YOLOv8n-GSS-Based Surface Defect Detection Method of Bearing Ring

Area of Science:

• Industrial Engineering
• Computer Vision
• Machine Learning

Background:

• Industrial bearing surface defect detection is challenged by complex backgrounds, multi-scale defects, and inadequate feature extraction.
• Existing methods often struggle with accuracy and efficiency in real-world industrial settings.

Purpose of the Study:

• To propose an improved YOLOv8-GSS defect detection method for industrial bearing surfaces.
• To enhance accuracy, robustness, and detection speed for multi-scale defects.

Main Methods:

• Implemented lightweight GsConv convolutions in the neck module to reduce computational costs.
• Integrated SENetV2 attention mechanism in the backbone to improve feature representation.
• Utilized a self-built dataset of industrial bearing ring surface images for training and validation.

Main Results:

• Achieved 97.8% AP50 accuracy.
• Improved detection accuracy for large, medium, and small defects by 2.4%, 3.6%, and 2.3% respectively.
• Reached a detection speed of 115 frames per second (FPS).

Conclusions:

• The improved YOLOv8-GSS method demonstrates superior performance compared to mainstream algorithms.
• The proposed enhancements effectively address challenges in industrial bearing surface defect detection.
• The method offers significant improvements in both accuracy and detection speed for industrial applications.