Diabetic Retinopathy is a diabetes complication that affects eye. It is an ophthalmic disease that reparations retinal blood vessels.Diabetic retinopathy happens due to the presence of huge amount of glucose in the blood vessels affects the retinal microvasculature. Pre-emptive symptoms of diabetic retinopathy are helpful for identifying the vision loss. There are several stages in predicting diabetic retinopathy. The stages are normal, mild, moderate, severe and proliferative diabetic retinopathy. The ophthalmologists observe the patient fundus images to diagnose the fatal disease and is found to be in error. The computer vision methods are proposed to detect diabetic retinopathy stages. However, these methods are not able to encode the complex macular edema feature and classify DR stages in a very low accuracy. In this paper,hundred and one deep convolutional neural network ResNET 101 model is proposed to encode the macular edemafeatureand improve the classification for all the five stages of DR. The training set is 413 (80%) and testing set is 103 (20%) is considered for analysis.The proposed experimental results show that detects different stages of diabetic retinopathy and performs better when compared to the existing techniques. ResNET 101 deep convolutional neural network is implemented, tested and the accuracy is compared with the ResNET 50 algorithm.