A Small Object Detection Algorithm Based on Dual Attention Feature Fusion
A high-precision small target detection algorithm in real sea area was proposed.Based on YOLOv3 as the basic model,the algorithm preserved more detailed feature information of small-scale targets by adding high-resolution feature layers,and improved the multi-scale feature fusion structure of YOLOv3 by using jumping residual connection to reduce the loss of target features caused by fea-ture fusion.Meanwhile,the information interaction between feature maps of different scales was en-hanced.A dual attention feature fusion module was designed to aggregate local features and global features and refine the fused feature maps to reduce the aliasing effect of the fused feature maps.The results show that the proposed method can effectively improve the detection accuracy of the model,es-pecially the detection accuracy of small-scale targets by about 15％,and the detection time has not in-creased significantly.
万方数据
维普
