Target Detection Algorithm for UAV Aerial Images Based on BMGS-YOLOv8

Aiming at the problem of detection difficulties caused by the small size of targets in UAV aerial images, which are easy to be occluded, as well as the densely populated targets such as pedestrians and vehicles, an improved BMGS-YOLOv8 algorithm is proposed in this paper. The algorithm first introduces the Biformer attention mechanism in the backbone network, which captures both global and local features by filtering key key-value pairs, and improves the efficiency of small target detection while reducing the amount of computation. Secondly, GSConv convolution is used to replace the standard convolution in YOLOv8, which effectively reduces the number of model parameters. Experimental results on the VisDrone2019 dataset show that compared with the benchmark model YOLOv8n, BMGS-YOLOv8 improves the mAP50 and mAP50-95 metrics by 0.8% and 0.2%, respectively, while the amount of model parameters is reduced by 7%.