Blindness can result from diabetic retinopathy (DR), a serious side effect of diabetes mellitus that requires prompt diagnosis and treatment. Using deep learning techniques to classify retinal fundus images (RFI) is a potential method for the early diagnosis of diabetic retinopathy (DR) in patients. The use of advanced deep learning techniques, including deep belief networks and DMLOPs, to transfer learning models is investigated in this work in order to classify RFIs and identify early indicators of DR. The suggested approach seeks to achieve high accuracy in differentiating between various phases of DR and healthy retinal states by utilizing a dataset of labeled retinal fundus pictures. The approach incorporates data augmentation and preprocessing techniques to enhance model robustness and generalization. Deep learning models are evaluated based on performance criteria like specificity, sensitivity, and accuracy to determine how effective they are. According to the results, these methods might greatly increase the early identification of DR, which may facilitate prompt intervention and improved management of the ocular health of diabetic patients.