A photovoltaic surface defect detection method for building based on deep learning

Abstract

The inspection and diagnosis of building engineering involve health monitoring of buildings and related facilities, and the utilization of renewable energy, such as solar energy, is crucial for smooth operation of modern construction projects. The detection of solar panel defects is related to the reliability and efficiency of building photovoltaics and has become a field of concern. Using deep learning to detect defects can improve the stability of building photovoltaics. However, achieving a balance between algorithm accuracy and reasoning speed requires further study. This paper presents an improved algorithm based on YOLO-v5, named YOLOv5s-GBC, which improves accuracy and inference speed. This demonstrates the advantages of fast and accurate photovoltaic defect detection. Based on the classical YOLO-v5 algorithm, the attention mechanism and bidirectional feature pyramid network were adopted to improve the accuracy of defect detection. Then, the lightweight module GhostConv and the Gaussian error linear unit activation function were used to reduce the number of model parameters and improve the reasoning speed. Further, the defect dataset of electroluminescence images proposed by the 35th European Photovoltaic Solar Energy Conference and Exhibition was used to verify the effectiveness of the proposed method. The experimental results show that YOLOv5s-GBC is superior to the original method in many evaluation indices, i.e., the accuracy and inference speed were increased by 2% and 20.3%, respectively. In conclusion, YOLOv5s-GBC exhibited better performance compared to other deep learning methods.