At present, the defect detection of solar cells still depend on manual completion. The common surface defects are classified into four types: appearance defects, colour defects, cracks, and silk screen defects. For the detection of solar cell surface defects, there are hidden issues, low accuracy, and manual dependence. Therefore, fast and accurate locking of defects is the most critical step for improving the detection of surface defect defects. The convolutional neural network (CNN) with multilayer network architectures has a strong ability to extract the data features of the surface image and good pertinence to the detection of surface defects of solar cells. In this paper, it focuses on the key technology of surface defect detection for solar cells based on CNN. It will provide the technical support for establishing automatic intelligent detection system to detach from manual detection. The main research contents are: Firstly, by summarizing various imaging methods and common defect types on the surface of solar cells, the imaging methods and defect models are analysed to establish the relationship between them. Secondly, It is designed that a good visual acquisition system of solar cells by analysing the mechanism of common defect types and combining the surrounding environment to collect the image of battery surface in real time. Finally, the acquired image features are extracted and the new feature image reconstructed by using CNN. Then the reconstructed image and defect image of the algorithm are compared to detect the surface defects. The experimental results demonstrate that this research content of technology can be used for on-line detection of solar cell surface defects. It has higher detection accuracy, lower algorithm complexity and real-time performance than the traditional method.