersive and interactive environment for patients. The use of VR in rehabilitation allows for taskspecific training, provides feedback, and allows for the simulation of real-life activities. This approach can lead to improvements in motor function, cognitive skills, and overall quality of life for patients. In this paper, we present a VR-based system for hand rehabilitation after stroke. This system was designed and developed in close collaboration with clinicians specialized in functional rehabilitation. It consists of a serious game represented by a virtual environment simulating a real-life activity, in which the patient is encouraged to interact with the virtual objects that comprise the virtual environment while playing the game, performing various tasks necessary to achieve their rehabilitation goals. The system uses a Leap motion controller as a motion tracking device, which captures the patient’s movements and transmits them to the virtual environment. To increase patient motivation and engagement, the virtual environment provides immediate feedback on the patient’s movements and performance in real time. This system was tested on a group of patients to evaluate its acceptability and usability.

ABSTRACT. In this study, a finite time synergetic control will be used to demonstrate the dynamic control of a car-like mobile robot (CLMR). To get the steering velocity and linear velocity to converge to references, use the FSC control. In order to provide controlled system stability, Lyapunov synthesis is used..

ABSTRACT. This paper investigates the attitude control of a perturbated three degrees of freedom (3-DOF) helicopter. From the practical side, the 3-DOF helicopter is a laboratory testbed system that operates on the same principle as that of a real tandem-rotor helicopter. Inspired by the theory of meta-heuristic optimization a hybrid controller is proposed by skillfully combining a Gray Wolf Optimization (GWO) algorithm, a Radial Basis Function Neural Network (RBFNN), and a ProportionalIntegral-Derivative (PID) controller. Given the nonlinear coupled dynamics of the system, the designed control structure allows for maintaining tracking performance and dealing with the nonlinearity problem. Furthermore, the GWO algorithm can usefully provide optimal gains whereas the RBFNN has self-tuning mechanism to achieve accurate control. The efficacy of the proposed controller is demonstrated with experimental tests performed on the Quancer’s 3-DOF helicopter laboratory setup. A comparative study is made involving the proposed PID-GWORBFNN strategy and two other controllers. The obtained results show that the suggested controller yields performance improvement regarding transient response and robustness.

ABSTRACT. Sign language or non-verbal language, which is based on hand gestures, is considered as a crucial communication system that facilitates interaction between the hearing-impaired people and the hearing community. The objective of the present study is to identify static and dynamic hand gestures through implementing a sign recognition system using two descriptors namely Low Frequency Descriptor (LFD) and Improved Weber Binary Coding (IWBC). The classification system is based on SVM one-against-all approach with RBF Kernel. The recognition system used was tested on four datasets, comprising a dynamic one which is Sebastien Marcel’s Dynamic dataset and three static ones which are Jochen Triesch’s dataset, Arabic Sign language 2018 dataset (ArSL 2018) and Arabic Sign language dataset 2001 (Halawani). The results revealed that higher recognition rates were achieved using IWBC with 97.07% on Arabic Sign language 2018 dataset and 95% on Jochen Triesch’s dataset. Regarding LFD, a higher recognition rate of 100% was obtained on Sebastien Marcel’s Dynamic Dataset.

ABSTRACT. The mission of parking attendants includes several tasks. One can mention, the supervision of automobile access, the management of parking facility, and the promotion of parking areas safety. Therefore, an intelligent system which allows carrying out some tasks only by speaking will definitely help attendants to complete their mission properly. The solution we propose activates the main computer to listen and promptly recognizes commands spoken in Arabic by the parking attendant in real-time. It seamlessly performs various tasks, including (i) opening and closing the main barrier, (ii) controlling the lights in the parking area automatically, and (iii) triggering and stopping theft alarms. The solution is based on two main phases: (i) querying the computer in real time through an impulsive waveform such as hands clapping, (ii) and recognizing voice commands. The first phase is accomplished using impulsive sound identification technique based on normalized version of power sequences. However, the second phase uses Google's application programming interface. The software was designed using Python language. We controlled external hardware such as parking barrier, light bulb, and alarm device using Arduino board. The execution time of the proposed software is nearly real-time and ranges from 5 to 6 seconds. The average subjective evaluation score conducted by 6 users is 4/5.