Research on Small Target Detection Algorithm for Outdoor Complex Environment Based on STB-YOLOv8

The task of object detection in complex outdoor scenes faces multiple challenges: differences in lighting conditions, dynamic changes in the object background, and incomplete targets due to occlusion. Existing research often focuses on the YOLO algorithm and optimizes it within the field of convolutional neural networks. However, the research on applying the Transformer algorithm to the field of image processing has a theoretical basis and good prospects, but still lacks extensive research and experiments. This paper proposes an improved YOLOv8n architecture, which replaces the C2f module located in the deep part of Backbone with Swin Transformer Block (STB) to take advantage of Transformer's global feature extraction capability. In addition, a small target detection head is added to capture the rich location information in the shallow layer of the network to improve the detection performance of small target objects. Simulation results show that the improved algorithm can effectively improve the detection effect, and the recall rate, F1-score, mAP50 and mAP50-95 are better than the original YOLOv8n model, increasing by 1.961%, 1.375%, 0.676%, 3.82% respectively.