针对遥感图像目标检测的改进YOLOv5s算法

Improved YOLOv5s Algorithm for Remote Sensing Image Target Detection

针对遥感图像中背景复杂、小目标分布密集以及易受环境因素影响导致检测性能不佳的问题,提出一种改进的YOLOv5s目标检测算法。首先,通过设计一种混淆鉴别注意力机制(Confusion-Distinguishable Attention,CDA)来避免目标与背景之间的混淆,提高对目标信息的关注度,能够有效提升目标检测的准确性和健壮性。其次,在原结构的颈部添加小目标检测层,解决小目标分布紧密、漏检的现象,从而提高算法的多尺度目标检测性能。最后,在DOTA数据集中进行实验和验证。实验结果表明,所提算法能够明显提高遥感图像目标检测的平均准确率。

Aiming at the problems of complex background,dense distribution of small targets and easy to be affected by environmenta l factors in remote sensing images,an improved YOLOv5s target detection algorithm is proposed in this paper.First,a Confusion Distinguishable Attention(CDA)mechanism is designed to avoidc onfusion between the target and the background,improve the atetntion to the target information,and effectively improve the accuracy and robustness of the target detection.Secondly,a small target detection layer is added to the neck of the original structure to solveh te phenomenon of tight distribution of small targets and missing detection,so as to improve the multi-scale target detection performance of the algorithm.Finally,the experiment and verification are acrried out in DOTA data set.Experimental results show that the proposed algroithm can significantly improve the average accuracy of remotes ensing image target detection.