Faculté des mathématiques et de l'informatique
Permanent URI for this communityhdl:123456789/17
Browse
7 results
Search Results
Item Intégration de l’Apprentissage Profond pour l’Optimisation du Routage dans les Réseaux FANETs(university of bordj bou arreridj, 2025) - KHIER Dounia; BELMOUMENE HoudaThis graduation thesis falls within the domain of mobile wireless networks, spe cifically Flying Ad Hoc Networks (FANETs), and proposes an intelligent approach to improve the stability of drone communications. The main objective is to design, mo del, and evaluate a predictive routing protocol capable of distinguishing stable links from unstable ones, in order to ensure better Quality of Service (QoS).The work be gins with an in-depth study of drone architectures in FANETs, followed by a critical analysis of conventional routing protocols and QoS management methods. Based on this analysis, we proposed an enhancement of the P-OLSR protocol by integrating a model based on Machine Learning and Deep Learning techniques to predict the Ex pected Transmission Count (ETX) metric, a key indicator of link quality.The proposed model was trained and validated using a dataset simulating FANET link characteris tics. A series of simulations was conducted to evaluate the performance of our protocol in terms of connection stability, throughput, latency, and routing reliability.The final performance of the model on the test set shows an accuracy of 98.58%, with a pre cision of 98.37%, a recall of 98.38%, and a low loss of 0.0612, demonstrating strong generalization ability.Simulation results also confirm the superiority of the proposed protocol, with a stability rate of 98.8%%, a reduced loss of 3%, a useful transmission rate of 97%, a resilience of 6%, and 92% active links, thus outperforming OLSR and Predictive-Optimized Link State Routing (P-OLSR).Item Traduction des documents arabes par les transformes(university of bordj bou arreridj, 2025) metaai ilhem Yahiaoui hadjer; Yahiaoui hadjerIn recent years, the field of machine translation has seen a remarkable development thanks to the rapid advancement of artificial intelligence (AI) technologies, especially with the emergence of deep learning-based Transformers models. These models have contribu ted to improving translation quality, especially when it comes to languages that are not standardized or lack sufficient linguistic resources, such as Algerian Darija. This study aims to build and evaluate a deep learning-based machine translation mo del for translating texts from Algerian Darija to English. To achieve this goal, a dataset containing sentences written in Algerian Darija and their corresponding English trans lations was collected and processed, and a model based on the Transformer architecture was trained using this data.Item Classification des IRM cérébrales Pathologiques avec Optimisation(university of bordj bou arreridj, 2025) Djebarni ilyesThe automatic classification of brain MRI images represents a fundamental challenge in the medical field, as it effectively assists practitioners in diagnosing cerebral pathologies. In this study, we explored several artificial intelligence approaches for multi-class brain tumor classification by evaluating different models : a traditional Convolutional Neural Network (CNN), the ResNet-18 model with and without optimization, the same model combined with the symbolic DRB classifier, and the Vision Transformer (ViT) model in both its base and optimized versions. Model optimization was performed using the Adam algorithm, known for its rapid convergence. Experimental results demonstrated varying performance across the tested models. The optimized CNN achieved a notable accuracy of 98.02%.confirming its effectiveness despite its relative simplicity. The optimized ResNet 18 delivered excellent performance with 99.08% accuracy and F1-scores exceeding 0.98 for most classes, indicating strong generalization and optimal model tuning. In contrast, its DRB classifier-integrated version without optimization reached 91.61%, highlighting both the potential and limitations of this combination without prior refinement. The ViT model’s case was particularly remarkable : without optimization, it achieved only 25.30% accuracy, reflecting poor initial learning capability. However, after optimization, its performance surged to 99.79%, making it the best-performing model in the study. This result clearly demonstrates the crucial importance of optimization in machine lear ning systems, particularly for advanced architectures. In summary, this work highlights the complementary roles of model architecture, classifier selection, and optimization tech niques in developing high-performance intelligent systems for brain MRI classification. The findings open promising avenues for integrating these models into clinical diagnostic tools to enhance accuracy and efficiency in medical diagnosisItem Brain MRI Image Segmentation Using YOLO and U-Net Deep Learning Models(university of bordj bou arreridj, 2025) Dhikra YOUSFIThe precise detection of brain tumors is crucial in medical practice, and the use of au tomated segmentation techniques has the potential to improve this significantly. The paper presents a comparative evaluation of U-Net and YOLOv8 deep learning models for automatic brain tumor segmentation from magnetic resonance imaging (MRI). U-Net, with its ability to achieve pixel-level accuracy, performed superiorly in terms of Intersection over Union (IoU) and Dice coefficient, indicating its robustness in boundary delineation. In contrast, YOLOv8 had better precision and recall, thus being more appropriate for fast segmentation. The com parison was made with the standard metrics of precision, recall, IoU, Dice coefficient, and F1 score, providing a well-rounded evaluation. The findings show that U-Net is better at gener ating precise segmentation boundaries, whereas YOLOv8 performs better when it comes to detecting tumors quickly and accurately. This comparison provides important insights for the determination of the most appropriate model depending on specific application requirements.Item Classification des émotions à partir de signaux EEGàl’aide de techniques d’apprentissage profond(university of bordj bou arreridj, 2025) BELARBI Chaima; KEDJOUTI AmelIn recent years, human emotion recognition from EEG signals has seen substantial progress, largely driven by advances in deep learning techniques. Unlike traditional methods based on facial expressions, EEG signals provide greater objectivity and robustness against voluntary manipulations. In this study, we propose an automatic emotion classification approach based on EEGsignals, utilizing the EEG Brainwave Dataset and the eeg-dataset-emotions. Two deep learning architectures—1D Convolutional Neural Network (1D CNN) and Long Short-Term Memory (LSTM) networks—were implemented to extract spatial and temporal features, res pectively. The experimental results demonstrate that the combined use of these models en hances the accuracy and reliability of emotional state recognition.Item Deep Learning Algorithms for Remote Sensing(university of bordj bou arreridj, 2024) Bouketir Hadda; Habbeche DouniaDeep learning has revolutionized the analysis of data collected from unmanned aerial vehicle (UAV) imagery, allowing for more profound insights, precise analysis, and enhanced data extraction. This advancement has significantly contributed to the refinement of semantic segmentation techniques. Particularly convolutional neural networks (CNNs) have emerged as powerful tools in this domain, outperforming traditional methodologies. Nonetheless, challenges persist, including feature extraction, class imbalance issues, overfitting, and vanishing gradients that hinder deep neural network training, consequently impacting segmentation performance. To address these challenges, we propose a novel approach by integrating the U-Net architecture with ResNet34 backbone leveraging its strong feature extraction capabilities. These features are further improved by using trained weights from the ImageNet dataset. We train and evaluate the proposed model on several UAV datasets, including Aerial Semantic Segmentation, LandCover.ai, UAVid, and AeroScapes. We achieve remarkable performance, higher accuracy, precision, recall, F1-score, and miou compared to other methods.Item Suivre les rumeurs dans les réseaux sociaux(2024) LAALAOUI Mouna; LAHRI SarahThe emergence of the Internet has transformed global communications, making information accessible at unprecedented speeds. With the emergence of social media, people can now connect, share and interact immediately with a variety of content. However, the ease with which information can be exchanged also facilitates the rapid spread of unverified rumors and false information. In this work, we aim to track and detect rumors, and to this end, we will present a model based on a deep learning approach using LSTM and RNN algorithms in order to obtain the best possible classification and more accurate and valid results.