Underwater Target Detection Algorithm Based on Improved YOLOv5

Abstract

 In order to solve the problem of low accuracy of traditional methods in underwater target detection and recognition, an underwater target detection algorithm based on YOLOv5, adding Convolutional Block Attention Module (CBAM) and Bidirectional Feature Pyramid Network (BIFPN) is proposed in this paper. In this method, CBAM is introduced into the output of YOLOv5 network model, and the attention module is used to fuse and enhance the features in channel dimension and spatial dimension, so as to enhance the features of target area features and improve the detection accuracy of small targets. At the same time, the simplified BIFPN module is used to replace the original enhanced feature extraction network in the Neck to improve the feature extraction ability of the network for different scales. The experimental results show that the mAP_0.5 is 84.2%, which is 0.9% higher than YOLOv5s model, meeting the needs of underwater target detection tasks.