A method of particleboard surface defect detection and recognition based on deep learning

Abstract

ABSTRACT Particleboard is a material for furniture and other wooden industrial products, directly impacting the service life of the final product. In this study, an innovative dual-attention mechanism, the Dense Connection Idea (DC-DACB), which combines Ganomaly and ResNet networks to detect and identify surface defects on particleboards was proposed. The model demonstrates superior performance in fine-grained direction. Evaluation metrics including precision, recall, average F1Score, and mAP are employed. Comparative analysis with Faster-RCNN and Yolo v5 baseline models reveals a 2.4% and 2.2% improvement in mAP for particleboard surface-defect detection. Moreover, the model exhibits excellent accuracy (93.1%) in recognizing five common defect types: shaving, scratches, chalk marks, soft spots, and adhesive spots. Comparative analysis with SE-Net, SA-Net, the original CBAM, and self-attention further supports its effectiveness in particleboard surface-defect detection. The integration of artificial-intelligence detection technology enables the timely detection of production process issues, reduces wood resource waste, and benefit the production of the enterprise.