RFE-YOLO: A Study on Photovoltaic Module Fault Detection Algorithm Based on Multimodal Feature Fusion

Area of Science:

• Electrical Engineering
• Computer Vision
• Renewable Energy Systems

Background:

• Photovoltaic (PV) module operational status is critical for power generation efficiency.
• Accurate fault detection is essential for intelligent operation and maintenance (O&M) of PV power plants.
• Single-modal image analysis faces limitations in complex PV environments.

Purpose of the Study:

• To develop a multimodal dataset and a novel deep learning model for precise PV module fault detection.
• To address the perceptual limitations of single-modal image analysis in PV systems.
• To enhance the intelligent O&M of centralized PV power plants.

Main Methods:

• Construction of an RGBIRPV multimodal dataset for PV power plants.
• Proposal of the RFE-YOLO model featuring RC, FA, and EVG modules for enhanced feature extraction and fusion.
• Utilization of CBAM-based attention mechanisms and GSConv for efficient multimodal data processing.

Main Results:

• The RFE-YOLO model demonstrated significant improvements over YOLOv11n, with a 2.9% increase in precision, 1.8% in mAP@50, and 1.5% in F1 score.
• The model effectively integrates visible and infrared data through differentiated feature enhancement and adaptive fusion.
• Achieved synergistic representation of shallow details and deep semantics with low computational cost.

Conclusions:

• The RFE-YOLO model provides an effective technical solution for rapid and accurate PV module fault detection in real-world conditions.
• The developed multimodal dataset and model enhance the intelligent O&M capabilities of PV power plants.
• This approach overcomes the limitations of single-modal analysis for improved PV system performance.