Automatic Extraction of Water Bodies From High-Resolution Remote Sensing Imagery Using Deep Learning and GIS in Dubai

Abstract

Extraction of water bodies from high-resolution remote sensing (HRRS) imagery has gained more attention recently. Several approaches, methods, and technologies were developed to delineate water bodies from different remote sensing imagery varying in spatial, spectral, and temporal characteristics. This study puts forward an intuitive approach to extract the water bodies from high-resolution imagery using an integrated deep learning method to GIS modeling in Dubai. For this study, training data was extracted first. Next, a selected deep learning model for object detection is introduced. Then, the deployment of this model in various areas across Dubai is comprehensively analyzed, including recognition, classification, detection, counting, and quality estimation. Finally, limitations and recommendations to take into consideration are summarized. As a result of this study, 1635 features were successfully detected across Dubai Emirate. Furthermore, the evaluation tests performed comparing the generated results to the reference data suggest that higher accuracy for water extraction is achieved from the high-resolution remote-sensing images than traditional approaches such as supervised classification and photo interpretation. In contrast, the average overall accuracy reached 98% in urban areas and 99% in rural areas, respectively. This novel approach offers good opportunities for the Geographic Information Systems Centre (GISC) at Dubai Municipality, under multiple land-use scenarios, to reduce the heavy amount of expensive human labor involved in updating the records through field visits or photo-interpretation manual techniques.