UAV Aerial Image Target Detection Based on Improved YOLOv8

In view of the small target scale, low efficiency of target detection, inconsistent target scale, low detection accuracy, and serious problems of missing detection and misdetection in UAV shooting scenes, a small target detection algorithm based on improved YOLOv8 is proposed. Firstly, on the basis of the original YOLOv8, the branch of detecting large-scale targets is cut to reduce parameter calculation. Secondly, deformable convolution is introduced to adapt to input features of different scales to reduce the rate of missed detection. Finally, GAM attention mechanism is introduced to further improve the detection precision of the algorithm. Experiments show that the algorithm has obvious improvement effect on VisDrone data set, such as accuracy increased by 5.3%, recall rate increased by 2.3 %, mAP @ 0.5 increased by 3.3%, mAP @ 0.5: 0.9 increased by 2.1%, F1 value increased by 3.5%, and parameter quantity decreased by 750186.