Abstract: Against the backdrop of rapid advancements in both data accumulation within large-scale sky survey projects and information technology, the automated classification of massive galaxy images using intelligent methods has become a cornerstone for understanding galaxy formation, evolution, and the cosmic environment. The accuracy of galaxy classification models is significantly correlated with the scale and balance of training samples, yet traditional data augmentation methods exhibit limited effectiveness in enhancing complex galaxy images. The study proposes a diffusion model-based algorithm for generating galaxy images, which produces galaxy images with realistic morphological features and textural details. The classification model trained using the augmented dataset demonstrates superior precision, recall, and F1-score compared to the model trained on the original dataset, with maximum improvements of 4.01%, 4.3%, and 5.06% in the three metrics respectively. These high-fidelity, clear generated images significantly improve the accuracy of galaxy morphological classification models. This process provides novel insights for future research on galaxy morphology classification in large-scale sky survey initiatives. Continued exploration of generative artificial intelligence models will provide powerful tools for astronomical research, enabling humanity to gain deeper insights into the mysteries of the universe.