Research on target detection method of side-scan sonar image based on improved Yolov8
In view of the fact that existing target detection methods are difficult to adapt to the high noise,multi-distor-tion,and feature-poor characteristics of side scan sonar images,we proposed a side scan sonar target detection method based on an improved Yolov8.In the network training stage,a RCS-OSA module was introduced into the main body of Yolov8 to further enhance the feature extraction ability of the main body of Yolov8.In the inference stage,the feature ex-traction ability of the network was enhanced by reparameterized convolution,which was simplified into a single branch to reduce memory consumption.Then the BiFPN was used to replace the feature fusion module of Yolov8,and by repeatedly applying top-down and bottom-up multi-scale feature fusion,the fusion results of different scale features were further optimized,thereby improving the adaptability to multi-scale features.The experimental results showed that the proposed method outperformed the original Yolov8 network in all quantitative and qualitative evaluations,with an average precision mean(mAP)increased of 6.3％.
万方数据
维普
