ABSTRACT. Recently, advanced beamforming antennas using engineered electromagnetic materials, such as FSS, EBG and metamaterials, are one of the attractive research topics, which have received much interest from many research and industrial groups worldwide. Using engineered electromagnetic materials, novel antenna systems with beamforming capability can be developed and used us an enabling technology to improve the performance of future wireless communication, radar and space systems at microwave and mm-wave bands. These approaches are based on using advanced artificial periodic electromagnetic structures, including electromagnetic band gap (EBG) structures, frequency selective surfaces (FSS), matasurfeces and metamaterials. The objective is to design and implement new compact, low profile, and low-cost antenna systems with high performances in terms of beamforming capability, high gain, and efficiency. With these features, they can be used to improve the performances of future wireless communication systems. For instance, they allow saving energy, increasing the coverage area, and reducing interference, which lead to good transmission quality. In this talk, I will first give a brief introduction on the wireless communication systems, presenting their potentials and challenges. Second, I will give an overview on periodic electromagnetic structures and their applications in advanced antenna designs. Third, I will present a few configurations of FSS-based reconfigurable antennas using various electromagnetic periodic structures. To show the beamforming feature of these antennas, some examples of simulated and experimental results will be presented and discussed. Finally, concluding remarks will be given.

ABSTRACT. Localisation and the estimation of the six degrees of freedom (6 DoF) object's position and orientation are essential tasks in numerous applications, such as augmented and virtual reality (AR and VR). It enables machines to gain a 3D understanding of their surroundings, empowering them to execute tasks like identifying objects, tracking their movement, interacting with them, and manipulating them. There is a growing trend in developing generalisable '6DoF' object pose estimators, which possess the ability to estimate the position and orientation of an unfamiliar object without the need for specific training or fine-tuning on that particular object during testing. By comparison, the previous 6 DoF object pose estimators focused primarily on specific categories of objects. This paper addresses the challenge of 6 DoF phantom head pose estimation from a single RGB image for augmented reality localisation. Using a local Magnetic Resonance Imaging (MRI) dataset, we reconstruct and print a 3D phantom head. This later is the target object for our AR localisation. Based on the comparison study including the accuracy and inference runtime scores, the Gen6D model achieves promising performance. Implementing 6 DoF pose estimation in deep learning stands to enhance and prove valuable in facilitating more effective interactions within the realm of augmented reality.

ABSTRACT. Hand recognition is crucial in various applications, such as gesture recognition, augmented and virtual reality. Hand gestures have the potential to facilitate human-computer interaction and enhance its convenience. The characteristics of gestures, such as their orientation and shape, differ among individuals, resulting in non-linearities in the problem. Recent studies have demonstrated the superior performance of Convolutional Neural Networks (CNNs) in accurately representing and classifying images, further emphasizing their effectiveness in this context. This paper proposes a novel deep learning model that is augmented using a transfer learning technique. Further, we trained our model using hand gesture recognition datasets of near-infrared images captured by the Leap Motion sensor. The proposed model is composed of ten classes that represent personalized gestures. The model with augmented data achieved an accuracy 99.12\% which is nearly 2\% higher than the model without augmentation (97.87\%). The results demonstrate the robustness and efficiency of our model compared with the state-of-the-art models.

ABSTRACT. This study aims to decrease the need for human intervention in CNN design by proposing a hybrid algorithm (GAPSO) that combines particle swarm optimization (PSO) and genetic algorithm (GA). Instead of fixed-length coding, we utilize a variable-length coding technique to represent population particles, and the particle update mechanism integrates GA principles. Our approach enables rapid convergence towards the optimal CNN architecture, resulting in high accuracy within a short period. We conducted experiments on five benchmark datasets for classification tasks and compared our algorithm with recent techniques from the literature. The results demonstrate the efficiency and practicality of our algorithm in autonomously discovering the optimal CNN architecture while minimizing human involvement.

ABSTRACT. Artificial intelligence (AI) has had a dramatic impact on a wide range of scientific disciplines, including healthcare. Deep learning (DL) algorithms have found application in the field of medical imaging in recent years, with promising results in disease detection and diagnosis. Cervical spine fractures are among the most common injuries among these illnesses, demanding quick identification for appropriate treatment. Recent advances in AI and DL open up the possibility of automated recognition and categorization of cervical spine fractures via medical imaging. The purpose of this study is to look into the capabilities of deep learning (DL) algorithms, specifically the convolutional neural network (CNN) architecture, in the automated detection and classification of cervical spine fractures utilizing medical imaging datasets. Our findings have the potential to greatly improve early detection and treatment of cervical spine fractures, resulting in better patient outcomes

ABSTRACT. In recent decades, medical imaging has emerged as a vital field in medicine, playing a crucial role in diagnosis. Computer Assisted Diagnosis (CAD) systems have become instrumental in this arena, employing sophisticated algorithms to extract crucial information from medical images. This study presents an innovative brain cancer detection system utilizing statistical classification methods. The approach involves three key stages: firstly, the identification of regions of interest through Gradient Vector Flow (GVF) Snake models and mathematical morphology techniques; secondly, the characterization of these regions using morphological and textural parameters; and finally, employing this characterization as inputs for a Bayesian network to classify malignant and benign cancer cases. Experimental validation of the proposed approach yielded impressive results, including a 100% sensitivity rate and a classification accuracy exceeding 98% for tumor segmentation. These findings underscore the high efficacy of the proposed CAD system, showcasing its potential in enhancing cancer diagnosis and patient care.

ABSTRACT. age estimation is a challenging problem in the field of facial image recognition and in biometrics features. Tradition- ally, convolutional neural networks (CNNs) have been the go-to approach for image analysis tasks in computer vision. However, the transformer architecture, originally introduced for NLP tasks, has demonstrated exceptional performance in capturing long-range dependencies and modeling complex relationships between elements in a sequence. This success inspired researchers to explore its application in computer vision, leading to the emergence of Vision Transformers.Therefore, this study aims to investigate the effectiveness of utilizing the standard Vision Transformer (ViT) model for predicting age classification from facial images using the AdienceDB dataset. Pretrained and fine- tuned ViT models (B/16) on ImageNet we are adopted for the Classicfication task . The proposed approach achieved promising preliminary results compared to the state of the art.

ABSTRACT. Glaucoma leads to permanent vision disability by damaging the optic nerve, which transmits visual images to the brain. The fact that glaucoma doesn't exhibit any symptoms as it progresses and can't be halted in later stages makes early diagnosis critical. Although various deep learning models have been applied to detect glaucoma from digital fundus images, the scarcity of labeled data has limited their generalization performance, along with their high computational complexity and specialized hardware requirements. In this study, progressive transfer learning with preprocessing techniques is proposed for the early detection of glaucoma in fundus images. The performance of this approach is compared against transfer learning and convolutional (deep learning) neural networks using three benchmark datasets: a cataract database, a glaucoma database, and the ORIGA database. The experimental results demonstrate that reusing pre-trained models from ImageNet and applying them to a database containing the same disease leads to improved performance compared to using databases with different diseases in progressive transfer learning. Additionally, applying preprocessing techniques to the databases further enhances the results.

ABSTRACT. The pivotal role of satellite imagery in our understanding of the world has led to a significant achievement: the capability to capture visuals from space using satellites. However, current optical sensors face challenges in delivering high-resolution images that encompass spatial, spectral, temporal, and radiometric dimensions. In response, the technique of Pansharpening is explored, merging high-spectral-resolution multispectral (MS) images with high-spatial-resolution panchromatic (PAN) images to create an improved spatial and spectral (HRMS) images. Generally, such methods utilize convolutional neural networks trained to implicitly map the MS and PAN inputs, producing a fused product with the highest fidelity. However, despite the remarkable fusion performance compared to traditional approaches, to the best of our knowledge, they suffer from two major issue that make them, to some extent, unsuitable for some remote sensing applications. First, deep learning-based methods generally rely upon a very deep and complex architecture, resulting in a high cost of parameters and computation resources. Second, They are more likely to be tailored to a single specific satellite system, lacking from generalization from unseen satellites. Therefore, in this paper, we introduce a novel global two-stream light weight method for Pansharpening, called TL-PNN, capable of combining data from diverse satellite types across the world. The experimental results show that the approach yields very competitive results on different datasets using the same model, while limiting the complexity of the model.

ABSTRACT. In an era of technological progress and pervasive digitalization across various sectors, the necessity for information security has risen to the forefront. In this digital age, the copyright verification stands out as a critical method for ensuring information security. In this paper, a promising solution for copyright verification is proposed, integrating the trustworthiness of biometric technologies with the power of chaos. In this study, we introduce a novel approach where features are first extracted, encoded, and encrypted before being seamlessly integrated into the cover image using the SM2LSB-based watermarking technique. A notable advantage of our method is the compact size of the encoded feature vector, which effectively reduces distortions in the cover image. Notably, our system optimizes the parameters of the chaos system to identify the most optimal embedding locations, further enhancing the method’s efficacy. The results obtained using the PolyU MSP database, comprising 300 persons, clearly demonstrate that our scheme outperforms several existing methods with a higher identification rate and substantially reduced distortion ratio.

ABSTRACT. Video compression is an essential operation in real time applications for Ultra High Definition (UHD). Several techniques for video compression exist nowadays, but the Versatile Video Coding (VVC) standard is the latest one. VVC adopt many new and developed techniques such LMCS filter, motion estimation, intra prediction and Context-based Adaptive Binary Arithmetic Coding plus (CABAC+). CABAC+ is considered as sole entropy coding adopted in VVC video codec but it undergone modifications and improvement compared to that used in High Efficiency Video Coding (HEVC). CABAC is operate with three main modes: Bypass (BM), Terminate (TM) and Regular (RM). Additionally, the heavy calculation of RM results a long critical path delay and consume more resources. In this work, we propose software and hardware implementation of RM CABAC decoder in the latest VVC standard. First, the algorithm is implemented on MATLAB software for manage and controlling the decoding process. Second, the Regular mode CABAC algorithm is implemented, synthesized and emulated. Finally, comparison results is carried out between that obtained with ModelSim and that obtained with Matlab in order to ensure the accuracy of our proposed design.

ABSTRACT. Brain tumor diagnosis and differentiation between various types of brain tumors are complicated and difficult tasks. To address these issues, automatic methods are desirable . This study proposes a method for classifying brain tumor images into three types of the most common tumor (glioma, meningioma, and pituitary tumor) in T1-weighted contrast enhanced MRI (CE-MRI) images. There are two scenarios proposed. First, the images are segmented using the U-net Convolutional Neural Network to crop the tumor region of interest (ROI). Second, two pre-trained Dense Convolutional Network (DensNet121) are used to extract features from the original images and cropped images. Then, these features were concatenated and passed to softmax classifier in order to classify the brain tumor. The proposed method based on features concatenation at output of pre-trained DensNet121 models produced 98.86 % accuracy on testing samples and achieved the highest performance in the detection of brain tumor as compared to existing state-of-the-art methods.

ABSTRACT. In this paper, we present a method to solve cells overlap and close boundaries problem during blood cells images segmentation with deep learning techniques. The CNN encoder- decoder based models have been developed recently to apply various segmentation tasks, however, loss of essential aspects in the data is a common issue during the model training, especially when it comes to low level features. Our work aims to fix this issue by dedicating a new class for cell boundaries and then segmenting them using U-Net. This method is helpful in complex imaging environment where the objects have close boundaries or overlap one over the other which makes it competitive for the model to recognize the corresponding segmentation contour.

ABSTRACT. The latest multi-view video coding standard, MV-HEVC, efficiently represents multi-view video sequences, reducing the required network bitrate, and incorporates error resilience tools to maintain video quality even under adverse network conditions. This paper investigates the impact of random error propagation in the MV-HEVC Network Abstraction Layer (NAL) and presents a comprehensive analysis of MV-HEVC bitstream sensitivity under error-prone networks. The study identifies the most sensitive data within the MV-HEVC bitstream and proposes an intra-refresh transcoding scheme to enhance the robustness of the MV-HEVC bitstream transmission. Experimental results ensured better video quality under adverse network conditions with an improvement of 22 dB in the perceptual quality of received multi-view video with minimal performance overhead.

ABSTRACT. The extraction of roads from satellite images is of particular importance in the field of remote sensing, as it is an interesting area of research for many different fields, including urban planning, road map updating and autonomous driving. This task is particularly complex because roads, due to their small size, winding configuration and ability to be obscured by the shadows of buildings, pose a sensing challenge for deep convolution neural networks (DCNNs). To meet this challenge, our study based on a semantic segmentation neural network explores and compares the capacity of LinkNet algorithm and their combination. The proposed approach, named DLinkNet34 combines the advantages of residual learning, dilated convolution that will enlarge the receptive field of features without reducing the resolution of feature maps, and a pretrained ResUnet34 encoder. The advantages of these two models are twofold. On the one hand, the residual units integrated into its architecture facilitate the formation of deep networks, thus improving their performance. On the other hand, rich hop connections within the network enable better information propagation, while reducing the number of parameters required. To evaluate the performance of D-LinkNet34, we conducted experiments using satellite images, comparing the results with other models such as U-Net and LinkNet34. The results clearly demonstrated the superiority of D-LinkNet34, achieving low losses and better visual results, particularly in complex environments where road geometry, occlusions and lighting conditions vary. The obtained results confirms the relevance of our choice in solving classification tasks for road detection in large-scale satellite images.

ABSTRACT. Free-Space Optical (FSO) communication is a promising solution for high-speed data transmission, yet its performance can be significantly affected by atmospheric conditions. This study investigates the impact of Return to Zero-Differential Quadrature Phase Shift Keying (RZ-DQPSK) modulation generation methods on FSO link performance under turbulent environments. Two distinct RZ-QDPSK transmitter schemes, one employing a serial configuration and the other utilizing a parallel configuration, are rigorously analyzed. Through the simulations under OptiSystem 14, the performance of these transmitters has been assessed, providing insights into their efficacy and suitability in challenging FSO scenarios. The study demonstrates the superiority of the parallel configuration in maintaining lower Bit Error Rates (BER) ensuring resilient communication and extending the operational range of the system. The findings underscore the significance of strategic modulation scheme selection for optimizing FSO system performance in challenging environmental conditions.

ABSTRACT. Device-to-Device (D2D) communications is one of the promising technologies for future fifth-generation cellular networks (5G). The first step to enable D2D communication is to discover neighbor devices and choose which will establish a D2D communication link. The limited battery life, discovery resources and the growing number of devices bind the current discovery protocols, so we need to find a new discovery protocol that goes with the next 5G cellular network. In this paper, we propose a new discovery protocol “MEC Discovery” based on Mobile Edge Computing (MEC) for D2D communication in 5G. The MEC server will discover neighbor devices for a User Equipment (UE) when requested and keep track of discovered devices in case other UEs ask for the same information with D2D mode in the near future. Our simulation results indicate a huge dropping in power consumption by the UE and reduced interferences caused by many devices doing discovery simultaneously in the same area.

ABSTRACT. The Internet of Things (IoT) has emerged as a transformative technology, revolutionizing critical domains such as healthcare, transportation, and energy management by enabling seamless connectivity, real-time data insights, and enhanced automation. Therefore, considerations for both intrinsic and extrinsic constraints are essential in the design of application protocols for IoT networks. IoT nodes often have limited processing power, memory capacity, and battery life. Furthermore, these nodes are interconnected wirelessly within a topology that undergoes dynamic and random changes. This in-depth study aims to analyze the performance of MQTT-SN (Message Queuing Telemetry Protocol for Sensor Networks) and CoAP (Constrained Application Protocol), which are widely regarded as prominent options for the IoT application layer. The analysis focuses on evaluating the control overhead and total energy consumption associated with these protocols under various scenarios of topologies and densities of IoT networks. The simulation results reveal a significant difference in the generation of control packets at the network level between CoAP and MQTT-SN. Specifically, CoAP exhibits a higher control overhead, which varies based on the network density, while maintaining equivalent energy consumption for both protocols. Moreover, CoAP demonstrates superior efficiency in transmitting application data messages compared to MQTT-SN under similar network constraints.

ABSTRACT. This article discusses the use of High Altitude Platform Stations (HAPS) in Intelligent Transport Systems (ITS) communication, with the objective of enhancing sustainability, efficiency, and safety in transportation applications. In the first section, the properties and capabilities of tethered HAPS are explored. In order to underline HAPS' benefits, such as its low latency and excellent reliability, a comparison is presented between HAPS and satellites. The study then analyses the effectiveness of HAPS by taking into account variables including surface coverage, attenuation of atmospheric gases, and data rate. A case study of the highway of the Hauts Plateaux in Algeria is offered to demonstrate the actual application of HAPS in ITS as a solution.

ABSTRACT. In a new work in this article, we studied a Trapezoidal antenna structure with MEMS switches, we presented the reflection coefficient and their radiation pattern to show the frequency responses, in the case without switches and in the case of the opening and closing switches (ON-OFF). We try through this study, to highlight all antenna directions without changing its location or changing its direction, to obtain the frequency and the radiation pattern that we want, we open or close one or more switches. At the beginning we presented the results before placing the switches we see that the antenna generates three resonance frequencies at 3.92 GHz, 7.84 GHz and 10.88 GHz with bandwidths of 240 MHz, then the results after installation of the switches on the antenna, the simulation results obtained by HFSS.

ABSTRACT. In this work, we are interested in designing an antenna array using SIW technology, with the basic element being a rectangular patch antenna. The main goal is to improve the radiation pattern. This technology allows us to produce this type of antenna at a lower cost and easily integrate it into telecommunications systems. The antenna was designed in the HFSS simulator on a FR-4 Epoxy substrate with a dielectric constant of εr=4.4 and a thickness of 1.6mm. Three structures have been proposed: the array antenna (AR2), the array antenna (AR4), and the SIW array antenna (AR4 SIW). We proposed inserting metal vias between the AR4 antenna's four elements to direct the antenna's radiation towards the main lobe. The antenna resonates at two different frequencies: 6.1GHz with a matching coefficient S11 = 17.77 dB and 6.85GHz with a matching coefficient S11 = 16.63dB.

ABSTRACT. Neonatal seizures pose a significant risk to infants and can have severe consequences if not detected and treated promptly. Electroencephalography (EEG) recordings are commonly used for monitoring and analyzing brain activity in newborns suffering from seizures. However, visual inspection of EEG signals can be error-prone and time-consuming, which can delay intervention. Deep learning algorithms, such as CNNs, have shown promise in classifying EEG signals for neonatal seizure detection. However, the multidimensionality of EEG recordings presents a challenge for using time-frequency representations (TFRs) as input for deep learning models. To address this challenge, recent research has explored alternative approaches, such as using a subset of the EEG channels that convey essential information about the brain’s activity. In this context, this paper proposes a novel approach for neonatal seizure detection using a minimized number of EEG channels and a TFD based on kernels with compact support. The proposed method employs a VGG19-BiLSTM mixed model to extract spatial and temporal features from the input data. The effectiveness of the proposed approach is evaluated on a publicly available Helsinki dataset, and the results demonstrate that this approach has the potential to significantly enhance the precision and practicality of neonatal seizure detection, potentially leading to more effective diagnosis and treatment outcomes.

ABSTRACT. Biopotentials are electrical signals generated by living organisms; they provide valuable information about the physiological processes occurring within the body. In several cases, such as electroencephalography, the level of the output biopotential signals is in the order of microvolts, and the measurements are easily corrupted by external interferences. In this paper, the design of a low-noise EEG amplifier is presented. The developed biopotential amplifier is composed of a low-noise instrumentation amplifier with a Driven Right Leg (DRL) circuit, a bandpass filter, and a programmable gain amplifier. Simulation results have shown an exhibit of a low input-referred noise of 0.174µVrms integrated from 0.1 to 100 Hz. After analog processing and filtering, the signal is sampled and digitized by a commercial data acquisition card interfaced to a PC. Additionally, the development of a wireless EEG acquisition system powered by a thermoelectric energy scavenger is proposed.

ABSTRACT. Diabetes is a chronic disease that represents a major challenge for global health. However, current technological advances allow us to monitor blood glucose levels in real time throughout the day, enabling better monitoring and faster intervention when needed. In this work, we aim to improve current methods of classifying patients at risk of diabetes using glycated hemoglobin test (HbA1c). The eleven features extracted from synthetic continuous glucose monitoring (CGM) data were used with two classifiers, namely decision tree (DT) and Random Forest (RF), to classify patients into four categories: healthy, moderate pre-diabetic, severe pre-diabetic and diabetic. The results obtained are promising, with an overall accuracy of around 92%.

ABSTRACT. The achievement of closed-loop glycemic control, and hence the automation of insulin administration, is highly required in the implementation of an Artificial Pancreas (AP) system. Various control theory approaches have been employed in many attempts to design an appropriate control algorithm. In our approach, we adopted a Constrained Linear Model Predictive Controller (CLMPC). Our simulations have shown that the CLMPC algorithm performs exceptionally well in maintaining blood glucose levels for patients across different scenarios. It effectively manages measurement and process noise, which is crucial for the practical implementation of the system. Moreover, the incorporation of feedforward capability in the controller has shown significant benefits, enabling proactive planning for future meals, and optimizing blood glucose levels. Our CLMPC is more effective and very simple than the ones presented in the literature, making it suitable for smartphone applications that support diabetes self-management. Therefore, our suggested intelligent control algorithm can be used as a core component of any Internet of Medical Things (IoMT) solution in addition to the present Glucose level control case.   

ABSTRACT. Gliomas are the most common type of primary brain tumors. The automated segmentation of these tumors from 3D magnetic resonance images (MRIs) plays a pivotal role in diagnosing, monitoring, and treatment planning for this disease condition. In this paper, we present a novel 2D U-Net architecture designed to automatically detect and segment three different tumor subregions, including edema, enhancing tumor and necrosis, in preoperative multimodal 3D brain MRI scans. The architecture of our convolutional neural network (CNN) model consists of three encoders, facilitating feature extraction from four different modalities of brain MRI images at multiple scales, as well as a decoder to efficiently upscale learned features while preserving spatial information. The proposed CNN model was tested and evaluated on 5880 multimodal brain MRIs of both, Brats'2023 training and validation datasets, which represent 1470 distinct adult patients with high and low-grade gliomas. It achieved an average Dice score of 0.92 for whole tumor (WT), 0.92 for tumor core (TC), and 0.90 for enhancing tumor (ET), when evaluated on the BraTS'2023 training dataset. On the validation dataset, the model's mean Dice scores for WT, TC, and ET were 0.89, 0.88, and 0.73, respectively.

ABSTRACT. Heart disease remains a significant global health concern, demanding swift and precise diagnosis. Traditional methods like stethoscope-based auscultation face difficulties in distinguishing complex heart sounds. To tackle this issue, we present an innovative method that combines signal processing and deep learning techniques to analyze Phonocardiogram (PCG) signals. Our approach involves segmenting PCG signals, extracting envelopes using mathematical morphological operations, and transforming them into image representations. A Convolutional Neural Network (CNN) is employed to classify these images, achieving impressive results with 99.00% training accuracy and 98.00% validation accuracy, along with minimal respective losses of 0.01% and 0.13%. During the final testing phase, our model demonstrated an accuracy of 98.00% and a loss value of 0.17%, reinforcing the robustness of our methodology.

ABSTRACT. In recent years, there has been a rising interest in integrating virtual reality (VR) technology into various medical services, specifically as a tool to support motor rehabilitation exercises for the upper extremities in stroke patients. VR has gained attention due to its potential to provide an interactive, engaging, and motivating environment for patients undergoing rehabilitation. This technology offers unique opportunities to enhance traditional rehabilitation approaches, providing a novel and immersive way to deliver rehabilitation exercises for stroke patients. The purpose of this paper is to introduce a new virtual reality system that aims to assist hemiplegic patients in their functional rehabilitation following a stroke. The system provides a partially immersive experience where users can interact with a virtual environment by tracking their hand movements using a leap motion controller motion sensor. An avatar of the user's hand is generated within the virtual environment, mimicking the movements of their real hand. The system is designed to deliver repetitive and goal-oriented tasks in the form of a serious game to facilitate the rehabilitation process. An initial evaluation was conducted to assess its acceptability among patients. The results revealed that patients responded positively to the game, expressing that it could be a useful tool in their functional recovery

ABSTRACT. Applications

ABSTRACT. The effectiveness of photovoltaic energy systems is contingent upon climatic variables, particularly sun irradiance and temperature. These considerations have prompted significant advancements in the development of novel Maximum Power Point Tracking (MPPT) techniques. The presence of partial shadowing condition gives rise to significant challenges, including a reduction in the output power generation and the emergence of several peaks in the PV characteristic; Also, it effects on the power quality injected to the grid. Therefore, the optimization of the global maximum power point (GMPP) under partial shading conditions is a challenging issue. In order to address this issue. This work provides a comparison of MPPT procedures utilizing the Grey Wolf optimization (GWO) method and the Cuckoo Search (CS) method. Furthermore, it explores the impact of these MPPT algorithms on the total harmonic distortion (THD) when implemented under identical system characteristics and climatic conditions. Using MATLAB simulation, the results demonstrate that before injecting power from PV, the MPC achieves a good THD Valeu of 1.87%, and that after injecting power, GWO outperforms CS in response time and efficiency, and has a lower THD value of 4.68%.

ABSTRACT. Digital holography (DH) is an imaging technique that allows the recording and the reconstruction of the complex optical field (complete information) in both amplitude and phase. Its main advantage is the possibility to modify the recorded wavefront to get free-aberration version. This technique represents the best contender for metrological applications especially in mechanical fluids where it gives an additional dimension for the determination of velocity field of different kind of fluid flow. In order to extend the fluid field of investigation , two synchronized cameras CCD and CMOS with different characteristic (pitch and overall pixel size) are used for recording two views holograms; this task is implemented under Labview interface which also permits to record successive several hologram with a predefined time interval. The holograms are obtained with the help of holographic setup in off-axis configuration of two views (direct and orthogonal). In order to obtain the total field of view, the two reconstructed holograms must have the same output resolution

ABSTRACT. Deep learning algorithms offer distinct advantages over other machine learning methods, enabling the exploitation of advanced techniques to analyze brain MRI scans. In a recent research study, a combination of ResNet-50 and DenseNet-201 convolutional neural network models was utilized to extract crucial features from brain MRI images, which were then fed into a compact classification model. The proposed model demonstrated remarkable results in accurately classifying the stages of Alzheimer's disease. It outperformed all other approaches using the same dataset and achieved an outstanding accuracy of 99.9 \% in categorizing the four cases, affirming the effectiveness and promising potential of this proposed model in the precise diagnosis and classification of Alzheimer's disease.

ABSTRACT. Visual cryptography, introduced in 1994 by Moni Naor and Adi Shamir, is a technique to encrypt images in a manner where the decrypted image becomes visible to the human eye. This involves dividing an image into shares, and encrypting each using distinct random masks. Overlaying these shares reveals the original image. The Vernam Cipher, using the XOR operation, encrypts by combining the secret image and a random mask, yielding the encrypted image. Decryption involves overlaying the encrypted image with the original mask, utilizing XOR to unveil the original image. Our study applies this method to binary, grayscale, and color image representations, thoroughly evaluating their respective strengths with the goal of achieving genuine visual cryptography. This exploits the human visual system's innate ability to directly decrypt encrypted messages.

ABSTRACT. In the last decade, the increasing popularity of image sharing applications over the web has led to a huge rising in the size of the personal image collections. While conventional content-based image retrieval systems suffer from the well-known sematic gap between low-level image features and high-level semantics, using textual information associated with images could be a good alternative. Therefore, in order to facilitate the navigation through these collections, and extracting meaningful information from them rapidly and accurately, semantic clustering of images based on textual information could help performing such an important task. In this study, we present a comparative study of several semantic similarity metrics for image datasets clustering. In particular, we evaluate the performance of eight measures namely Path, Resnik, Wu-Palmer, Lin, Jiang-Conrath, Leacock-Chodorow, Li, and Wpath. We conduct our experiments on three public datasets. The experimental results revealed that Resnik and Wpath Similarity measures whith accuracy (78% and 77.67% respectively) outperform the remaining metrics and yield more coherent and fast clustering solutions.

ABSTRACT. We analyze the energy requirements for assessing the heating load of building systems using the k-Nearest Neighbor (KNN) method for classification. Unsupervised transformations of the dataset are employed in learning algorithms. The system is implemented and simulated in Anaconda, with performance testing conducted on a real dataset comprising 8 features and 768 instances to classify heating load magnitudes. Two distinct data representation classes are generated separately from the original dataset (excluding the target) based on captured heating load magnitudes. The unsupervised transformations of the dataset yield 4 and 8 classes, respectively. To evaluate performance, various test and training sizes are employed, comparing measurement outcomes and predicting class label descriptions for the two dataset representations as the nearest neighbor parameter (k) varies from 1 to 19. Simulation results indicate that accuracy is influenced by factors such as test size, the k parameter, and dataset class count. The training and test accuracy rates reach maximum values of 95% and 85%, respectively. Keeping k constant, the test size has a negligible impact on accuracy. The classes exhibit two critical regions: 1) Region I, where accuracy fluctuates with increasing k in the interval [1-5], and 2) Region II, where accuracy declines with increasing k in the interval [5-10]. In Region III (k[10-19]), accuracy remains relatively constant.

ABSTRACT. Machine Learning (ML) play a significantrole in the design and optimization of new generation optical systems that are intelligent, adaptive, and efficient, enabling a wide range of new applications in the field of optical telecommunications. In this work, ML algorithm is proposed to classify the two common intensity modulation formats, such as Non-Return-to-Zero (NRZ) and Return-to-Zero (RZ) used in Wavelength Division Multiplexing (WDM) system. ML-based modulation classification involves to classify a signal based on its features that can include statistical features, such as the mean and variance of the signal, spectral features such as the power spectral density(PSD), autocorrelation function, constellation diagram, Bit Error Rate (BER), Q-Factor and Signal-to-Noise Ratio (SNR). The proposed system was trained on a dataset of signals with the known modulation schemes, NRZ and RZ which is pre-processed to extract features, notably BER, Q- Factor and SNR. Once the dataset is prepared, an ML algorithm was trained and will output a predicted modulation format based on the learned relationship between the features and the modulation format to classify new signals. The obtained results based on Pearson correlation matrix and VIF (Variance Inflation Factor) factor indicate that from a dataset of Q-Factor as a function of the distance link, the ML-based modulation classification help in optimizing the design of WDM systems by allowing for more precise control and tuning of the used modulation format.

ABSTRACT. Abstract—Despite the increasing speed of development  in 3D medical image segmentation based on deep learning  techniques, it is important to consider the requirements for performing this process due to the large number of information contained in the data. So far, researchers have proposed several methods to avoid excessive demand placed on computational resources. The objective of this work is to address this issue by leveraging 2D based Convolutional Neural Network (CNN) architectures to segment 3D medical images in order avoid resource overload. To achieve this, it has been divided the work into three phases. The first one consists of splitting each 3D image alongside depth axis. Next, it has been trained the extracted slices through U-Net; one of the most common CNN architectures in the biomedical image segmentation field. To end up with reconstructing the predicted slices back to volumetric images. In the MSD-Spleen dataset, our best F1-Score, IoU, and Accuracy on the validation set were 0.840 and 0.89 respectively. Moreover, this method has shown efficiency to train volumetric images using limited resources.

ABSTRACT. Anti-personnel and anti-tank landmines create a lot of human and material damage in the Sahel countries affected by terrorism. Explosive mine detection methods are based on tools handled by human operators and target industrial metal mines. These methods are risky and limited because the types of mines most commonly used in the Sahelian context are mainly homemade and are encased in various local materials such as metal, plastic, glass, ceramic or wood. This paper presents a solution for remote sensing of artisanal mines buried in the ground using infrared thermography. A DJI Phantom 4 quadcopter equipped with a FLIR thermal camera and a GNSS sensor performs an automatic low-level flyover of the potentially mined road. Thermal images of the road are collected with an overlap rate of 80% and referenced with the GNSS sensor. Photogrametry algorithms are used to process the thermal images to detect and locate anomalies related to the presence of buried mines. Despite the limitations due to environmental influences, the model showed a detection rate of 75% during flights at an altitude of 6m and a speed of 3m/s. The experimental results show a good correlation between the thermal contrast of the mathematical model and the cooler areas containing a mine-related chemical substance.

ABSTRACT. Classification of surface crack images is an ever-evolving area of research that aims to develop machine learning models capable of identifying and categorizing cracks in materials. This study focuses on comparing the performance of different machine learning algorithms applied to crack image classification to find the most appropriate classification technique. To conduct this comparative study, several machine learning algorithms are selected, such as neural networks (NN), support vector machines (SVM), decision trees, random forests and k-nearest neighbors (KNN). ), Each of these algorithms is trained and evaluated on a dataset of previously annotated crack images.

ABSTRACT. Modern Standard Arabic (MSA) serves as the official language across all Arab countries, employed in administrative, educational, official broadcast, and press settings. However, in everyday informal communication, each Arab country adopts its own variant dialect, commonly known as colloquial dialect. This colloquial form has emerged as the dominant language on social media platforms due to the widespread usage of such platforms over the past decade. Consequently, there is a growing demand for language resources and natural language processing (NLP) systems that cater to these dialects. This paper details the development of an Algerian Arabic dialect dataset, with a focus on average colloquial dialect to ensure effective performance on standard Algerian dialect. Carefully curated, the dataset encompasses a diverse collection of high-quality audio extracted from social media platforms. Employing a fine-tuning approach, we leverage existing ASR systems, particularly the Whisper model by OpenAI for its noise robustness and adaptability to varying audio quality. The paper delves into the process of efficiently collecting speech data, while also outlining potential applications and future directions. Researchers in speech analysis and natural language processing will find this work invaluable for advancing their knowledge of Algerian Arabic and fostering accurate ASR system development tailored to the dialect.

ABSTRACT. This study presents the CIAF-PFNLMS algorithm, an abbreviation for Proportionate Fast Normalized Least Mean Square with Competent Individual Activation Factors. This algorithm provides a competent method for calculating individual activation factors. This guarantees that the gain remains proportional to its associated coefficient even when the coefficient is inactive, in accordance with the concept of proportionality. This proves very beneficial in scenarios involving sparse impulse responses, leading to significant performance improvements. Additionally, the algorithm smoothly integrates the Fast Normalized Least Mean Square (FNLMS) adaptation gain into its framework, making it effective in environments with dispersive impulse responses and capable of accurately tracking echo path variations. In our study, we performed simulations to validate the capabilities of the proposed algorithm. We evaluated its performance across a range of scenarios, including sparse and dispersive environments, non-stationary situations, and compared it with established algorithms

ABSTRACT. The risk of compromising data confidentiality and quality during transmission, especially in the case of images, has led to the use of various encryption algorithms to secure transmitted images while ensuring their original quality. Satellite images, in particular, often have large volumes and high redundancy. Considering that satellites operate within strict limits in terms of computing power and resources, their operation in harsh environments can introduce radiation-induced defects in onboard electronics, including encryption systems. The objective of our work is to analyze the quality of a specific type of satellite image, which is Interferometric synthetic aperture radar (InSAR) interferogram, encrypted using a Feistel network-based cryptosystem with Cipher FeedBack (CFB) mode. Several metrics are used for this analysis, based on objective methods using a complete reference image, which consists of two original interferograms. The quantitative measures employed include Structural Similarity Index (SSIM), SHA-256 hash function, Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and statistical analysis such as encrypted interferogram histogram analysis.

ABSTRACT. In this paper, we exploit the improved chaotic Colpitts configuration to generate chaotic signals in the frequency band 1-11 GHz using a new type of bipolar transistor BFP420 with a threshold frequency of 25 GHz. The numerical analysis of the mathematical model under MATLAB allows to plot the different chaotic temporal responses and state spaces. The circuit simulation of the oscillator is made under the ADS simulator with the aim of studying its frequency characteristics. In the second part of this paper, we propose a synchronization method based on nonlinear feedback control.

ABSTRACT. Early detection of breast cancer can lead to a significant increase in survival rates, making it one of the most common cancers among women worldwide. The exponential growth in the number of mammograms performed each year is making computer-aided diagnosis of breast cancer increasingly necessary. Various researches have been conducted either to automatically detect lesions in images using computer-aided detection systems or to provide a second opinion about the lesion detected through diagnostic systems assisted by computer. Medical imaging is a crucial technique that ensures a superior quality of clinical diagnosis. The interpretation of the medical image is still a problem. This led to the design of several mammographic image processing methods all based on the American College of Radiology (ACR) Breast Imaging Reporting and Data System (BI-RADS) classification to classify the masses into two categories namely benign and malignant. The purpose of this article is to develop an automated system that can help detect and classify different pathologies in mammographic images.

ABSTRACT. Blood transfusion is a medical procedure that involves transfusing blood or one of its components from one or more donors into a patient. Digital technology and machine learning have played a crucial role in the blood field and have provided real prospects for the production and distribution of blood products. In this study, we propose supervised machine learning techniques for the multi-label classification of blood products transfusion in patients with hematologic diseases. We used three multi-label approaches from the problem transformation category to create a decision support system for blood products: Label Power Set (LP), Binary Relevance (BR), and Classifier Chain (CC). Multi-label classification using the Problem Transformation approach is a flexible approach. In this study, we used data from different hospitals in haematology departments and blood transfusion centers to explore the application of contemporary supervised learning algorithms in blood product prediction. The experiment was performed by calculating the Hamming loss and Accuracy to facilitate the classification of blood products. As a result, the prediction model achieved an area under the ROC curve of 99.80%, a Hamming loss of 0.30, and an accuracy of 98.8%. The proposed model has been developed to provide accurate and fast results that can save patients’ lives.

ABSTRACT. In the field of pharmacology, standardized recognition systems for medical labels in Algeria are lacking due to the absence of an approved model by the Minister of Health. This study proposes an AI-based system that automatically identifies medical labels. Two approaches are introduced: the Tesseract-based approach using traditional detection and segmentation techniques, and the VisionAI-based approach with an AI custom-trained model. Experimental evaluations show the VisionAI-based approach achieves an accuracy of approximately 95.79% in label detection and content extraction. This research presents a promising solution for efficient medication management in Algeria.

ABSTRACT. Electromyography is an important diagnostic tools used for neuromuscular disorders’ classification. Amyotrophic Lateral Sclerosis (ALS) is one of the most severe and dangerous neurological disorders. In this paper, we propose to use time–frequency analysis of EMG signals and Convolutional Neural Networks (CNNs) for efficient classification of ALS and normal EMG signals. In the proposed approach, the Polynomial Cheriet-Belouchrani distribution (PCBD) is herein used to provide the TF signatures related to the analyzed EMG samples. The obtained results show that the designed classifier based on the PCBD and the CNN model outperforms the best known classifiers related to ALS disease identification, attaining an average test accuracy of 99%.

ABSTRACT. Agriculture is a sector that represents a real indicator of the country’s development, and it is a sector that affects national security. The palm tree is a significant, affordable, and nutrient-rich food source that gives millions of people access to wholesome food globally. It can get a lot of illnesses, much like other plants. DL has been effectively implemented in agricultureand has been employed in many other industries. This paper examines the latest advancements in the use of deep learning (DL) for the diagnosis of palm tree diseases. Particularly, convolutional Networks (CNN) are often utilised in agricultural research due to their strong image processing ability.

ABSTRACT. Universal Filtered MultiCarrier (UFMC) waveform has emerged as a promising choice for Multi-Carrier Modulation (MCM) within fifth-generation (5G) systems. Its appeal lies in its ability to offer a lower Bit Error Rate (BER) and reduced latency compared to alternative waveforms. A significant challenge common to all MCM systems is the high Peak-to-Average Power Ratio (PAPR). UFMC signals tend to exhibit substantial peaks, leading to distortions when they surpass the linear range of the high-power amplifier, consequently causing a notable deterioration in BER. Numerous past research studies have attempted to address this issue by reducing PAPR for UFMC systems. However, these solutions often proved challenging to implement, overly complex, or necessitated parallel information transmission to the receiver. In light of these challenges, this paper presents an innovative alternative approach that can be easily implemented without the need for secondary information transfer to mitigate PAPR. This method involves applying the L₂-by-3 technique belonging to the Discrete Sliding Norm Transform (DSNT) to the UFMC system. Simulation results indicate that the proposed approach surpasses most other methods in terms of PAPR reduction, achieving improvements of up to 60%.

ABSTRACT. Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social communication and repetitive behaviors. Diagnosis of ASD is currently based on behavioral assessments, which can be subjective and time-consuming. In this study, we aimed to develop a more objective and efficient method for ASD classification using functional Magnetic Resonance Imaging (fMRI) by extracting the most correlated features. We got fMRI data from Autism Brain Imaging Data Exchange (ABIDE I) that is intended to aid in developing computational models of autism and to facilitate the discovery of biomarkers for the disorder; we used this dataset with different machine and deep learning algorithms to classify individuals based on their brain activity patterns. Our results showed that by using fMRI, it is possible to accurately classify individuals with and without ASD with high sensitivity and specificity. This study provides a promising direction for using fMRI and machine learning in the diagnostic process of ASD.

ABSTRACT. This paper addresses the challenge of efficiently offloading heavy computing tasks from ground mobile devices to the satellite-based mist computing environment. With ground-based edge and cloud servers often being inaccessible, the exploitation of satellite mist computing becomes imperative. Existing offloading algorithms have shown limitations in adapting to the unique characteristics of heavy computing tasks. Thus, we propose a heavy computing task offloading algorithm that prioritizes satellite proximity. This approach not only reduces energy consumption during telecommunications but also ensures tasks are executed within the specified timing constraints, which are typically non-time-critical. Our proposed algorithm outperforms other offloading schemes in terms of satellites energy consumption, average end-to-end delay, and tasks success rates. Although it exhibits a higher average VM CPU usage, this increase does not pose critical challenges. This distance-based approach offers a promising solution to enhance energy efficiency in satellite-based mist computing, making it well-suited for heavy computing tasks demands.

ABSTRACT. Breast cancer continues to pose a significant public health challenge with increasing prevalence. Accurate early detection is crucial for effective treatment and improved patient outcomes. This paper introduces a novel approach for breast Micro-Calcification mammography classification, distinguishing between benign and malignant cases. The proposed methodology combines image pre-processing for denoising, deep transfer learning for feature extraction, and classifier using global pooling and a Multi-Layer Perceptron. Notably, two feature extractors, Xception and DenseNet201, are combined. The study demonstrates the potential of this approach with impressive results, achieving a maximum accuracy of 98.45%.

ABSTRACT. In this article, we highlight image segmentation in the field of medical diagnosis, with a special emphasis on the accurate identification of retinal vessels in fundus images. Achieving accurate segmentation is essential for effective screening and monitoring of eye diseases. To enhance segmentation performance, the study presents a deep learning approach that utilizes the U-Net architecture. This method involves data augmentation preprocessing, where green channel extraction is performed along with Contrast Limited Adaptive Histogram Equalization (CLAHE) enhancement on two important databases: DRIVE and CHASE-DB1, to train the U-Net model. The research used Google Colab as the development environment and Python as the programming language. Experimental results confirm the effectiveness of this approach, providing promising avenues for further improvement and innovation in retinal vessel segmentation using the U-Net architecture.

ABSTRACT. Collecting a dataset for a specific plant disease is a very complex, costly, and time-consuming process, involving the use of expensive equipment and monitoring the evolution of the disease on a plant-by-plant basis over several crop cycles. Overall, data collection will be simpler if we can form a classifier disease model for a single plant and then use it for other plants that may have the same disease and the same symptoms. The proposed work consists of a convolutional neural network (CNN) binary-based classification of black rot disease applied to apples and grapes. The employed methodology is based on training and testing six types of CNN models, namely efficientnet_v2, inception_resnet_v2, inceptionV3, mobilenet_v2, VGG-16, and our proposed CNN model for the purpose of black rot disease detection. The dataset contains a single type of plant leaf. Then, the model with high accuracy is applied to another plant showing the same disease leaf with common symptoms. We divided the experiment into 3 main approaches, providing respective accuracies of 100%, 97.43%, and 93.16%. The obtained results show that the CNN-based architectures allow the classification of apple and grape black rot leaf diseases with high accuracy. The proposed approach is promising and encouraging in the context of developing CNN models so that Data collected for specific plant diseases are automatically applied to all plants susceptible to infection by the same diseases. This approach will enable researchers involved in the early detection of plant diseases to speed up their work by optimizing computing resources during the model-learning phase and collecting a disease-specific database for a single plant as quickly as possible.

ABSTRACT. This academic work evaluates and compares the performance of various convolutional neural network (CNN) architectures in classifying thyroid nodules into two categories, malignant and benign, using ultrasound images. The dataset comprises 269 cases of benign lesions and 526 cases of malignant lesions. Given the limited dataset size, we employ a progressive learning approach with three established CNN models: VGG-16, ResNet-50, and EfficientNet. Initially pre-trained on ImageNet, these models undergo further fine-tuning using a radiographic image dataset related to a different medical condition but similar to our domain. Different levels and fine-tuning strategies are applied to these models. We employ a supervised softmax classifier for classifying lesions as malignant or benign, with the exception of the VGG-16 model. For the VGG-16 model, we evaluate two additional classifiers, Support Vector Machine (SVM) and Random Forest. The results obtained demonstrate the possibility of easily transitioning from the classification of one disease to another, even with a limited number of images, by leveraging the knowledge already acquired from another extensive database. This endeavor has important implications for the application of deep learning in medical image classification tasks.

ABSTRACT. The Alzheimer’s Disease (AD) is a progressive incurable neurodegenerative disease and one of the most challenging health problems that scientists are facing since decades. Despite the current diagnostic approach for AD involving clinical, neuropsychological, and neuroimaging assessments, this conventional diagnosis may be impractical and time-consuming given the significant burden of the disease. Furthermore, the identification of diagnostic factors from brain images is critical, considering the subtle changes in the brain and the variability in visual interpretation among clinicians. A computer-aided diagnosis (CAD) system based on Artificial Intelligence (AI) technologies is considered as a potential tool to mitigate this problem and to assist medical professionals in decision-making, providing a more reliable automatic detection. Our research goal is to find the maximum accuracy rate of AI-based diagnostic classification using neuroimaging data. In this paper, a hybrid machine learning (ML) model is proposed on a problem of recognition of demented subjects from healthy ones in accurate manner using the three anatomical sections of structural Magnetic Resonance Imaging (sMRI) dataset. The model has achieved 91.6% classification accuracy.

ABSTRACT. Object detection models based on region proposals, whether those using classic methods or deep neural networks, have provided cutting-edge results and achieved an interesting speed/accuracy ratio. The state-of-the-art methods using local feature detectors to generate region proposals incorporate descriptor information or combine the detected interest points with other methods to generate proposals. In this work, we aim to explore the efficiency of proposals generated only according to the position of the generated SURF keypoints. We propose an object detection approach namely SURF-Fast-RCNN, based on FAST-RCNN framework, that integrates SURF local detector. Instead of using the proposals produced by the selective search method, each 2-combination of interest points is considered as a candidate box which we call keypoint anchors. Based on the intersection over union (IoU) value of each keypoint anchor with ground truth boxes, we discard the excess anchors and assign labels to the retained ones denoting their class labels. The retained anchors are fed into the detector for training, followed by NMS operation to suppress the overlapping boxes. Our SURF-Fast- RCNN model is trained and tested on different region proposals on the MS-COCO dataset examples. With 1000 region proposals, the performances of our model reached an accuracy of about 82% and a mean average precision of 84.04%, which are encouraging results.

ABSTRACT. This article discusses the adaptive acquisition process analysis performed by a Global Navigation Satellite System (GNSS) receiver. Signal acquisition determines the presence or absence of a GNSS signal by comparing the tested signal to an adaptive threshold. The main objective of this work is to optimize the thresholding to obtain a constant false alarm rate (CFAR) in both the statistical order CFAR ( OS-CFAR)( Order statistic) and the CFAR cell averaging and modified likelihood ratio test detectors (CMLD-CFAR)( Censoring mean level detector) in the Rayleigh Fainting Canal. In GNSS acquisition, the driver and data channels may have different thresholds. Therefore, optimization focuses on two scale factors (T1 and T2) for each detector using the "AND" and "OR" fusion rules. Because of their performance in various problems, simulation results demonstrate the best way to improve detection in the GPS (Global positioning system) system

ABSTRACT. The recursive least squares (RLS) algorithm is widely used in various adaptive filtering applications, mostly due to its rapid convergence rate. The forgetting factor is a crucial parameter in this algorithm. For a fixed value for this parameter, there is a tradeoff between misalignment and tracking. Nevertheless, the use of a variable forgetting factor (VFF) approach gives a better compromise between the performance criteria of the RLS algorithm. Nonetheless, the algorithm's computational complexity presents a notable challenge, particularly in scenarios that involve lengthy adaptive filters, such as acoustic echo cancellation (AEC). This paper incorporates the concept of a variable forgetting factor into the fast normalized least mean squares (FNLMS) algorithm. The FNLMS algorithm demonstrates performances comparable to the RLS algorithm while maintaining a computational complexity similar to that of the normalized least mean squares (NLMS) algorithm. Based on simulation results, the proposed VFF-FNLMS algorithm exhibits better performances in terms of convergence speed and tracking ability when compared to both FNLMS and NLMS algorithms.

ABSTRACT. In synch with the artificial intelligence era and particularly in the biomedical field, biomedical signals like electrocardiographic (ECG), electromyographic (EMG), and Electroencephalogram (EEG) are being used in various applications, such as artificial hand and arterial pressure. However, identifying a patient's ailment is still a challenge. In this paper, we have utilized three empirical mode decomposition techniques to minimize the impact of additive noises on noninvasive biomedical signals. These methods are the classical empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), and the complete ensemble empirical mode decomposition with additive noise (CEEMDAN). Using the correlation coefficient, we conducted an extensive simulation and detailed comparative study between the noisy and reconstructed signals. The results show that the CEEMDAN method is the most effective in reducing noise compared to the other two methods.

ABSTRACT. The utilization of electroencephalography (EEG) in scientific investigations related to biometric authentication has spurred intricate experiments aiming to recognize and confirm individuals based on brain activity in specific contexts. Recently, EEG has proven to be a viable signal for biometric authentication, offering significant advantages such as resilience against fraudulent attempts and immunity to coercion or pressure. Leveraging EEG signals derived from brain activity holds promise for addressing contemporary security challenges in conventional knowledge-based user authentication. This paper delves into cutting-edge approaches within EEG-based authentication. The analysis encompasses 12 distinct experiments involving EEG-based user authentication, exploring various facets including pre-processing techniques and classification methods employed in EEG biometric authentication. The study also zooms in on artificial intelligence techniques integrated into the classification process and their corresponding outcomes. The scrutinized papers span from 2019 to 2023, encompassing diverse experimental protocols and algorithms employed for classifying EEG signals. This review engenders a discourse on advanced biometric authentication methods, offering a comprehensive overview of the most recent biometric authentication techniques rooted in machine learning and deep learning, utilizing EEG data. Anticipated outcomes include the provision of a roadmap for forthcoming research endeavors, emphasizing the development of efficient and effective EEG-based user authentication methods while upholding privacy standards.

ABSTRACT. In this paper, we propose a novel approach for Phonocardiogram (PCG) signal classification using a BiLSTM model with Mel Frequency Cepstral Coefficients (MFCC) features extracted from short PCG segments. The issue of handling PCG signals of varying lengths is addressed by segmenting the audio signal, allowing for feature extraction and organization with a fixed dimension that is compatible with the input layer of the BiLSTM model. Our approach achieves state-of-the-art performance while utilizing only two BiLSTM layers, making it an efficient and lightweight model for embedded applications. The combination of MFCC features, a handcrafted feature extraction method, with a BiLSTM architecture, addresses the issue of feature engineering for improving PCG signal classification performance. Our study is the first work in the literature to explore the potential benefits of using MFCC features with a BiLSTM model for PCG signal classification. The proposed approach has the potential to significantly impact the healthcare industry by improving the accuracy and efficiency of PCG signal classification, aiding in earlier diagnosis and treatment.

ABSTRACT. IoT devices are on the verge of permeating our lives more extensively than mobile phones and will possess access to the most sensitive personal data, including social security numbers and banking information. With the increasing number of devices, security challenges escalate, further complicated by the fact that security solutions often are not compatible with most embedded devices. These devices must ensure the security of stored data, guarantee secure communication, and defend against cyber-attacks. The objective of our paper is to explore how to fortify our IoT device, exemplified by a Raspberry Pi, utilizing a Firewall, IDS, and SSL/TLS. Initially, it will establish a connection with the Losant platform (IoT platform), and we will verify its seamless operation. Subsequently, we will progress with the Raspberry Pi, involving three stages: installing a Firewall, implementing an IDS, and confirming the default presence of the SSL/TLS protocol. Finally, we endeavored to scrutinize the traffic between them to demonstrate that no unauthorized attempt will be made to breach our Raspberry Pi.

ABSTRACT. Ship detection in SAR (Synthetic Aperture Radar) imagery plays a crucial role in maritime surveillance, enabling the monitoring of vessel activity, ensuring maritime safety, and supporting security operations. In this work, we propose a new CFAR (Constant False Alarm Rate) ship detector for a heterogeneous Weibull sea clutter, specifically designed for SAR imagery. This detector is an extension of the MAD (Median Absolute Deviation) detector to ship targets, a robust measure of the variability to achieve a high DR (Detection Rate) [1]. Extensive evaluations using real SAR images show that the proposed MAD-CFAR ship detector exhibits excellent detection performances in terms of DR and CPU (Central Processing Unit) time. The experimental results show the effectiveness of the proposed CFAR ship detector, highlighting its potential for real-time of large-scale maritime surveillance applications. This research work contributes to advancing ship detection capabilities in SAR imagery and holds great promise for enhancing maritime security and situational awareness.

ABSTRACT. The intermittent nature of renewable energy sources makes it extremely challenging to predict the quantity of energy produced. This is due to the continuous variation in weather conditions, including the amount of solar radiation at the production site, wind speed, temperature, rainfall, and other factors. In our work, we conducted an analytical study using a massive dateset collected by the Capricorn FLX 202004 weather station, which is affiliated with the Laboratory of Hydraulic and Hydrological Engineering and Water Resources Development at the University of Biskra, Algeria (LARHYSS). In this analytical study, we utilized deep learning method , which has proven its ability in analyzing big data. We performed this process to facilitate the fieldwork carried out by companies interested in investing in clean electricity production in the region. Through this statistical weather forecasting technique, we provided valuable insights for researchers in the field of designing emerging energy production networks to identify the best locations for constructing hybrid power stations that combine solar and wind energy.

ABSTRACT. Convolutional Neural Networks (CNNs) have gained significant popularity in image classification tasks, yet achieving their optimal design remains a challenge due to the vast array of possible layer configurations and associated hyperparameters. Selecting the best CNN model for a given task often demands considerable time investment in training numerous models. To address this issue, we propose an automated method for CNN architecture design, utilizing pre-trained models like the Backbone, and employing Gray Wolf Optimization. This approach automatically generates and evaluates candidate CNN architectures for classifying plant seedlings. Our objective is to distinguish between weed and crop seedlings. Additionally, we introduce a gray wolf representation to encode CNN architectures and their hyperparameters. Our method combines the strengths of transfer learning from pre-trained models to extract meaningful image features with the optimization capabilities of the Gray Wolf Optimization (GWO) algorithm. By leveraging these techniques, our method achieves exceptional accuracy, surpassing state-of-the-art methods with a validation accuracy of up to 97.83%. This innovative approach offers a transformative tool for enhancing the accuracy of CNN models, tailored specifically to the dataset at hand.

ABSTRACT. Early Recognition of plant diseases is required to provide appropriate treatments and protect agricultural production. Automated image detection techniques can provide early, fast, accurate, and inexpensive detection results. However, scientific research on date palm diseases is even rare due to the limited number of databases, including the Bayoud disease. In this paper, a new database, Biskra Algeria Palm Fronds Database (BAPFD), containing 400 images of date palm fronds is proposed, it includes images of both healthy and Bayoud disease-affected fronds. The Deep Learning (DL) technique is then used to classify our database and recognize Bayoud disease in date palm fronds. The performance of our approach is measured in terms of accuracy, loss, precision, and recall. The mentioned experimental results show that the proposed approach based on DL achieved high performance in recognizing Bayoud disease in date palm fronds, up to 100% for precision and recall and in order of 4 × 10−4 for loss. The results show also that the model is capable to accurately classify infected and not infected fronds images even with a small percentage of bayoud. This demonstrates that our database contains significant information concerning bayoud diseases, which enables DL to design an accurate model that aids in correctly classifying palm fronds and identifying Bayoud disease. We further acknowledge the effectiveness of the DL approach in the recognition of healthy and not healthy fronds.

ABSTRACT. Fatigue is a significant safety issue that has not been fully addressed, primarily due to its nature. As a general rule, it is tough to measure or observe it. Most current systems rely on external factors to inform the user of a problem and rescue the user after an accident. These methods cannot detect facial expressions, head yawning, or eye blink frequencies. According to the National Safety Council (NSC), drowsy driving accounts for about 100,000 crashes, 71,000 injuries, and 1,550 fatalities a year. In addition, a study from the AAA Foundation for Traffic Safety found that drowsiness contributed to up to 9.5% of all crashes. Drowsiness detection is a safety technology that can prevent accidents caused by drivers who fall asleep while driving and reduce damage to those who work on computers for extended periods. For all these reasons, drowsiness detection technologies have attempted to prevent such incidents by predicting if a person is falling asleep based on various inputs. Our research focuses on a Computer Vision technique for drowsiness detection. The work that is being proposed will be carried out in multiple stages. The first is face detection using MTCNN. In the next stage, a Swin Transformer is used to extract facial features for classifying the driver’s drowsiness level. The effectiveness of our approach is evaluated on the NTHU Drowsy Driver Detection (NTHU-DDD) video dataset. The results demonstrate that the proposed method outperforms state-of-the-art methods with 76.54% accuracy.

ABSTRACT. Accurate disease classification utilizing endoscopic images indeed poses a significant challenge within the field of gastroenterology. This research introduces a methodology for assisting medical diagnostic procedures and detecting gastrointestinal(GI) tract diseases by categorizing features extracted from endoscopic images using Vision Transformer (ViT) models. We propose three ViT-inspired models for classifying GI tract diseases using colon images acquired through wireless capsule endoscopy (WCE). The highest achieved accuracy among our models is 97.83%. We conducted a comparative analysis with three pre-trained CNN (Convolutional Neural Network) models namely, Xception, DenseNet121, and MobileNet, alongside recent research papers to validate our findings.

ABSTRACT. Person re-identification (PRe-ID) is a challenging task in computer vision field, that has been an fertile research area in the last few years. It aims to identify persons across different non-overlapping camera views.In this paper, We propose a novel PRe-ID systemt hat combines tensor feature representation and multilinear subspace learning. Our method exploits the power of pretrained Convolutional Neural Networks (CNNs) as a strong deep feature extractor, along with two complementary descriptors, Local Maximal Occurrence (LOMO) and Gaussian Of Gaussian (GOG). Then, Tensor-based CrossView Quadratic Discriminant Analysis (TXQDA) is used to learn a discriminative subspace that enhances the separability between different individuals. Mahalanobis distance is used to match and similarity computation between query and gallery samples. Finally, we evaluate our approach by conducting experiments on three challenging person datasets, VIPeR, GRID and PRID450s.

ABSTRACT. Driver drowsiness is identified as a leading factor behind traffic accidents and fatalities across the world. To deal with this critical public concern, researchers are developing a variety of driver-centered drowsiness detection systems. Nevertheless, the majority of studies investigating fatigue levels rely on either visual-based monitoring techniques involving physiological indicators or one single measure of signals. This paper presents a comparative study aimed at identifying the most reliable and promising method for detecting the stages of drowsiness using physiological signals. The study explores various techniques, both traditional and deep learning methodologies. The traditional methods focus on extracting features into temporal representation using time series and frequency representation through Fast Fourier Transform, Discrete Cosine Transform, and Wavelet analysis. The second approach utilizes spectrogram transformation, which converts a one-dimensional signal into a visual representation, subsequently employed within a deep learning model. The different approaches explored in this study have produced a multitude of results, however, the most promising, achieving an accuracy of 88%.

ABSTRACT. The recognition of text and characters in images, commonly called OCR (Optical Character Recognition) is still today a relevant field of image analysis. In this work, we tackle the text recognition problem by developing a system for reading Algerian drug labels. The processing of the label’s text starts by detecting and extracting the text regions by using MSER and SWT algorithms. The text detection procedure locates all text zones based on text properties. It selects the areas resembling letters, numbers, and some special characters, then it forms the words. The word-forming approach uses the STROCK algorithm which assembles nearby and adjacent detected text regions to determine the word regions that feed the recognition module. Our text reading model converts pixel text into string format (word or sequence of words). It uses a deep learning model consisting of a CNN, RNN, and a CTC layer to process visual data, recognize individual words from drug labels as strings of characters, and convert them into digital text. The experiments carried out on our VIG_Net dataset where our model trained on 30000 Algerian drug labels have allowed us to reach an acceptable word recognition accuracy of 63%.