Advancing spatial mapping for satellite image road segmentation with multi-head attention

Remote sensing imaging is an interesting field, particularly in road areas. Road segmentation has become crucial in several areas, such as transportation network optimization, urban planning, and image analysis. We proposed in this study an upgraded mixed-scale UNet network (MAP-UNet) with a multi-head attention mechanism to identify and delineate road networks within aerial images. This upgraded model identifies and delineates road networks within aerial images. Modified MAP-UNet aims to enhance the efficiency of road segmentation through the integration of multi-scale features and attention mechanisms. We performed a comparison using the most recent methods. Our proposed approach achieves recall (76.18%), precision (80.30%), and IoU (63.00%) threshold overtime on the DeepGlobe dataset.