An Improved Target Recognition Algorithm for Urban Management Cases Based on YOLOv8

To address high computational costs and poor small-target detection in urban management, this paper proposes an efficient target recognition algorithm based on YOLOv8n. The algorithm enhances the backbone with a module based on BiFormer's two-layer routing attention to filter out irrelevant features, improve detail focus, and reduce missed detections. It incorporates a lightweight module with GSConv in the Neck for efficiency and replaces the loss function with SIoU to improve angle matching between predicted and actual boxes. Experimental results show that the improved model significantly outperforms common models like YOLOv5s, Faster R-CNN, and SSD, achieving 93.8% precision and 93.5% [email protected]. The algorithm balances high performance with a low parameter count, meeting accuracy and timeliness requirements for urban image target detection.