GS-YOLOv8: An improved UAV target detection algorithm based on YOLOv8

In recent years, drone technology has developed rapidly and is gradually being used in the field of target detection, but there are problems such as a large number of targets, complex scenes, and a high proportion of small targets from the perspective of UAVs. In this paper, we introduce a novel small target detection model upon the basis of YOLOv8. Adding the GAM attention mechanism before Upsampling enhances the feature extraction capability in both spatial and channel dimensions. Feature fusion is performed before detecting the output, and the output part of the detection is extracted from the new fusion layer. The experimental results shows that the improved model target detection mAP is increased to 26.6%, which is 0.8 percentage point improvement on YOLOv8. The improved model has better performance in UAV aerial target recognition.