The paper introduces a novel Wavelet-Attention Swin model for automatic diabetic retinopathy (DR) classification. The model integrates a WT Attention-Db5 Block, which uses Discrete Wavelet Transform (DWT) to focus on high-frequency domains while preserving low-frequency information. This block enhances feature extraction, leading to improved classification accuracy. The model is evaluated on the APTOS 2019 dataset, achieving high accuracy for both binary and multi-class classification. For Swin-T, training and validation accuracies are 99.14% and 98.91%, respectively. For Swin-B, binary classification accuracy reaches 99.01% (training) and 99.18% (validation), with a test accuracy of 98%. In multi-class classification, the model achieves 93.19% (training) and 86.34% (validation) accuracy, with a test accuracy of 86%. The model outperforms existing methods in terms of accuracy and efficiency, demonstrating its potential for early DR detection and treatment. The study highlights the effectiveness of combining wavelet analysis with attention mechanisms in medical image analysis, particularly for DR classification. The proposed approach contributes to the development of new theories and improves the efficiency of existing models, with applications beyond DR classification. The model's performance is validated through extensive experiments on the APTOS 2019 dataset, showing its robustness and adaptability to different classification tasks. The study also discusses the importance of early DR detection in preventing vision loss and the potential of deep learning in addressing gaps in medical image analysis. The model's integration of wavelet attention and Swin Transformers provides a promising solution for accurate and efficient DR classification.The paper introduces a novel Wavelet-Attention Swin model for automatic diabetic retinopathy (DR) classification. The model integrates a WT Attention-Db5 Block, which uses Discrete Wavelet Transform (DWT) to focus on high-frequency domains while preserving low-frequency information. This block enhances feature extraction, leading to improved classification accuracy. The model is evaluated on the APTOS 2019 dataset, achieving high accuracy for both binary and multi-class classification. For Swin-T, training and validation accuracies are 99.14% and 98.91%, respectively. For Swin-B, binary classification accuracy reaches 99.01% (training) and 99.18% (validation), with a test accuracy of 98%. In multi-class classification, the model achieves 93.19% (training) and 86.34% (validation) accuracy, with a test accuracy of 86%. The model outperforms existing methods in terms of accuracy and efficiency, demonstrating its potential for early DR detection and treatment. The study highlights the effectiveness of combining wavelet analysis with attention mechanisms in medical image analysis, particularly for DR classification. The proposed approach contributes to the development of new theories and improves the efficiency of existing models, with applications beyond DR classification. The model's performance is validated through extensive experiments on the APTOS 2019 dataset, showing its robustness and adaptability to different classification tasks. The study also discusses the importance of early DR detection in preventing vision loss and the potential of deep learning in addressing gaps in medical image analysis. The model's integration of wavelet attention and Swin Transformers provides a promising solution for accurate and efficient DR classification.