Rotating target detection method based on improved YOLOv5

The invention discloses a rotating target detection method based on an improved YOLOv5. The method has certain universality in the rotating target detection direction, and the patent takes remote sensing image target detection as an explanatory case. For the problem that excessive background information is extracted in the feature extraction process, an attention mechanism is introduced and improved, and a convolution kernel with a smaller receptive field is adopted to give attention to a model space according to the features of a target in a remote sensing image. A residual module adopts a C3 structure to construct a deeper network, SPP is used for fusion of different scale features, an angle classification thought and a CSL label are introduced, and a Gaussian function is adopted as a window function, so that the model has the capability of learning an angle distance. 本发明公开了一种基于改进的YOLOv5的旋转目标检测方法。该发明在旋转目标检测方向上具有一定的通用性，该专利以遥感图像目标检测为说明案例。对于特征提取过程中提取到过多背景信息的问题，本文引入了注意力机制，并且对其做出改进，针对遥感图像中目标的特点，采用拥有更小感受野的卷积核来赋予模