RGS-YOLO: A lightweight solution for surface defect detection in wind turbines

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Wind turbine defects impact energy generation and can cause significant damage.
• Existing defect detection methods face challenges with low efficiency and high costs in industrial settings.

Purpose of the Study:

• To design a lightweight and efficient defect detection model for wind turbines.
• To improve the accuracy and reduce the computational cost of wind turbine defect detection.

Main Methods:

• Developed a novel Cross Stage Partial with 2 convolutions and feature fusion-Receptive-Field Attention Convolution (C2f-RFAConv) module for enhanced feature extraction.
• Integrated Group Shuffle Convolution (GSConv) into the neck network for a lightweight yet accurate model.
• Designed a detection head (SCConv Head) using Spatial and Channel reconstruction Convolution (SCConv) for efficient classification and detection.

Main Results:

• The proposed model reduced size by 1.16 MB and floating-point operations by 3.5 G compared to YOLOv8n.
• Achieved a 3.7% improvement in mean Average Precision (mAP).
• Demonstrated effective lightweight performance with enhanced feature extraction capabilities.

Conclusions:

• The designed lightweight model offers a promising solution for efficient and accurate wind turbine defect detection.
• The integration of RFAConv, GSConv, and SCConv modules significantly enhances model performance and efficiency.
• The model's reduced size and computational cost make it suitable for real-world industrial applications.