YOLOX-DG robotic detection systems for large-scale underwater concrete structures

Abstract

Large-scale complex underwater concrete structures have structural damage and the traditional damage detection method mostly uses manual identification, which is inaccurate and inefficient. Therefore, robotic detection systems have been proposed to replace manual identification for underwater concrete structures in ocean engineering. However, the highly corrosive and disruptive environment of the ocean poses great difficulties for the application. Here, we develop a manta ray-inspired underwater robot with well controllability to establish the damage datasets of underwater concrete structures, proposing the YOLOX-DG algorithm to improve the damage detection accuracy, and integrating the model into the robotic detection systems for underwater concrete damages. Eventually, the system is used for ocean testing in real applications (i.e., underwater marine harbors around the East China Sea), and satisfactory detection performance is obtained. The reported manta ray-inspired robotic detection system can be used to accurately monitor and analyze the underwater regions.