Research on Hot Spot Fault Detection Method Based on Infrared Images of Photovoltaic Modules in Complex Background

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Detecting hot spot faults in photovoltaic modules is challenging due to low target pixel proportion and background interference.
• Existing methods struggle with multi-scale hot spot targets and balancing detection speed with accuracy.

Purpose of the Study:

• To propose an effective method for detecting hot spot faults in complex environments.
• To improve the accuracy and efficiency of photovoltaic hot spot fault detection.

Main Methods:

• Integrating U-Net for image segmentation to remove background noise and highlight panel contours.
• Developing a YOLOv8-based detection network incorporating deformable convolution (DCN) for multi-scale target adaptability.
• Designing the C2f_Ghost module to optimize network parameters for faster inference speed.

Main Results:

• The proposed method accurately detects hot spot faults, achieving an accuracy of 88.5%.
• Comparative analysis showed superior performance over SSD, YOLOv5, YOLOv7, and baseline YOLOv8.
• The integration of U-Net and YOLOv8 effectively addresses challenges of low target visibility and background interference.

Conclusions:

• The U-Net and YOLOv8 integrated approach provides a robust solution for photovoltaic hot spot fault detection.
• The method demonstrates significant improvements in accuracy and efficiency for complex environmental conditions.
• This technique lays a foundation for enhanced reliability and performance monitoring of photovoltaic systems.