An automatic detection model for cracks in photovoltaic cells based on electroluminescence imaging using improved YOLOv7

The increasing interest in photovoltaic (PV) energy plants, one of the renewable energy sources, is because of its clean, environmental-friendly and sustainable energy production. Early detection of faults in PV modules is essential for the effective operation of the PV systems and for reducing the cost of their operation. In this study, an improved version of You Only Look Once version 7 (YOLOv7) model is developed for the detection of cell cracks in PV modules. Detecting small cracks in PV modules is a challenging task. These cracks can occur during production, installation and operation stages. Electroluminescence (EL) imaging test procedure is often used to detect these cracks. Defective images with linear and star cracks obtained from EL are collected. The ghost module and global attention mechanism (GAM) are integrated into the backbone of the proposed YOLOv7 model to enhance the network’s learning capability and reduce the number of network parameters. This integration aims to achieve efficient detection results with more features extracted from our network. An evaluation of the proposed YOLOv7 model’s ability to detect in PV cell cracks was conducted by comparing it with popular YOLO models. The improved YOLOv7 model achieves 88.03% of precision, 74.97% of recall, 80.97% of F1-score, and 84.02% of mean average precision (mAP). The experiments validate that the developed YOLOv7 model yields impressive results in detect the PV cracks.