The accurate and timely diagnosis of diabetic retinopathy (DR) is critical for effective treatment and management of this progressive disease. In this study, we propose a deep learning model called U-Net for the classification and identification of the severity of diabetic retinopathy. The U-Net model utilizes a convolutional neural network (CNN) architecture and focuses on analyzing blood vessel thickness and dilation, which are early signs of retinopathy. The model trained on a specifically curated dataset called FIVES, designed for this purpose. Benchmarking the U-Net model against various existing approaches in the field, our results demonstrate its exceptional performance, achieving an accuracy of 98.48%. This accuracy surpasses the majority of other methods, positioning the U-Net model as the most accurate approach among those considered. This high accuracy suggests that the U-Net model can reliably diagnose diabetic retinopathy, making it a valuable tool in the healthcare domain. Early detection of diabetic retinopathy is crucial to effective treatment. In addition, the U-Net model's high accuracy enables the identification of retinopathy. This facilitates timely intervention and improves patient outcomes. Additionally, the scalability and accessibility of DL models allow for the deployment of the U-Net model in various healthcare settings, including remote or underserved areas, where access to specialized ophthalmologists may be limited.