Latent Diffusion Model-Based T2T-ViT for SAR Ship Classification

Recently, deep learning methods have been applied to ship classification in Synthetic Aperture Radar (SAR) images. However, because of the problem of imbalanced and insufficient samples in the SAR ship datasets, accurately identifying SAR ships still poses challenges. In this paper, we propose an improved T2T-ViT model based on the latent diffusion model, which expands the data set through image generation, and adds the SE attention mechanism to adjust the channel weight. To evaluate the effectiveness of the proposed method, training and experiments were conducted on the OpenSARShip 2.0 dataset. Our proposed model, in accordance with experimental results, achieves better recognition accuracy compared with existing models.