Target Recognition and Detection in Side-Scan Sonar Images based on YOLO v3 Model

In recent years, the target recognition of sonar images is an important link in the realization of underwater target search, path planning, hostile target reconnaissance, seabed rescue, seabed texture survey, etc. Therefore, improving the accuracy of target detection on sonar images is very important for underwater detection. The paper studies the problems of low accuracy, low efficiency, and low missed detection rate in sonar image target detection. Deep learning is applied to sonar image recognition in this paper. The convolutional neural network and You Only Look Once version 3(YOLO v3) model are used to perform one-time neural network processing on side-scan sonar images. The Darknet53 network is used as the backbone extraction network, and the bounding box is used at the same time. For the low-credibility bounding box, multiple high-credibility maximum bounding boxes are selected to improve the success rate of image feature extraction. It is proved by simulation that the YOLO v3 model can effectively identify targets from side-scan sonar images.