Target Detection and Recognition of UAV Aerial Images Based on Improved YOLOv5

Aiming at improving the efficiency and accuracy of the target detection for UAV aerial images especially the small targets, an improved YOLOv5 method is proposed. Based on the original YOLOv5 deep learning framework, a prediction head and convolution block attention model are introduced to effectively extract the attention area and detect small targets. Moreover, several data augmentation methods are adopted to enrich the dataset. A self-established aircraft targets dataset is utilized for the training and verification of the proposed model. And the experiment results show that the improved YOLOv5 method can effectively detect and recognize the small target. Meanwhile, this algorithm improves the detection accuracy and speed without greatly increasing the number of model parameters, which can be applied for real-time target detection in the military and civil fields.