Target Detection for Forward Looking Sonar Image based on Deep Learning

Target detection is an important part of underwater target search and hostile target reconnaissance. This paper studies the problems of sonar image in target detection, such as strong noise, poor image quality and low resolution. This paper applies deep learning to sonar image recognition, in which YOLOX (You Only Look Once X) from YOLO series is used. YOLOX uses CSPDarknet, a backbone feature extraction network, and constructs FPN feature pyramid for enhanced feature extraction. Finally, Yolo Head is used to obtain prediction results. Simulation results show that the algorithm mentioned in this paper can recognize targets effectively from forward looking sonar images.