Génie Eléctriques
Permanent URI for this communityhttps://dspace.univ-boumerdes.dz/handle/123456789/58
Browse
4 results
Search Results
Item Very high resolution multi-spectral remote sensing image cassifcation(Universite M'Hamed Bougara Boumerdès : Institut de Génie Eléctrique et Eléctronique, 2023) Semcheddine, Belkis Asma; Daamouche, Abdelhamid(Directeur de thèse)As an immediate consequence of the improvements in remote sensing sensors in terms of spatial resolution, and the increase of the amount of information recorded, traditional image segmentation tools became limited in accurately capturing the different land cover objects. This thesis is one more contribution in dealing with the complexity of very high resolution remote sensing image segmentation. We propose an improved version of matched filters, a tool to enhance and extract spatial information in VHR remote sensing images, in the aim to increase the class separability. With the help of particle swarm optimization, we designed an adaptive scheme that tailors matched filters kernels to supposedly meet the requirements of any study area. Traditional matched filters are in fact an out-dated method, mainly due to the choice of the kernels coefficients, which were specifically set for each experiment individually. Moreover, and mainly because of this limitation, matched filters were rarely employed as 2D filters. Our proposed approach automatically generates these kernels, and have shown considerable classification accuracy improvement. Additionally, our work includes various techniques for VHR image segmentation, where we explored textural information, data clustering and deep learning. Furthermore, we conducted a comparative study, contrasting pixel-based (SVM) with object-based (U-Net) image classification methods.Item Heterogeneous face matching(Université M'Hamed Bougara Boumerdès : Institut de génie électrique et électronique, 2023) Mahfoud, Sami; Daamouche, Abdelhamid(Directeur de thèse)Facial recognition technology faces a complex task in heterogeneous face matching, which involves computing the similarities between face images obtained through different modalities. Overcoming this challenge would enable the efficient matching of a vast collection of frontal photos from various visible face databases, such as passports, driver’s licenses, and mug shots, with face images acquired through alternate modalities, including hand-drawn face sketches, near-infrared photos, three-dimensional faces, and low-resolution photos. Heterogeneous face matching is a vital area of research that holds significant implications for law enforcement and digital entertainment applications, particularly when it comes to facial photo-sketch matching. This thesis puts forward two key contributions to the algorithms used in heterogeneous face matching, with a specific focus on matching face photos to hand-drawn face sketches. The first contribution is a framework designed for matching handdrawn face sketches to face photos. The precision of state-of-the-art face sketch recognition can be significantly improved by utilizing the Image Quality Assessment (IQA-Fusion) technique, which involves the fusion of Multi-Image Quality Assessment metrics. This approach enables the search for criminal suspects’ faces using hand-drawn face sketches based on verbal descriptions of a person’s appearance. The IQA-Fusion technique does not require training, which resolves the issue of the high cost associated with collecting large-scale photo-sketch pairs for training matches based on deep learning. The second key contribution focuses on a technique called cGAN-FSS, which involves using conditional Generative Adversarial Networks to generate facial sketches from facial photographs. The cGAN-FSS framework is capable of producing high-quality face sketch synthesis while maintaining a high level of accuracy in identity recognition. This contribution is considered valuable because it aids witnesses in visually recognizing criminal suspects and helps to bridge the gap between face photos and sketches in the pre-processing phase of facial recognitionItem Optimizing system reliability of stochastic-flow networks(Université M'Hamed Bougara : Institut de génie électrique et électronique, 2021) Aissou, Abdallah; Daamouche, Abdelhamid(Directeur de thèse)Components assignment problem (CAP) is used for searching the best set of components that maximize the system reliability subject to one or more constraints. Many research works have been devoted to the CAP under one constraint such as assignment budget, or total lead-time. Components assignment problem (CAP) based on three constraints namely assignment cost, total lead-time, and system reliability is never discussed. Therefore, this thesis investigates a new components assignment problem called optimal components assignment problem (OCAP) that takes into account three constraints: assignment cost, total lead-time, and system reliability. Subsequently, new and efficient optimization approaches are proposed: i) the first approach is based on a random weighted genetic algorithm (RWGA) is proposed to solve the OCAP and determine the most optimal solution characterized by maximum reliability, minimum assignment cost, and minimum total lead-time ii) the second approach is based on multiobjective particle swarm optimization (MOPSO) is presented to solve the multiobjective components assignment problem (MCAP) problem. The results revealed that MOPSO is more efficient in comparison with other optimization approaches based on single or multiobjective genetic algorithms. In addition, there is no need to convert the problem into minimization or maximization and normalize the solutions based on RWGA or NSGA approachesItem Automatic methods for the analysis and recognition of the Electrocardiogram of the electrocardiogram(Université M'Hamed Bougara : Institut de génie électrique et électronique, 2021) Belkadi, Mohamed Amine; Daamouche, Abdelhamid(Directeur de thèse)Cardiac diseases rank first in the cases of death all over the world; Electrocardiogram (ECG) bears valuable information about the person health state. Therefore, ECG became a standard tool for heart disease exploration. Beats segmentation is a necessary step before disease type identification. The segmentation is based on the QRS detection. In this thesis, we proposed three different methods for ECG segmentation. First, an optimized Pan-Tompkins algorithm is developed, in which the parameters of the benchmark algorithm are optimized using the particle swarm optimization (PSO). Second, the QRS is detected in the time-scale domain; the stationary wavelet transform is applied to the filtered ECG signal to enhance the QRS wave, and then thresholding is carried out to extract the wanted signal. Finally, a machine learning technique is used to identify the QRS. In particular, a deep learning autoencoder is trained by standard datasets for the purpose of QRS detection