Biofouling recognition and boundary tracking control for underwater cleaning robots.

Underwater climbing robots have shown outstanding potential for cleaning underwater structures. Improvements in robotic intelligence can further facilitate their development in engineering applications. In this study, we propose a deep-learning-based biofouling recognition algorithm and a computer-vision-based boundary tracking method for an underwater cleaning robot (UCR) to achieve navigation. A Fast-SCNN network model accelerated by TensorRT was developed to accurately recognize areas covered by biofouling, achieving a fast processing speed of 0.03 s per image. Subsequently, a contour point extraction algorithm was introduced to obtain the biofouling boundary. Furthermore, a tracking cruise control method was developed to guide the robot for boundary tracking based on the position of the boundary relative to the robot. The experimental results demonstrated the capability of UCR to automatically recognize biofouling and track the boundary for automated cleaning operations. • TensorRT-optimized Fast-SCNN network can realize real-time biofouling recognition. • A contour points extraction algorithm is presented to locate biofouling boundary. • The robot is steered to track biofouling boundary by visual guidance. • Experimental results verify the feasibility of the algorithms and control method. [ABSTRACT FROM AUTHOR]