Improved Ship Detection Algorithm Based on YOLOX for SAR Outline Enhancement Image

Synthetic aperture radar (SAR) ship detection based on deep learning has the advantages of high accuracy and end-to-end processing, which received more attention. However, SAR faces many problems, such as fuzzy contour, complex background, large scale difference dense distribution small targets. To solve these this paper proposes a method with ultra lightweight YOLOX. Aiming at problem speckle noise blurred contour caused by special imaging mechanism SAR, feature enhancement frequency sub-band channel fusion makes full use information is proposed. requirement light-weight algorithms for micro-SAR platforms unmanned aerial vehicle defect spatial pooling pyramid structure damaging features, an ultra-lightweight performance backbone Ghost Cross Stage Partial (GhostCSP) dilation convolution (LSDP) designed. characteristics diversity unbalanced after in images, four layers are used to fuse contextual semantic attention enhancement, finally improved target YOLOX (ImYOLOX) formed. Experimental tests Ship Detection Dataset (SSDD) show that proposed achieves average precision 97.45% parameter size 3.31 MB model 4.35 MB, its ahead most current algorithms.