Diabetic retinopathy(DR), a disease that affects the eye by causing damage to the blood vessels of the light sensitive retina. Diabetic retinopathy at the earlier stage might cause no or very minimal vision issues but if not cured it might lead to loss of vision as well. Diabetic retinopathy, particularly in people of working age, is the main cause of blindness among diabetic patients in developing countries. Normal, mild, moderate, severe, and PDR (Proliferative Diabetic Retinopathy) are the five stages of diabetic retinopathy.. Usually, trained experts look into the colored fundus image of the eye to examine the stage of DR. But this process is manual and is highly time consuming. It could also sometimes lead to error-prone results. Therefore, various Computer Vision approaches are proposed to detect the stage of DR in the fundus image, but they lack in particular classifying the DR at lower stages. The proposed methodology compares various CNN models (Densenet201, InceptionV3, Resnet50) along with processing specifically to extract maximum features from fundus image to classify the stage of DR with the publicly available kaggle dataset. The proposed methodology compares the algorithms based on Quadratic Kappa Score where the obtained score for Densenet201, InceptionV3 and Resnet50 are 0.931,0.956 and 0.926 respectively.