Aerial insulator defect detection method based on CWSP-YOLO.

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Unmanned Aerial Vehicle (UAV) technology is rapidly advancing, making aerial imagery crucial for intelligent power equipment inspection.
• Traditional insulator defect detection methods suffer from high false/missed detection rates and insufficient multimodal data fusion.

Purpose of the Study:

• To propose an improved YOLOv11-based model for enhanced insulator defect detection in power equipment.
• To address the limitations of existing methods by integrating multimodal data and advanced deep learning techniques.

Main Methods:

• An improved YOLOv11 model was developed, incorporating multimodal data fusion.
• Key features include cross-modal collaboration, wavelet-optimized C3k2 modules, channel attention mechanisms, and PIoU v2-based dynamic gradient optimization.

Main Results:

• The proposed model achieved a mean average precision (mAP) of 84.77% on a self-built dataset.
• Performance metrics included 94.53% accuracy and 82.38% recall, with a processing speed of 24 FPS, meeting real-time requirements.

Conclusions:

• The developed multimodal YOLOv11 model significantly improves insulator defect detection for UAV-based power inspection.
• The system provides reliable and efficient support for real-time power equipment monitoring.