Target detection in sonar image based on faster RCNN
Jie Fang, Wang Pingbo · 2020
Aiming at the limitation of underwater target detection method, a detection method based on Faster RCNN is proposed. Using a variety of convolution neural networks to extract image features, constructing a region recommendation network (RPN) to extract the region that may contain the target, and then using the detection network to determine the target area. The application of transfer learning method to ensure that the limited amount of data can also be trained and achieve good results. The experimental results show that the average accuracy of the proposed method in underwater target detection task can be improved by about 14% compared with the basic RCNN method and about 18% higher than the more new YOLO method.