YOLOv7-Driven Visual Inspection System for Edge Banding Defects in Panel Furniture

Current quality inspection of edge banding in panel furniture heavily relies on manual screening, which is labor-intensive, subjective, and inefficient. To address this challenge, we propose a YOLOv7-based visual inspection system by integrating machine vision and deep learning. A dataset containing 1,887 images of six defect types (e.g., open glue, chipping, uneven trimming) was constructed, with annotations generated via LabelImg. Data augmentation strategies (rotation, scaling, cropping) were applied to enhance model robustness. The YOLOv7-Tiny model achieved a mean average precision (mAP) of 74.8 % at 57.63 FPS, outperforming traditional
methods and demonstrating superior speed-accuracy trade-offs. Experimental results on real-time industrial camera data validated the system’s capability to detect defects with high precision (82.1 %) and recall (75.4 %). This framework significantly reduces production costs and provides a scalable solution for automated quality control in furniture manufacturing.