The retinal malfunctioning caused during diabetes is popularly known as Diabetes retinopathy (DR) which is a severe disease of the retina and is emerging as a threat cause of less sightedness in the world. Eye using optical coherence tomography (OCT), and photographs of the fundus and for appraise of the thickness and the structure of the retina, furthermore for the discernment of edema, bleeding, and scarring. Deep learning models are particularly used for the analysis of OCT images, or fundus images, extraction of features that are unique to each period of study. and, in ramification, the classification of the images, and the stage of the disease. During this work, it is deep the neural network (CNN), with 24 convolutional layers and 4 fully-connected layers for the analysis of fundus images, and to distinguish between the control, moderate degree (i.e., a coalescence of mild and moderate Non-proliferative (D (NPDR) and severe degree (i.e., a group of severe NPDR, and keeps the DR (PDR)) the validation of the precision and accuracy of 88% -89%, a sensitivity of 87% -89%, and a specificity of 94% to 95% and the quadratic weighted Kappa scores of 0.91–0.92, with a 5-fold and 10-fold cross-validation methods, respectively. The last preprocessing phase has been carried out for both the class-specific image size and the data on the growth rate were to be applied. The proposed approach is significantly more accurate in the objective diagnosis and classification of diabetic retinopathy, which eliminates the need for retinal imaging and extends access to the retina, or treatment. This technology allows for early diagnosis, as well as the purpose of the monitoring of the progression of the disease, which may help in the optimization of medical therapy in order to denigrate the loss of visual function.