In this paper,we propose an improved YOLOv5-based object detection method for radar images,which have the characteristics of diffuse weak noise and imaging distortion.To mitigate the effects of noise without losing sp...In this paper,we propose an improved YOLOv5-based object detection method for radar images,which have the characteristics of diffuse weak noise and imaging distortion.To mitigate the effects of noise without losing spatial information,an coordinate attention(CA)has been added to pre-extract the feature of the images,which can guarantee a better feature extraction ability.A new stochastic weighted average(SWA)method is designed to refine generalization ability of the algo-rithm,where the medium mean is used instead of their average value.By introducing an deformable convolution,both regular and irregular images can be proceeded.The experimental results show that the improved algorithm performs better in object detection of radar images compared with the YOLOv5 model,which confirms the effectiveness and feasibility of our model.展开更多
基金supported by the National Natural Science Foundation of China(6227323662136006+1 种基金62073215)Key R&D Projects in Hainan Province(ZDYF2024GXJS009).
文摘In this paper,we propose an improved YOLOv5-based object detection method for radar images,which have the characteristics of diffuse weak noise and imaging distortion.To mitigate the effects of noise without losing spatial information,an coordinate attention(CA)has been added to pre-extract the feature of the images,which can guarantee a better feature extraction ability.A new stochastic weighted average(SWA)method is designed to refine generalization ability of the algo-rithm,where the medium mean is used instead of their average value.By introducing an deformable convolution,both regular and irregular images can be proceeded.The experimental results show that the improved algorithm performs better in object detection of radar images compared with the YOLOv5 model,which confirms the effectiveness and feasibility of our model.
