基于视觉注意和语义信息特征融合的遥感图像目标检测方法

The Remote Sensing Object Detection Based on Visual Attention and Semantic Information Feature Fusion

针对遥感图像目标检测方法中存在的特征提取不充分、语义信息表达能力弱、小目标检测准确率低和定位不准确的问题,提出了一种基于YOLOv5和Swin Transformer的改进策略。实验结果表明,与传统方法以及其他改进策略方法相比,文中的方法在公共数据集DOTA和自建数据集SkyView上均表现出更高的检测准确率,性能优势显著。

Considering insufficient feature extraction,weak semantic information representation,low detection accuracy for small targets and inaccurate localization in remote sensing object detection methods,an improved strategy based on YOLOv5 and Swin Transformer was proposed.Experimental results show that,compared with the traditional method and other improved strategy methods,the proposed method shows higher detection accuracy rating on the public data set DOTA and the self-built data set SkyView,and the performance advantage is significant.