Research On Ship Detection In SAR Images Based On Fully Convolutional One-stage Networks

A pixel-level denoising and semantic enhancement detection model based on fully convolutional one stage is proposed in order to solve the problems that the ship target is easily submerged in complex background,the edge is blurred by speckle noise and the small-scale ship target is easily lost after multiple convolution in SAR images.Firstly,the asymmetric convolution module is used to obtain features of different dimensions and further fuse them,so that the feature map contains more features that can reflect the information of ship targets.Secondly,the Transformer Encoder module is used to improve the context information between the ship target and the feature image,and enhance the dependency relationship between the ship target and the image.Finally,in the post-processing stage of feature map,a pixel-level denoising module is designed,which uses feature map convolution to generate prediction mask and further generate attention map.The feature map of each dimension is guided pixel by pixel,the weight of pixels is updated,the target area information is activated,and the non-target area information and speckle noise are suppressed.The test results on the open data set SSDD show that the detection accuracy reaches 96.73%,among which the detection accuracy for small-scale ships reaches 96.85%,the detection accuracy for large-scale ships reaches 96.64%,the detection accuracy for offshore scenes reaches 98.53%,and the detection accuracy for coastal scenes is90.00%,which verifies the effectiveness and generalization ability of the model.There are 31 figures,9 tables and 64 references in this paper.