A Bionic Robotic Fish Detection Method by Using YOLOv3 Algorithm

Bionic robotic fish has attracted increasing attention due to its potential applications. However, rapid positioning and precise control is a great challenge for the research of bionic robotic fish. In order to realize the bionic robot fish target detection, this paper proposes a bionic robot fish target detection method based on YOLOv3 with global vision. Firstly, the target detection data set of bionic robotic fish was generated and annotated. The data set included 4,074 sample images, including the motion states of robotic fish turning, swimming straight, stationary, and lighting changing. Secondly, The improved K-means clustering method was used to obtain a priori bounding box size suitable for robotic fish. Furthermore, by modifying and testing different internal parameters, a suitable robotic fish target detection model was obtained and tested. The experimental results showed that even in the case of deformation, light changing and water flow fluctuation, accurate detection can still be achieved. It verified the effectiveness of the proposed bionic robot fish target detection method based on YOLOv3.