Solar photovoltaic module defect detection based on deep learning

Defect detection for photovoltaic (PV) modules is crucial in their production process, but the dataset quality and complex defects limit the accuracy and speed of the detection. In this paper, a solar PV module defect detection method was investigated using electroluminescence (EL) images. To reduce useless information in the EL images, a PV module segmentation method was proposed to segment PV cells from PV modules. Next, aiming at the insufficient sample size and the imbalance between classes in the dataset, a hybrid data augmentation method was proposed. Then, we proposed an improved YOLOv8n model for PV cell defects with different shapes and small sizes. Experiments showed that the proposed model has good comprehensive performance compared with other SOTA models, with mAP50 reaching 0.943 at only 7.6 G Flops. In addition, the proposed method can complete the defect detection of a PV module EL image containing 144 PV cells within 3 s. Overall, the proposed method meets the requirements of accuracy and real-time detection, providing a feasible solution for defect detection in PV modules.