Abstract: Rapid seismic damage assessment for buildings is of a great significance for post-earthquake emergency rescue and recovery. Deep learning methods provide a key means for rapid seismic damage assessment. However, due to the limited availability of strong motion data, this method faces the challenge of insufficient samples with high destructive power. Therefore, this study proposes an intelligent and rapid seismic damage assessment method for buildings based on data augmentation and on deep learning. The method uses continuous wavelet transform to construct spectrum-compatible ground motion to augment the strong ground motion data. The augmented strong ground motion dataset and deep learning algorithm are then adopted to predict seismic damage of buildings. The method proposed is demonstrated using an RC frame and is compared with the widely used data augmentation method that involves adjusting ground motion acceleration amplitudes. The research results indicate that: The strong motion data augmentation method proposed in this study has been shown to generate strong motions with less variability in duration and their effects on structural responses compared to the amplitude modulation method; The spectrum-compatible strong motion data augmentation method proposed can improve the accuracy of deep learning prediction and provide an important means for strong motion data augmentation; and compared to traditional modulation adjustment methods, the method proposed in this study has a higher accuracy in predicting severe damage to buildings, providing an important support for the rapid seismic damage assessment method utilized in building structures.