Diabetic retinopathy is one of the major ocular diseases that has the potential to cause vision loss and is ranked as one of the most common complications associated with diabetes. However, it can be halted if identified early and intervened on quickly. This study looks into the application of advanced deep learning techniques for the diagnosis and classification of diabetic retinopathy at its five different stages by using retinal images that portray a spectrum of mild to severe diseases. We evaluated many prominent Convolutional Neural Network (CNN) models, including DenseNet, Inception, MobileNet, VGG-19, and ResNet. These models were meticulously calibrated for the task and performed quite well. To guarantee optimal model performance in the presence of imbalanced data and to mitigate overfitting, we implemented diverse data augmentation strategies, including rotation, flipping, and zooming of photos throughout the training process. This study concentrates on the automation of retinal fundus image processing, a crucial instrument for the diagnosis and monitoring of diabetic retinopathy. We demonstrate that deep learning models can assist healthcare practitioners in the accurate and swift detection and staging of diabetic retinopathy (DR). This may lead to faster clinical decisions and timely treatments, thus protecting vision and improving the quality of life for individuals with diabetes.