AI Remote Sensing: Development of a Novel Road Extraction Approach Using Open Street View

Roads play a crucial role for government to plan infrastructure development. Smart Geographic Information Systems is used to design and implement a novel road extraction approach using Open Street map and deep learning techniques. The current study proposes an innovative approach for producing vast quantity of training data automatically and effectively, which is utilized to extract roads using open street map and Inception-v4 ResNet architecture. The performance of our approach is assessed using dice coefficient and loss function. The training accuracy and the validation accuracy were 91.52% and 89.73%, respectively. Moreover, an open toolkit is developed to be accessible to researchers and developers.