Mapping urban road networks using semantic approach

Abstract Over the last decades, urbanization and its impact on the demand for infrastructure have drawn much attention in global studies and policymaking. Road infrastructure has contributed to a compound annual growth of approximately seven percent. This quick growth of road infrastructure has brought several challenges. Meanwhile, with the availability of high-resolution remote sensing images, extraction of accurate road information has become a fundamental challenge in image processing and geospatial-related technologies. This study proposes a novel framework for road extraction and automated vectorization process. The proposed model excavates contextual information from high-resolution images to restore the topological information of road features. Further, we evaluate our model’s extraction capability on online and acquired datasets. The experimental results demonstrate that the proposed model outperforms the state-of-the-art architectures by obtaining 97.42% and 97.18% overall accuracy and 77.15% and 72.66% IoU for the Ottawa Road Imagery and Indian dataset, respectively. The extracted road features are further mapped, restoring the geospatial information. This analysis would help develop the database for the urban observatories and help sustainably plan future developments.