Line-Focused Laser and YOLOv5-Based High-Precision Defect Detection for Reflective Surfaces

Abstract

This paper addresses surface defect detection for parts with highly reflective surfaces, proposing a machine vision-based line-focused laser inspection method. This method leverages the reflective and curved features of part surfaces, utilizing a line-focused laser to mitigate halo and reflection issues common in traditional lighting methods. By collecting and analyzing reflected laser line images, the system effectively detects and classifies surface defects. To enhance detection efficiency and accuracy, this study integrates a deep learning-based YOLOv5 model trained on an expanded dataset. A series of controlled experiments on 5086 defect samples demonstrate that YOLOv5 achieves a mean Average Precision (mAP) of 96.35%, significantly outperforming YOLOv3 and traditional vision-based approaches. The tested defect types include scratches, pits, and varying degrees of surface roughness, ensuring a comprehensive evaluation of detection performance. Specifically, YOLOv5 shows a 10.3% reduction in inference time compared to YOLOv3 while maintaining superior detection performance. The system processes images of 5496×3672 pixels in 0.744 seconds, meeting industrial demands for real-time, high-precision defect detection.