Understanding Defect Detection in Solar Cells: Challenges and Deep Learning Approaches with Electroluminescence Imaging

Solar energy has gained considerable attention in recent years due to the pressing need to minimize carbon emissions and fight climate change. Photovoltaic (PV) solar cells are crucial in the harvesting of solar energy, but their efficiency and longevity potential is subjected to the effects of defects (e.g., microcracks, finger interruptions). Identifying and diagnosing these defects is crucial for keeping systems running in optimal performance. Although current systems can achieve good performance using Convolutional Neural Networks (CNNs) with encoder-decoder structures, they do not perform well when modelling long-range dependencies, e.g., in situations with complex backgrounds or ambiguous pseudo-defects. Transformers are good at capturingglobal dependencies at the cost of losing fine-grained local structural information. In detail, this paper presents ahybrid model of CNNs and transformers to eliminate the limitations of existing SSL techniques by leveraging thestrength of EfficientNet and transformers architectures. This design integrates local and global features to enhancedefect detection capabilities, offering a robust solution for improving the performance and reliability ofphotovoltaic systems