Object detection method for UAV image based on rotation frame and position attention
As to the problems of complex UAV imaging background,large ground target size and different angles,object detection method for UAV image based on YOLOv5 model and oriented bounding box(OBB)mechanism was proposed in this paper.Firstly,the position attention module is added to the backbone network to make the model focusing on learning the texture features of positive samples.Secondly,a larger feature map is output in the feature enhancement network to further enhance the de-tection ability of small objects.Thirdly,the object frame positioning loss function with rotation angle is introduced at the output end in order to detect multi-angle objects more accurately.Finally,the training set is expanded to increase the number of samples and expand the diversity of the training set.The ex-perimental results show that the improved model improves the mean average precision(mAP)by 8.3%compared with the original model,and it has good generalization in a variety of complex environments,which is significantly superior to the current mainstream deep learning methods.Moreover,the FPS of the improved YOLOv5 model is 32 frames per second,and it can output detection results in real time.
万方数据
维普
