Diabetic Retinopathy (DR) is a common eye condition that occurs in people with Diabetes Mellitus. If this illness is not treated, there is a greater chance of blindness, either entire or partial. DR has broad range of levels and to identify and categories these levels in its early stages are challenging so it is important to have an automated system that can identify DR timely to prevent this disease from developing, but automated DR detection system has its own short comings such as limited data, small dataset and distortions in images due to size, colour and others. To overcome this problem, we proposed ResNET50 model with customized classification block to accurately identify DR and categories its levels. During the experiment we used two Datasets i.e. DR benchmark dataset and APTOS-19 and we validate the effectiveness and reliability of our proposed approach and the results of our experiment demonstrate that our proposed model generated better results to accurately identify DR and its states.

Index Terms- Image Classification, Diabetic Retinopathy, Resnet50, Fundus Images and Deep Learning