Discovering Roads from High-Resolution Satellite Imagery

Urban roads are prone to issues including insufficient road construction, poor topological connection, and low accuracy in satellite images because of the surrounding ground characteristics, such as building and tree shadows. It is often difficult to accurately separate the road from its background due to the intricate arrangement of the buildings and the sensor's acquisition geometry. These issues will also exist for mountain roads, such as hill shadow or vegetation occlusion. Significant progress has been made in recent years in creating advanced algorithms that can accurately and automatically identify and extract roadways from satellite photos. This study suggests integrating the optimal feature selection and Decision Tree Method with Hierarchical Framework and Ensemble Learning (HFEL). Convolutional neural networks (CNNs) of three distinct varieties, including AlexNet, LeNet-5, and ResNet are utilised by the HFEL.