Cigarette appearance defect detection method based on EBF-YOLOv8

Detecting defects in cigarette appearance faces three key challenges: limited category coverage, difficulty in recognizing small-scale defects, and the trade-off between detection speed and accuracy. To address these issues, this study proposes an enhanced model, EBF-YOLOv8, based on YOLOv8n. The model integrates the Efficient Channel Attention (ECA) mechanism to strengthen feature channel representation and employs a BiFPN structure for bidirectional multi-scale feature fusion, thereby improving detection performance for small targets and complex defects. Experiments conducted on a self-constructed cigarette defect dataset demonstrate that EBF-YOLOv8 achieves 95.94% Precision, 89.93% Recall, and 93.8% mAP@0.5, outperforming mainstream object detection models. Moreover, it maintains high inference speed, meeting the requirements of real-time industrial inspection. This study provides an effective solution for automated cigarette quality inspection.