Road Detection from Remote Sensing Images using Machine Learning

Road detection from remote sensing images is vital in various applications, such as urban planning, transportation management, and navigation systems. In this work, we propose a methodology for road detection using deep learning architectures, including ResNet, U-Net, DenseNet, and MobileNet, with the DeepGlobe dataset. This work aims to accurately identify and delineate road pixels within remote sensing imagery. The objective is to compare the performance of different models and determine the most influential architecture for road detection. The proposed methodology involves data preprocessing, implementing the deep learning architectures, training the models, and evaluating their performance using metrics such as Intersection over Union (IoU), Structural Similarity Index (SSIM), and Peak Signal-to-Noise Ratio(PSNR). Experimental results show that ResNet achieves the highest accuracy, followed by U-Net, DenseNet, and MobileNet. Future work includes exploring multi-segmentation for detecting other features in remote sensing imagery, ensemble models, transfer learning, addressing occlusions and environmental variability, and efficient inference. By leveraging deep learning architectures and the DeepGlobe dataset, this work contributes to accurate road detection in remote sensing images, enabling improved urban and transportation planning decision-making.