A Component-Level Defect Detection and Real-Time Localisation Method for Photovoltaic Arrays Using UAV-Based Infrared...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Engineering
• Computer Vision for Infrastructure Monitoring

Background:

• Photovoltaic (PV) module defects, such as hotspots, shading, and diode failures, critically impair energy generation efficiency and introduce safety hazards.
• Effective and timely detection of these defects is crucial for maintaining the performance and reliability of large-scale solar power stations.
• Current inspection methods can be time-consuming, labor-intensive, and may not provide real-time feedback.

Purpose of the Study:

• To develop and validate a real-time detection and localization framework for PV module defects using infrared images captured by unmanned aerial vehicles (UAVs).
• To improve the accuracy and efficiency of defect identification compared to existing methods.
• To enable precise component-level localization of defects within PV arrays for prompt maintenance.

Main Methods:

• Construction of a dedicated dataset comprising 5583 infrared/visible images under standardized conditions.
• Development of an improved rotating-bounding-box detector (YOLO-CLO) based on YOLOv8-OBB, incorporating a C3m module and an LSCD-OBB detection head.
• Implementation of a multi-feature image processing pipeline (gradient, grayscale, temperature, morphological cues) for defect identification.
• Integration of a component-level localization strategy using GNSS metadata, Hough transform, and K-means clustering.

Main Results:

• The YOLO-CLO detector achieved high performance metrics: 99.1% mAP@0.5, 96.7% mAP@0.5:0.95, with 59.88 FPS, 8.52 M parameters, and 23.6 GFLOPs.
• Detection accuracies for specific defects reached 96.97% (hotspots), 100% (diode failures), and 88.89% (obstructions).
• The localization strategy accurately recovered the row-column index of defective modules, outperforming baseline methods like YOLOv5 and Faster R-CNN.

Conclusions:

• The proposed framework provides a highly accurate and efficient solution for real-time PV module defect detection and localization using UAV-based infrared imaging.
• The developed YOLO-CLO detector and the integrated image processing and localization pipeline demonstrate superior performance over existing approaches.
• This method is suitable for engineering deployment in large-scale PV power stations due to its low hardware dependency and high accuracy.