Verification of Dam Spatial Location in Open Datasets Based on Geographic Knowledge and Deep Learning

Abstract

Dams are important water conservation hubs; however, dam locations provided by open datasets are often unreliable. The aim of this article is to provide a single, geographic location-reliable dataset of dams for the scientific community by fusing existing dam datasets and verifying these locations. Using Southeast Asia as the case study, we propose an efficient and automatic method to verify dam locations and developed a process framework. First, the possible location of a dam was obtained by analyzing its geographic location characteristics. Then, the deep learning method was used to detect dams. Finally, a variety of geographic knowledge was applied to comprehensively verify the detection results to obtain accurate and reliable dam location information. The fused dam dataset we produced includes the locations of 4493 dams in Southeast Asia, which were verified using the proposed framework. The verification results were then evaluated via manual visual inspection. The verification accuracy of the framework was 86.7%. The experimental results show that the proposed framework can quickly and reliably verify dam spatial locations and provide solutions for the spatial location verification of other remote sensing objects.