| 摘要: |
| 针对合成孔径雷达(Synthetic Aperture Radar,SAR)图像中飞机目标尺度多样性及背景强散射干扰的问题，提出了一种基于坐标注意力和自适应特征融合的YOLOv4 SAR图像飞机目标检测算法。该方法首先在主干网络引入坐标注意力机制，以增强对于飞机散射点组合结构的聚焦能力以及抗背景干扰能力。其次，在特征增强网络中引入自适应特征融合机制，提高了对不同大小飞机的特征提取能力，同时改善了YOLOv4算法召回率和精确率不平衡的问题。最后，通过改进的 K-Means聚类针对飞机目标调整先验框的尺寸，提高了模型的定位精度。实验结果表明，改进算法召回率达到91.01%，精确率达到90.09%，AP0.5达到92.34%，分别较原YOLOv4算法提高2.49%，6.56% 和3.62%。 |
| 关键词: 合成孔径雷达(SAR) 飞机检测 注意力机制 特征融合 |
| DOI：10.20079/j.issn.1001-893x.221014002 |
|
| 基金项目:国家自然科学基金资助项目（61901445） |
|
| Aircraft Detection in SAR Images Based on Attention and Adaptive Feature Fusion |
| XIA Yifan,ZHAO Fengjun,WANG Yingjie,WANG Chunle |
| (1.Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100190,China;2.School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China) |
| Abstract: |
| For the problems of multi-scale aircraft targets and strong scattering interference in the background in synthetic aperture radar(SAR) images,an improved YOLOv4 SAR target detection algorithm based on coordinate attention and adaptive feature fusion is proposed.Firstly,the algorithm introduces the coordinate attention mechanism in the backbone network to enhance the focusing ability on the combined structure of aircraft scattering features and the resistance to background interference.Secondly,the adaptive spatial feature fusion mechanism is implemented in the feature enhancement network to improve the feature extraction capability for aircraft of different sizes,while improving the imbalance between the recall and accuracy rates of YOLOv4.Finally,the size of the prior anchor is adjusted for aircraft targets by improved K-means clustering to improve the localization accuracy of the model.The experimental results show that the improved YOLOv4 achieves 91.01 |
| Key words: synthetic aperture radar(SAR) aircraft detection attention mechanism feature fusion |
