Abstract: Aiming at the problems of defect detection for aerospace solar cells, the machine vision and deep learning are combined to detect the surface defect of solar cells. The aerospace solar cells images are obtained through the vision detection system and the aerospace solar cells defect dataset is constructed according to the enterprise’s defect classification standard. So as to solve the problem of low recall rate caused by information loss of convolution and down sampling, the slicing technique is used to obtain the partial defect images of solar cells and the sub-image dataset is constructed. In order to avoid the overfitting problem caused by insufficient dataset in the model training process, the appropriate image enhancement methods are adopted to expand the dataset for different defects. The YOLOX-s algorithm is improved by using depth wise separable convolution, optimizing the loss function, adopting bilinear interpolation up sampling, and introducing convolutional block attention module, and the best comprehensive defect detection model for aerospace solar cells has been obtained. The effectiveness of the improved model has been verified through comparison of multiple detection accuracy indicators between the models trained by different datasets, as well as ablation experiments. The superiority of the improved model for aerospace solar cells defect detection is verified through comparative experiments between similar mainstream models.