Browsing by Author "DOGHMANE, Rayene"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Detection et suivi de bords de routes dans les images d’UAV (Unmanned Aerial Vehicle)= Road edges detection and tracking from UAV Images
    (Université Badji Mokhtar Annaba, 2026-06-28) DOGHMANE, Rayene
    In recent years, with the advent of computer vision and deep learning, Convolutional Neural Networks (CNNs) have become foundational for image processing. CNNs excel at tasks such as object recognition, classification, and segmentation. Among these tasks, road extraction from UAV imagery has benefited from CNNs due to their ability to automatically learn complex features from images. However, for CNNs to perform optimally, the model architecture must align with computational constraints. U-Net is one of the most well-known deep learning methods for image segmentation and has demonstrated remarkable adaptability and performance on aerial images. It shows great promise for road detection. In this thesis project,an enhanced U-Net architecture is used to automatically detect road from UAV images. An opensource aerial road dataset is used for training the model. The model is improved by integrating new normalization layer. Image pre-processing techniques and augmentation methods are also used for improvement. At last, qualitative and quantitative statistics are used to illustrate superiority of our method compared to other baseline models.The experiments demonstrate that our enhanced U-Net model, called BGN- Net, achieved a Mean Intersection over Union (Mean IoU) score of 0.984 after just 25 epochs. This result indicates both faster convergence and superior segmentation quality compared to the selected baseline models. This thesis project is a research project carried out by the LASA Laboratory at Badji Mokhtar University of Annaba. The main objective of this thesis is to develop a methodology for automatic road detection from UAV images.