Marine Impact Water Column Signal Detection Algorithm Based on Improved Yolov3

Aiming at the problem of low detection accuracy and easy missed detection in YOLOV3 algorithm, an improved YOLO V3 based detection algorithm for water column target at impact point at sea is proposed. Firstly, Mish activation function is used as the activation function of basic components for network to express deeper information. Then，a new detection module is constructed, which divides the input feature layer into two branches to compress and splice the features to realize the information interaction between channels. Finally, the feature layers of different scales are fused by using the idea of multi-scale fusion for reference to improve the network generalization ability. At the same time, Mosaic data enhancement method is adopted to reduce training threshold and enrich the background and small target of detection objects. Experiment results show that the proposed algorithm improves average accuracy by 5.39% and speed by 2.6 frames/second on the same data set, which can meet actual needs.