Informatique

URI permanent de cette collectionhttps://dspacee.univ-temouchent.edu.dz/handle/123456789/746

Parcourir par

Voici les éléments 1 - 20 sur 98

ELEMINATION DU COLORANT LE BLEU DE METHYLENE PAR UN MATERIAU PROVENANT DES DECHETS MARINS
(2026) CHERIF Fatima Zahra; BENCHEKOR Hassiba
Identification Of An Email Author Using Machine Learning And Natural Language Processing
(2026) ELMEGUENNI Hadjar; MOHAMMED SMAIN Bouchra; BOUCHAKOUR .E
Emails are often used in cybercrime today, so it is important to verify the identity of the email author. This paper proposes different Machine Learning models: Naive Bayes (NB), Logistic Regression (LR) and Sup- port Vector Machine (SVM), to solve the problem of anonymous email author attribution. The main task is to find the author of an anonymous email among the many suspected targets, to verify if an email was in fact written by the sender. In this project, the models were trained and tested using the same email dataset where we analyze writing style, vocabulary usage, and tex- tual patterns, plus date and time information using supervised machine learning in addition to natural language processing techniques, with the intention of combining multiple factors to identify the author. The tests proved that the accuracy varies from one model to another indicating that some are better than others. In email authorship verifica- tion experiments, usually the average accuracy reaches 89.9%. while our model’s accuracy rate to a well distributed dataset is 89,06% for SVM , 81,77% for Naive Bays and 88.02% for Logistic Regression. And with a not so well distributed dataset the accuracy rate is 91,40% for SVM model, 76,74% for Naive Bays and for Logistic Regression 92.29%. Proving that a good identification system relies on two aspects, the model chosen for the task and the dataset veracity.
Decentralized Authentication based on Artificial Intelligence and Blockchain for Internet of Things Environments
(2026) BENADLA CHAIMAA; BACHIR NOUR EL IMANE CHAIMAÂ; MEDEDJEL MANSOUR
La croissance rapide des écosystèmes de l’IoT soulève des défis majeurs en matière d’authentification, d’évolutivité et de détection des menaces en temps réel, car les architectures centralisées traditionnelles restent vulnérables aux points de défaillance uniques et à la compromission des identifiants. Ce mémoire propose un framework hybride décentralisé combinant une gestion des identités basée sur la blockchain et une détection d’anomalies pilotée par l’IA. Le système s’articule autour d’une architecture à quatre couches comprenant des signatures ECDSA légères exécutées directement sur les appareils IoT, une API Go à haute concurrence pour la vérification des preuves à divulgation nulle de connaissance (ZKP), une blockchain Ethereum privée de type PoA (Proof-of-Authority) pour la gestion d’un registre immuable, et un auto-encodeur BiLSTM pour l’analyse des menaces en temps réel. Une évaluation expérimentale menée sur des déploiements à échelle variable de 20 à 200 appareils simulés démontre un taux de réussite d’authentification de 99 % avec une latence de bout en bout comprise entre 4,9 et 13,3 secondes. Entraîné sur un ensemble de données complet de 187 183 échantillons issus des ensembles IoT-23, CSE-CIC-IDS2018 et Bot-IoT, le modèle BiLSTM atteint un score F1 de 94,40 % (Precision = 95,28 %, Recall = 93,53 %, ROC-AUC= 0,9375), surpassant de 22,55 points de pourcentage un modèle de référence « Isolation Forest ». Enfin, le framework démontre avec succès un mécanisme d’atténuation automatisé en boucle fermée qui isole les appareils compromis en quelques secondes après la détection d’une anomalie.
Deep RL with Optimization-Based Refinement for Mobility Prediction in Smart Cities Using MEC and V2X Communication
(2026) BENKRAMA Yassine; BENZERBADJ Ali
The rapid growth of smart cities and the proliferation of Vehicle-to-Everything (V2X) communication networks have created an urgent need for traffic prediction systems that are simultaneously accurate, adaptive to non-stationary conditions, and deployable within the stringent latency constraints of Multi-Access Edge Computing (MEC) platforms. This thesis proposes and validates a hybrid two-stage framework for real-time directional traffic trend prediction. In the first stage, a Proximal Policy Optimization (PPO) agent equipped with a Temporal Convolution Network encoder and a multi-head self-attention spatial integration module is trained offline on a 200,000-timestep synthetic V2X dataset generated over the real urban road network of Nördlingen, Germany, learning to classify the dominant directional traffic surge (North, East, South, or West) over a 60-minute prediction horizon from 15 roadside unit observations. In the second stage, the frozen PPO policy is augmented at inference time by a Genetic Algorithm (GA) that continuously optimises a 4-dimensional logit bias vector against a sliding window of recent prediction outcomes, correcting for systematic directional errors introduced by abrupt traffic regime shifts without any gradient- based retraining. Experimental evaluation against supervised LSTM, ARIMA, Kalman filter, and random baselines demonstrates that the PPO+GA system achieves 78.0% accuracy and 79.26% Macro-F1 on the static test set, and maintains 46.8% accuracy across four abrupt regime transitions a 50% relative improvement over the frozen LSTM (31.2%). A closed-loop validation in the SUMO traffic simulator on the Belgrade road network, with 74 signalized intersections, yields a 33.3% vehicle throughput gain and a 10.4% average speed improvement over fixed-time signal control without any retraining, demonstrating cross- network generalization. All results confirm that the proposed framework meets the computational requirements of real-time MEC deployment with an inference latency of 54.59 ms, well within the 60-second operational window.
Détection et classification de nodules thyroïdiens bénins et malins sur images échographiques par approches hybrides de Deep Learning
(2026) Bentahar Chaimaa; Benbahi Asma; Beddad Fatima; Bousahba Nassima
The automatic detection and classification of thyroid nodules has seen significant development in recent years, thanks to rapid advances in medical imaging techniques and the emergence of deep learning methods, which have proven particularly effective in the field of computer vision. These methods are distinguished by their ability to automatically extract discriminating features from images, without requiring complex preprocessing steps, thus enabling the construction of rich hierarchical representations of visual information. This development has contributed to improving the efficiency of detection systems, making them better suited to the visual com- plexity that characterizes many medical pathologies, particularly in the field of ultrasound ima- ging diagnosis, such as the detection and classification of thyroid nodules. This is precisely what we examined in our thesis. We propose an automated system for detecting and classifying thyroid nodules into two cate- gories (benign/malignant) from ultrasound images. This system leverages deep learning archi- tectures combined using hybrid strategies to ensure high performance and satisfactory clinical robustness. We proposed five hybrid approaches based on the YOLOv8 and YOLOE models. These two models were initially trained individually. We also developed two cascade archi- tectures (YOLOv8→YOLOE and YOLOE→YOLOv8) and two prediction fusion approaches (NMS and WBF). Experiments were conducted on the ThyroidXL dataset, which comprises 11,545 ultrasound images from 4,093 patients. Individual evaluation shows that YOLOE achieves the best mAP@50 (90.1%) and YOLOv8 the best mAP@50-95 (61.1%), justifying their com- plementary nature. Among the hybrid approaches, the WBF+NMS fusion performs best overall, exhibiting the highest F1 score (92.9%) and the best recall for malignant targets (98.50%), re- flecting an optimal balance between reliability and comprehensiveness of detections. The WBF Fusion approach stands out with a very close malignant-class F1-score (89.52%) and a mAP@50 (97.32%), while the YOLOv8→YOLOE cascade achieves the best mAP@50 :95 (73.70%), in- dicating more precise spatial localization. Considering all the criteria, WBF+NMS is selected as the optimal hybrid architecture.
Traitement des images médicales par Deep Learning : segmentation des polypes colorectaux pour l’aide au diagnostic
(2026) Bentsabet Seloua Amira; Boulefred Radouane; Belgrana Fatima Zohra
Colorectal cancer is one of the most pressing global public health challenges, with nearly 1.9 million new cases diagnosed annually. Colonoscopy remains the gold standard for its preven- tion; however, polyp miss rates ranging from 22% to 40% persist due to the high morphological variability of lesions and operator fatigue. This thesis presents CANet+ (Colorectal Attention Network Plus), an encoder-decoder Deep Learning architecture combining a ConvNeXt-Tiny en- coder, an Attention U-Net decoder, CBAM attention modules, an ASPP module for multi-scale context modeling, and a deep supervision mechanism.Evaluated on three standard benchmarks Kvasir-SEG, CVC-ClinicDB, and CVC-ColonDB using a four-fold cross-validation protocol, CANet+ achieves Dice scores between 0.92 and 0.96 and IoU values between 0.87 and 0.92, demonstrating competitive performance against methods. These results open concrete prospects for integrating Deep Learning based computer aided diagnosis and segmentation systems into real world clinical endoscopic practice.
An Adaptive and Behavior-Aware Aggregation Strategy for IOT-Based Hierarchical Federated Learning Systems
(2026) BOUDAOUD Aya; BENAMAR Sara; Imad Fethi BENARIBI
Federated Learning improves privacy by keeping data decentralized, making it particularly suitable for distributed Internet of Things (IoT) environments where large numbers of edge devices collaboratively train machine learning models. However, the lack of visibility into clients’ local training processes creates vulnerabilities to Byzantine attacks that can compromise the global model. Existing robust aggregation methods still face limitations due to their stateless behavior and reliance on limited behavioral indicators, which makes them less effective against adaptive adversaries. Therefore, a dynamic, multi-dimensional, and history-aware trust mechanism is required to achieve more reliable aggregation. This thesis proposes TWA-HFL, a dynamic trust-aware aggregation framework for Hierarchical Federated Learning that continuously evaluates client reliability throughout the training process. The proposed ap- proach combines multiple behavioral indicators, including update similarity, peer consistency, magnitude stability, and historical behavior, into an adaptive trust score. A temporal regulation mechanism is intro- duced to monitor trust evolution over time by rewarding consistent clients and reducing the contribution of potentially malicious participants. Furthermore, trust-enhanced clustering and weighted aggregation are applied to select reliable updates and improve resilience against Byzantine attacks. Extensive experiments were conducted on MNIST, FashionMNIST, and CIFAR-10 datasets under various attack scenarios and malicious client ratios. The experimental results demonstrate that TWA-HFL consis- tently outperforms traditional and robust aggregation baselines, achieving higher accuracy and stronger robustness against adaptive Byzantine attacks.
Dynamic Resource Allocation for 5G Network Slices Using a Hybrid DRL–GA Approach
(2026) REGHAOUAT Douaa El Aldja; DAHMANI Maroua; BENZERBADJ A
Dynamic resource allocation between co-existing network slices is one of the most chal- lenging problems in 5G Radio Access Network management. In a Vehicle-to-Everything (V2X) highway scenario, a URLLC slice serving safety-critical collision-avoidance messages and an eMBB slice serving high-definition cameras and passenger internet access must share a fixed 100 MHz radio pool under non-stationary, multi-regime traffic conditions. Satis- fying the strict latency constraint of the URLLC slice and the load-dependent bandwidth requirement of the eMBB slice simultaneously, at every one second decision epoch, is an NP-hard problem that classical and standalone intelligent methods cannot fully resolve. This thesis proposes a Hybrid Deep Q-Network + Adaptive Genetic Algo- rithm (Hybrid DQN+GA) framework that addresses three structural limitations iden- tified in the state of the art: the discrete action ceiling of DQN-based agents, the cold- start inefficiency of standalone Genetic Algorithms, and the absence of a bidirectional co-adaptation loop between learning and evolutionary search. At every decision epoch, the DQN provides a fast, experience-driven, discrete allocation seed; the Adaptive GA re- finer searches the continuous bandwidth interval when a constraint violation is detected, using a multi-directional warm-start strategy seeded from the DQN output and domain- knowledge candidates; and the resulting refined allocation is fed back to train the DQN, forming a self-improving co-adaptation loop. The framework is evaluated in a simulated environment with five traffic regimes Nor- mal, Rush-hour, Accident, Peak-eMBB, and Night over 500 training episodes and 100 greedy evaluation episodes, against a Static Allocation baseline and a standalone DQN. The Hybrid agent achieves an overall QoS satisfaction of 75.34% (vs. 74.10% for DQN and 65.96% for Static), an eMBB satisfaction of 53.49% (+11.0% over DQN), a per-step violation rate of 0.493 (−4.8% over DQN), and a mean reward of 7.51. During safety- critical accident events, the Hybrid agent allocates 69 MHz to the URLLC slice 19 MHz more than the DQN— maintaining a latency of 4.8 ms under near-maximum load. All decisions are produced with a computational overhead below 1 ms per epoch, confirm- ing practical deployability on a Mobile Edge Computing server co-located with the base station.
Conception et réalisation d’un système intelligent de gestion automatisée des présences basé sur la reconnaissance faciale : ClassTrack DZ
(2026) YAGOUBI Asmaa; ZIGH Hanaa; MERAD BOUDIA
Manual attendance tracking in higher education institutions remains time-consuming, error-prone, and susceptible to fraud. This thesis presents the development of ClassTrack DZ, an intelligent automated attendance system based on facial recognition technology. The study begins with a theoretical overview of artificial intelligence and deep learning, followed by a critical analysis of state-of-the-art detection and recognition algorithms. A rigorous experimental methodology was employed to reproduce reference benchmarks while evaluating a key innovation: the integration of the YOLOv8 architecture to replace conventional detectors (Viola-Jones, HOG, MTCNN). Hybrid pipelines were compared across two datasets using standardized metrics (accuracy, precision, recall, F1-score, and inference time). Results demonstrate that YOLOv8 significantly accelerates multi-face de- tection while maintaining optimal accuracy. The YOLOv8 + Dlib configuration emerged as the most suitable compromise for real-time deployment in lecture halls. The modu- lar Python-based implementation ensures scientific reproducibility and features a secure pedagogical interface compliant with ethical standards for biometric data. This work validates the operational viability of facial recognition for modernizing academic admin- istration and paves the way for future enhancements, including liveness detection and embedded platform deployment.
Hybrid CNN–Transformer Architecture for Semantic Segmentation of High-Resolution Satellite Images
(2026) MEZADA Sarra; KRADEM CHARAH Chaimaa Nesrine; MESSAOUDI M.A
Semantic segmentation of high-resolution satellite imagery is a crucial task for applications such as urban planning and environmental monitoring. However mod- ern satellite images are very complicated, containing objects of different sizes and irregular boundaries. Convolutional Neural Networks (CNNs) are powerful tools for extracting local details such as textures and edges but they struggle to capture long- range contextual information. On the other hand, Vision Transformers (ViTs) are good at context capturing on the global level but lack local precision and typically require massive amounts of training data. The project is intended to address this gap by developing a hybrid CNN-Transformer architecture. The network combines the strengths of CNNs for local feature ex- traction with the power of Transformers for global context understanding, leading to improved segmentation performance, particularly for large objects and complex landscape classes. The proposed hybrid model is designed, implemented, and tested on annotated satellite datasets, with performance measured using pixel accuracy and the Dice coefficient. The results show that the hybrid framework, especially with transfer learning, provides a significant improvement in edge detection and generalizes better in different environments. This work shows the importance of AI in remote sensing and offers a solid tool for automated decision-making systems.
segmentation of Brain Tumors from Medical Images: An Artificial Intelligence Approach
(2026) Taibi Fatima; Doukali Fatima Zahra; bendiabdallah Mohammed Hakim
Ordonnancement des tâches éco-énergétique dans les environnements Cloud et le Fog Computing
(2026) Boudieb Chaimaa; Tebbat Serina; BOUAFIA Zouheyr
This thesis investigates the scheduling of IoT tasks in Cloud and Fog Computing environ- ments, with the objective of minimizing energy consumption and execution time while ensu- ring satisfactory Quality of Service (QoS). Various scheduling approaches were analyzed and compared, including traditional methods as well as intelligent techniques based on Reinfor- cement Learning. A simulation developed using SimPy and Python was used to evaluate the performance of these algorithms under different scenarios. The results demonstrate that Rein- forcement Learning-based approaches improve the overall efficiency of the system in terms of energy consumption, makespan, and deadline compliance. In particular, the Double Deep Q-Network (DDQN) approach showed superior adaptability in distributed and heterogeneous environments.
Modèle intelligent pour le diagnostic médical Application aux données multimodales
(2026) Bouhacida Ibrahim; Saïm Sofiane; BENOMAR MOHAMMED LAMINE
Plateforme Intelligente de e-Santé pour la Gestion des Rendez-vous Médicaux, du Carnet de Santé Numérique et la Prédiction des Maladies basée sur l’IA – MyVital –
(2025) BOUDIEB, Mohamed Anes; RACHEM, Hanane; BOUHALOUAN, Djamila
This thesis represents the final document of the end-of-study project for obtaining the Master’s degree in Cybersecurity and Artificial Intelligence. It focuses on the study, design, and development of MyVital, an intelligent e-health platform aiming to modernize the Algerian healthcare system. This platform enables users to manage their healthcare journey by simplifying online medical appointment booking, centralizing all their data in a secure digital health record, and assessing disease risks through a predictive module based on artificial intelligence. The ultimate goal of this work is to create a practical and integrated solution that facilitates communication between patients and healthcare professionals, while improving access to care and strengthening prevention. This document provides a solid foundation for future developments of the application, ensuring its continuous improvement and adaptation to user needs.
Détection et reconnaissance d'objets dans des Images sous marines basée sur le deep learning pour application robotique -Neptunet Net-
(2025) ghitri, hamza; Elhadj Mimoune, Mourad; belgrana, Fatima Zohra
In the context of underwater inspection and environmental monitoring, this thesis presents our start-up project “NeptuNet,” a deep learning-based system designed for the automatic identification of underwater gas leaks through marker object detection. The proposed system offers two complementary solutions: a lightweight model, YOLOv5n, suitable for real-time deployment on embedded hardware, and a high-precision model, Faster R-CNN with ResNet-50 backbone, intended for offline analysis. Our approach relies on using computer vision techniques to address the challenges posed by underwater images, such as low contrast and turbidity. The experiments focused on the detection of gas pipes (for navigation assistance), bubble plumes for identifying active leaks, and bubbles considered as a potential leak indicator. Finally, the system was deployed as a web platform, allowing users to test the detection models on their own images or videos.
Segmentation Sémantique des Images Satellitaires pour la Surveillance Environnementale
(2025) LECHLECHE, Maroua; MAKNI, Fatima Zohraa; MESSAOUDI, Mohamed Amine
This project focuses on the semantic segmentation of satellite images using deep lear- ning architectures. The task involves assigning a label to each pixel in an image to automa- tically identify the different elements present in a scene (bare soil, vegetation, urban areas, water bodies, etc.). Following an in-depth study of classical segmentation approaches and deep learning principles, several architectures were implemented and compared. Experiments conducted on an annotated satellite image dataset enabled the evaluation of each model’s performance using metrics such as accuracy, Dice coefficient, and confusion matrix analysis. The results show that models integrating pre-trained backbones provide finer and more generalizable segmentation, especially in visually heterogeneous areas. This work highlights the growing importance of artificial intelligence in the processing of remote sensing data and paves the way for automated decision-support systems in critical fields such as environmental monitoring, security, and risk management.
Predict Microsatellite Instability (MSI) and Microsatellite Stability (MSS) status in gastrointestinal cancers
(2025) Hibi, Lara; Mekadid, Hafsa Amina; Beddad, Fatima
In colorectal cancer (CRC), MSI status is the crucial biomarker that influences treatment approaches, especially immunotherapy. However, the current diagnostic techniques, such as polymerase chain reaction (PCR) and immunohistochemistry (IHC), are time-consuming and variable.This study uses histopathological images to automatically predict the Microsatellite Instability (MSI) and Microsatellite Stability (MSS) status in gastrointestinal cancers using a novel deep learning framework. We suggest an AI-powered approach, Deep learning models to address these issues by combining cutting-edge computer vision methods for nuclear feature segmentation and classification in histopathology slides.The results we achieve shows how deep learning can revolutionise the digital pathology by providing a reliable, accurate, and scalable substitutes for conventional MSI and MSS diagnostics. With implications for enhancing patient outcomes by quicker and more accurate diagnosis, this work is an important breakthrough towards achieving precision oncology.
Réduction de la consommation d’énergie dans le Cloud Computing.
(2025) Benbouha, Djihene; Hassini, Imene; Bouafia, zohir
Cloud computing is a promising technology that facilitates the execution of applications in various fields. It provides flexible and scalable services on demand with improved quality of service. Ho- wever, this flexibility leads to high energy consumption in data centers, which are the backbone of cloud infrastructure and whose energy needs are growing rapidly. Faced with this significant en- vironmental and economic challenge, reducing the energy impact of this infrastructure is a major priority. In this final year project, we focused on reducing energy consumption in cloud computing, in par- ticular by implementing three main approaches : dynamic voltage and frequency scaling (DVFS), dynamic consolidation of virtual machines, and a combination of these two techniques. We simula- ted these techniques using the SimPy library in Python and analyzed their impact on several key indicators such as energy consumption, execution time, and execution deadline compliance. Our results show that DVFS is suitable for reducing energy consumption (up to 52%), while dyna- mic consolidation optimizes energy consumption by up to 22%, and the combined method achieves a reduction of 49%. This combined approach yielded significant performance gains, particularly for high workloads.
Segmentation d’images urbaines par Deep Learning
(2025) BOUAALAOUI, CHAIMA; SLATNA, AICHA; SAIDI
Semantic segmentation is a major area of computer vision that aims to assign a precise label to every pixel of an image, thereby enabling a deep and comprehensive understanding of its visual content. Since the rise of deep learning, this field has witnessed significant advances with the introduction of Convolutional Neural Networks (CNNs), which have demonstrated remarkable efficiency in extracting and representing visual features. Among the most popular models is U‐Net, initially designed for precise segmentation in medical contexts, which we have adopted as the foundation of this work. Structural modifications and additional dedicated layers have been introduced to enhance its ability to segment dense visual environments, such as the urban scenes found in the Cityscapes dataset, renowned for its richness and precise annotations. These improvements have significantly increased the model’s precision, while making the learning process more stable, faster, and better suited to the diversity of modern urban environments. This work thus provides a solid methodological basis for designing models adapted to the precise segmentation of complex visual environments.
Deep Learning et l’Aide au Diagnostic Médical (Application à la Classification d’image médicale)
(2025) Bensafi, Basma; Soussi, Amina Nour El Houda; Benomar, Mohamed Lamine
Medical image classification has seen significant progress in recent years, driven by rapid ad- vancements in medical imaging techniques and the emergence of deep learning methods, which have proven highly effective in the field of computer vision. Deep learning is characterized by its ability to automatically extract features from data without requiring complex preprocessing steps, enabling the construction of rich hierarchical representations of visual information. These developments have enhanced the efficiency of classification systems, making them better suited to handle the visual complexity inherent in many medical conditions, particularly in medical imaging diagnostics, such as diabetic retinopathy classification. This is precisely the focus of our research. Accordingly, we designed an automated system to effectively classify diabetic re- tinopathy using deep learning networks, specifically leveraging transfer learning to ensure high performance and reasonable computational time. We proposed three architectures based on the convolutional models DenseNet121, ViT-B/16, and YOLOv8, which are renowned for their ef- fectiveness on the ImageNet dataset. Our contribution involves classification using two distinct strategies : Direct classification, where the model is applied directly to the images. Cascade classification, where images are classified in stages—first into 2 classes (binary), then 3 classes, and finally 5 classes. The experiments were conducted on two datasets : The first dataset used is APTOS2019, comprising 3,662 images distributed across five (5) distinct classes. The results demonstrated a clear superiority, achieving a maximum accuracy of 96.30%. The second dataset is the DDR (Dataset for Diabetic Retinopathy), consisting of 12,522 images also divided into five (5) distinct classes. The results were satisfactory, with dataset accuracy reaching 93% and 96% respectively.

Parcourir par

Dépôts récents