Research on Cigarette Butt Small Target Detection Using an Improved YOLOv8s Algorithm

To resolve the issues of missed detection and low detection accuracy in cigarette butt small target detection, an improved YOLOv8s algorithm is proposed to detect small cigarette butt targets. A small object detection layer was added to the neck network of YOLOv8s, enabling the integration of shallow and deep feature information. This enhancement promotes the reuse of effective features, ultimately improving the accuracy of small target detection. CoT attention mechanism is introduced after the SPPF of the backbone to increase the degree of attention paid to small targets effectively improving the problem of missed detection of small targets. The effectiveness of the added modules is verified through ablation experiments and comparative experiments. According to the comparison results of the baseline model YOLOv8s, the improved model's Precision increased from 86.2% to 90.0%, the Recall increased from 80.8% to 81.2%, and the mAP50 increased from 85.1% to 86.4%. Experimental results indicate that the improved YOLOv8s significantly enhance the performance of the baseline model. Its main performance metrics surpass those of other models in the YOLO series, as well as competing models like Faster R-CNN and SSD.