http//localhost:8080/jspui/handle/123456789/580 2026-06-09T12:46:58Z 2026-06-09T12:46:58Z BOUATOUATA, Chakib http//localhost:8080/jspui/handle/123456789/13628 2025-11-30T09:21:42Z 2025-06-06T00:00:00Z

AI-Driven Cardiac Diagnostics: Leveraging Vision Transformers for PCG Signal Classification BOUATOUATA, Chakib work presents a comprehensive approach to classifying cardiac sound (PCG) signals into normal and abnormal categories using hybrid deep learning and ma- chine learning techniques. Our goal is to develop a solution based on artificial intelligence techniques. The study leverages the PhysioNet/CinC Challenge 2016 dataset, which consists of recordings from multiple environments and patient cases, introducing significant variability in signal quality, duration, and class balance. A robust preprocessing pipeline was implemented to normalize and segment the sig- nals by patient. Time-domain and frequency-domain features were extracted to improve representation. n our work, we proposed a hybrid architecture that inte- grates multiple models, long short-term memory (LSTM) networks to model tem- poral context, and an attention mechanism to weight the most informative heart- beats. The most relevant segments were selected using attention weights, forming the final input to a random forest classifier. Experimental results demonstrate that the proposed system achieves promising classification performance, proving its effectiveness in handling PCG variability and imbalanced real-world data. This work highlights the potential of hybrid deep learning methods in automated car- diac listening systems

2025-06-06T00:00:00Z DJEDIDI, Abdelghani http//localhost:8080/jspui/handle/123456789/13560 2025-11-19T11:12:35Z 2025-06-09T00:00:00Z

Classification Audio pour les Maladies Respiratoires DJEDIDI, Abdelghani Les maladies respiratoires constituent un défi sanitaire mondial majeur, nécessitant des outils diagnostiques rapides, fiables et accessibles. L’analyse audio des sons de toux offre une alternative non invasive prometteuse. Ce projet vise à développer un système basé sur l’intelligence artificielle pour la classification automatisée des maladies respira- toires à partir de l’audio de la toux, en utilisant le jeu de données COUGHVID annoté par des experts. Face aux défis tels que la variabilité acoustique et le bruit, nous avons exploré diverses techniques de prétraitement et d’extraction de caractéristiques, en nous concentrant sur la combinaison des coefficients MFCC et des spectrogrammes Mel. Plusieurs modèles de classification ont été entraînés et évalués sur ces caractéristiques. Notre approche méthodologique a abouti à la sélection d’un modèle XGBoost opti- misé comme étant le plus performant. Les expériences menées ont rigoureusement validé son efficacité : le modèle XGBoost final a atteint une précision remarquable de 98.01% sur l’ensemble de test (et 97.51% en validation) pour la classification des conditions res- piratoires basées sur les diagnostics experts. Ces résultats surpassent significativement ceux d’autres modèles testés, tels que Random Forest (96.20% test) et les approches d’Apprentissage Profond standard (94.12% test). Ce travail contribue une méthodologie optimisée combinant des caractéristiques acous- tiques pertinentes (MFCC + Mel-spectrogramme) avec un classificateur XGBoost per- formant, démontrant une haute précision sur des données validées cliniquement. Ces résultats soulignent le potentiel de l’analyse audio de la toux pour développer des outils de dépistage évolutifs et économiques, renforçant ainsi la télémédecine et les stratégies de santé publique.

2025-06-09T00:00:00Z RAIS, Tameur http//localhost:8080/jspui/handle/123456789/13555 2025-11-19T09:52:21Z 2025-06-10T00:00:00Z

Video Tampering Detection Based on Deep Learning Techniques RAIS, Tameur Dashcams have become standard in modern vehicles, capturing continuous video footage that can serve as vital digital evidence in legal disputes or insurance claims. However, the reliability of these recordings is threatened by various forms of video tampering, including video forgery, frame deletion, frame insertion, and even deepfake-based alterations. These challenges demand robust and intelligent forgery detection systems capable of identifying manipulations at both frame and video levels. This master thesis presents a deep learning-based approach for automated video tampering detection in dashcam recordings. It begins with an in-depth theoretical analysis of digital video forgery, outlining its common forms and the limitations of traditional forensic methods. The study then focuses on the use of Convolutional Neu- ral Networks (CNNs), known for their effectiveness in visual anomaly detection, and proposes a novel detection pipeline built upon the ResNet50 architecture. The main contribution is the implementation of a custom model named BinaryCNN, a CNN-based architecture derived from ResNet50 and fine-tuned to classify dashcam videos into “original” or “forged” (specifically for frame deletion forgery). The de- tection pipeline includes video preprocessing, frame extraction, individual frame classification, and score aggregation to determine video-level authenticity. The system was trained and evaluated using the publicly available Video Forgery Dataset (Kaggle), which contains thousands of dashcam videos, divided into classes representing different tampering techniques: frame deletion, insertion, duplication, flipping, rotation, and zooming. For this work, we focused on a binary classification task: detecting frame deletion forgery vs. original. The dataset includes 1181 videos per class, split evenly between training and testing sets, and covers diverse driving conditions. The proposed BinaryCNN model achieved an overall accuracy of 79%, with a precision of 81%, recall of 78%, and F1-score of 79%, demonstrating its effectiveness in identifying forged videos under real-world conditions. These results confirm the capability of CNN-based models to tackle video forgery detection with high reliability. Additionally, a user-friendly web interface was developed using Flask to demonstrate the real-time application of this deep learning video tampering detection system. Users can upload a dashcam video and instantly receive a classification result. This research contributes to the growing field of automated forgery detection and highlights the potential of deep learning techniques, especially ResNet50-based CNNs, in building scalable, intelligent, and robust systems for video tampering de- tection.

2025-06-10T00:00:00Z TRIKI Tarek, KALLA Abdelmounam http//localhost:8080/jspui/handle/123456789/13553 2025-11-19T09:34:29Z 2025-06-07T00:00:00Z

Mefria : Une application super multiservice TRIKI Tarek, KALLA Abdelmounam Mefria est une application mobile multiservices innovante conçue pour simplifier le quoti- dien des utilisateurs en centralisant plusieurs services essentiels au sein d’une seule plateforme. L’application propose initialement la prise de rendez-vous pour les coiffeurs et les coiffeuse, et prévoit d’étendre ses fonctionnalités vers la livraison de repas, les courses alimentaires, le transport de personnes, ainsi que la livraison de colis. Mefria vise à répondre aux besoins croissants des citoyens en matière de praticité, de gain de temps et d’efficacité, tout en soutenant la digitalisation des petits commerçants, artisans et prestataires de services. L’application se distingue par son interface conviviale, son système de réservation intuitif, ainsi que son approche locale et adaptée au contexte algérien. Ce mémoire explore les aspects techniques, ergonomiques et économiques de l’application, tout en analysant son impact potentiel sur les habitudes de consommation et la transformation numérique des services de proximité.

2025-06-07T00:00:00Z