Surface defect detection of ceramic disc based on improved YOLOv5s

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Surface defect detection in ceramic disks faces challenges with small defects, size variations, and localization accuracy.
• Existing methods struggle to consistently identify diverse surface anomalies.

Purpose of the Study:

• To enhance the YOLOv5s algorithm for more accurate and efficient detection of surface defects on ceramic disks.
• To address limitations in detecting small defects and improve localization precision.

Main Methods:

• Improved YOLOv5s algorithm with modified anchor frame structure for better generalization.
• Integration of ECA attention mechanism to boost small target detection accuracy.
• Comparative analysis against YOLOv3 and YOLOv4 models under identical experimental conditions.

Main Results:

• Precision, F1 scores, and mAP increased by 3.1%, 3%, and 4.5% respectively.
• Accuracy for crack, damage, slag, and spot defects improved by 0.2%, 4.7%, 5.4%, and 1.9%.
• Detection speed increased from 232 frames/s to 256 frames/s, outperforming YOLOv3 and YOLOv4.

Conclusions:

• The enhanced YOLOv5s algorithm significantly improves surface defect detection on ceramic disks.
• The model demonstrates superior performance in identifying small defects and achieving real-time detection.
• This advancement offers a more robust solution for industrial quality control of ceramic materials.