Detection of Printing Defects on Wood-Derived Paper Products Using an Improved YOLOv8n

Paper-based printing materials originate from the wood-based value chain–wood–pulp–paper–printing—and their yield reflects the utilization efficiency of pulp and paper resources. In roll-to-roll printing production, small printing defects (e.g., missing prints, smudges, cracks) often cause rework and scrap, thereby increasing the consumption of wood-derived materials. To improve resource efficiency, this study proposes a lightweight, improved YOLOv8n model for real-time small-defect detection. The Efficient IoU (EIoU) loss is introduced in the bounding box regression stage to improve localization accuracy, and a Squeeze-and-Excitation (SE) channel attention mechanism is embedded in the feature fusion stage to strengthen feature representation for small printing defects. Evaluations conducted on datasets collected from real production lines demonstrate that, with 3.02 M parameters and 8.1 GFLOPs, the model achieves mAP@0.5 = 94.1%, Precision = 95.1%, Recall = 94.3%, and an inference speed of 100.2 FPS, outperforming the baseline model. The proposed method contributes to reducing rework and material waste, supporting the efficient utilization of wood resources and the sustainable development of the paper-based packaging industry.