An Improved Object Detection Method for Underwater Sonar Image Based on PP-YOLOv2

Forward-looking sonar is widely used in underwater obstacles and objects detection for navigational safety. Automatic sonar images recognition plays an important role to reduce the workload of staff and subjective errors caused by visual fatigue. However, the application of automatic object classification in forward-looking sonar is still lacking, which is due to small effective samples and low signal-to-noise ratios (SNR). This paper proposed an improved PP-YOLOv2 algorithm for real-time detection, called as PPYOLO-T. Specifically, the proposed method first resegments the sonar image according to different aspect ratio and filters the acoustic noise in various ways. Then, attention mechanism is introduced to improve the ability of network feature extraction. Finally, the decoupled head is used to optimize the multiobjective classification. Experimental results show that the proposed method can effectively improve the accuracy of multitarget detection task, which can meet the requirement of robust real-time detection for both raw and noised sonar targets.