This paper introduces the Wavelet Attention-Db5 Block (WT Attention-Db5 Block) for early detection of diabetic retinopathy (DR), a complication of diabetes that affects the eyes by damaging blood vessels in the retina. The WT Attention-Db5 Block focuses on high-frequency domain using Discrete Wavelet Transform (DWT), extracting detailed information while retaining essential low-frequency information. The proposed Wavelet-Attention Swin (WT-Swin) model, which integrates the WT Attention-Db5 Block with Swin Transformers, achieves significant improvements in classification accuracy on the APTOS 2019 dataset. For binary classification, the training and validation accuracies are 99.14% and 98.91%, respectively, for Swin-T, and 99.01% and 99.18% for Swin-B. For multi-classification, the training and validation accuracies are 93.19% and 86.34%, respectively, while the test accuracy is 86%. The WT-Swin model demonstrates promising results in classification accuracy, making it a valuable tool for early detection and treatment of DR, which is crucial for preventing vision loss. The study highlights the potential of combining wavelet analysis and attention mechanisms in medical image analysis, particularly for DR diagnosis.This paper introduces the Wavelet Attention-Db5 Block (WT Attention-Db5 Block) for early detection of diabetic retinopathy (DR), a complication of diabetes that affects the eyes by damaging blood vessels in the retina. The WT Attention-Db5 Block focuses on high-frequency domain using Discrete Wavelet Transform (DWT), extracting detailed information while retaining essential low-frequency information. The proposed Wavelet-Attention Swin (WT-Swin) model, which integrates the WT Attention-Db5 Block with Swin Transformers, achieves significant improvements in classification accuracy on the APTOS 2019 dataset. For binary classification, the training and validation accuracies are 99.14% and 98.91%, respectively, for Swin-T, and 99.01% and 99.18% for Swin-B. For multi-classification, the training and validation accuracies are 93.19% and 86.34%, respectively, while the test accuracy is 86%. The WT-Swin model demonstrates promising results in classification accuracy, making it a valuable tool for early detection and treatment of DR, which is crucial for preventing vision loss. The study highlights the potential of combining wavelet analysis and attention mechanisms in medical image analysis, particularly for DR diagnosis.