Master Informatique
Permanent URI for this collectionhdl:123456789/1389
Browse
4 results
Search Results
Item Eye tracking in simple visual search tasks(university of bordj bou arreridj, 2025) BOUHADDA KENZA; BOUDIAF FELLAEye tracking has become an essential technique for understanding human visual attention and behavior across a wide range of fields. One of the core challenges in this domain is ac curately predicting gaze during visual search tasks. As traditional models using handcrafted features often lack generalizability,Recent advances in deep learning offers a powerful alterna tives by enabling data-driven learning of complex spatial patterns in visual attention. This research introduces a deep learning-based eye-tracking system aimed at predicting vi sual attention through saliency maps, using the uEyes dataset which features a variety of image categories including desktop, mobile, web, and posters, along with corresponding human eye tracking data . the system employs a U-Net convolutional neural network optimized for pixel level saliency prediction. The model is trained and evaluated using a robust set of performance metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), Kullback-Leibler Divergence (KLD), Correlation Coefficient (CC), Histogram Similarity (SIM), and Accuracy. This system showcases the effectiveness of deep learning in modeling human visual behav ior in visual search tasks.Item Brain Tumor Detection Using U-Net and SVM(university of bordj bou arreridj, 2025) BENGUEZZOUMohammed; BENYAHIAOUI Mohamed AssilBrain tumors, particularly gliomas, pose a significant clinical challenge, requiring both precise localization and accurate grading to guide treatment. Accurate segmentation of tumor regions is a critical first step, enabling meaningful analysis and interpretation of the affected areas. In this project, we present a hybrid framework that first segments tumor regions in brain Magnetic Resonance Imaging (MRI) scans using a U-Net model trained on the Brain Tumor Segmentation dataset, and then classifies these regions as Low-Grade or High-Grade Gliomas with a Support Vector Machine (SVM) model based on features extracted from the segmented masks. On the held-out test set, our U-Net achieved an accuracy of 99.3%, while the SVM classifier delivered an overall accuracy of 93%.Item 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 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.