Recognition and classification of water surface targets based on deep learning

Abstract

Abstract Aiming at the shortcomings of low recognition rate and low calculation rate of surface unmanned ship in complex time-varying water surface environment, a surface target detection method based on Faster R-CNN is proposed in this paper. Firstly, the water surface image was enhanced by McCann Retinex method to improve the image quality under complex background. Secondly, a water surface target data set was established. Finally, based on Faster R-CNN algorithm, VGG, Resnet and Inception network structures were employed to test and analyze the data set. The results show that the detection method proposed in this paper can effectively complete the identification and classification of six categories of common targets on the water surface, which has significant guiding significance for the autonomous obstacle avoidance and maritime search and rescue of unmanned ships.