基于改进YOLOv5的SAR图像飞机目标检测

Detection of Aircraft Targets in SAR Images Based on Improved YOLOv5

针对当前SAR图像中飞机检测尺寸较小导致小目标检测率低、虚警率高的问题,提出一种基于YOLOv5的改进方法。先采用K-means聚类算法针对飞机小目标尺寸优化锚框,在主干网络融合Swin Transformer模块,同时引入自适应学习权重的多尺度特征融合机制和全局注意力机制(GAM),使网络跨越空间通道维度放大全局维度交互,提高模型捕获不同维度信息的能力;并且增加一个小目标检测层,提高网络对SAR图像飞机小目标检测能力。实验结果表明,相较于原YOLOv5方法,改进方法在SAR图像尺寸较小飞机目标的检测上具有更强的特征提取能力和更高的检测精度。

To solve the problem of low detection rate of small targets and high false alarm rate due to the small size of aircraft in SAR images an improved method based on YOLOv5 is proposed.First the K-means clustering algorithm is used to optimize the anchor frame for the size of the small aircraft target and the Swin Transformer module is integrated into the backbone network.At the same time the multi-scale feature fusion mechanism of adaptive learning weights and the Global Attention Mechanism(GAM)are introduced to make the network span the space channel dimension and amplify the global dimension interaction so as to improve the model s ability to capture information in different dimensions.A small target detection layer is added to improve the network s ability to detect small aircraft targets in SAR images.The experimental results show that compared with the original YOLOv5 method the improved method has stronger feature extraction ability and higher detection accuracy in the detection of small-size aircraft targets in SAR images.