As one of the most important renewable energy sources, so-lar energy is gaining more and more attention. However, in the manufacturing process, solar cells will have some surface de-fects, including broken gates, pasting spot, thick lines, dirty cells, missing corners, scratches, chromatic aberrations, etc. Solar cells with defects should be detect. In this section, the multi-spectral characteristics of solar cell surface defects are analyzed, and defect datasets are estab-lished. Then the solar cell CNN model and the multi-spectral solar cell CNN model are designed. The effect of model depth and convolution kernel size variation on the detection perfor-mance is discussed. The solar cell CNN m. Aiming at the wide variety of surface defects, various shapes, and severe background interference, the multi-spectral convo-lutional neural network model is proposed in this paper. Exper-imental results show that multi-spectral solar cell CNN model enhances the ability to extract multiple spectral information features, improves the ability to separ.
