Ves-YOLOv4:Vessel Target Detection Technology in Complex Sea Area
In order to solve the problem of low success rate of ship target detection in complex sea areas,a Ves-YOLOv4 based ship target detection network architecture is proposed in this paper.Mosaic data en-hancement and adaptive image scaling optimization are applied at the input side to improve the generalization ability of the architecture.Based on the attention mechanism,the feature attention module is built and embed-ded in CSPDarknet-53 to improve the feature extraction ability of the model in complex environment.The fea-ture fusion structure of PANet is optimized to solve the problem of feature disappearance caused by too deep network.Experiments are carried out in custom data sets.The results show that the complex vessel target de-tection algorithm used in this paper can identify target information such as bulk carriers,container ships,fish-ing boats,cruise ships and islands,and maintain high recognition accuracy and speed.On RTX3060,the mAP of the model can reach 85.2％and the FPS can reach 23.1,which can meet the basic requirements of re-al-time detection.
万方数据
维普
