A Research of Deep Learning on Target Detection of Underwater Sonar Images

Abstract

Deep learning techniques have shown success in recent years detecting visual targets. Whereas, these methods do not achieve the same excellent results in underwater sonar images target detection field due to the underwater sonar images differs greatly from the optical images. It is the primary objective of this study to apply typical algorithms to the domain of underwater sonar image object detection. The related algorithms Faster R-CNN, SSD and YOLO V5 which used in our experiments were introduced firstly. Then we make a comparative study on the effects of the typical target detection algorithms Faster-RCNN, SSD and YOLO V5 on underwater sonar images, and compared the effects of these typical target detection algorithms when applied to the field of target detection in images taken by underwater sonar equipment separately. On our underwater sonar images dataset, none of these algorithms performed as well as they did on VOC2007. And the results of comparing the expression of these several algorithms on the dataset might helpful for the researches of sonar technology and help to promote sonar's use in related detection tasks.