Insulator Defect Detection Based on ML-YOLOv5 Algorithm

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Current insulator defect detection methods face challenges in balancing accuracy, speed, computational parameters, and FLOPs.
• High-altitude insulator inspection requires efficient and accurate automated defect detection systems.

Purpose of the Study:

• To develop an enhanced insulator defect detection algorithm that optimizes speed and accuracy while reducing computational load.
• To improve the efficiency and applicability of deep learning models for real-time inspection of high-altitude insulators.

Main Methods:

• Proposed ML-YOLOv5 algorithm based on the YOLOv5 network.
• Incorporated depthwise separable convolution in the backbone and an improved C2f_DG module for feature fusion.
• Enhanced the Multi-scale Feature Pyramid Network (MFPN) and utilized knowledge distillation with YOLOv5m as the teacher model.

Main Results:

• Achieved a 46.9% reduction in parameter count and a 43.0% reduction in FLOPs.
• Maintained a frames per second (FPS) of 63.6, indicating high detection speed.
• Demonstrated good accuracy and detection speed on both CPLID and IDID datasets.

Conclusions:

• The ML-YOLOv5 algorithm effectively balances accuracy and speed for insulator defect detection.
• The proposed enhancements make the algorithm suitable for real-time, high-altitude insulator defect inspection.
• This work contributes to more efficient and practical automated visual inspection systems in the power industry.