Over the last few years, there has been a rise in the prevalence of Diabetic Retinopathy (DR), one of the most severe consequences of diabetes that can cause lifelong blindness. Analysis of diabetic retinopathy (DR) using color fundus images requires experienced physicians to determine the presence of DR. In order to detect issues in DR and its numerous stages utilizing retinal scans, many different artificial intelligence and deep learning techniques have been developed. The present study aims to develop a robust system for the automated identification and categorization of diabetic retinopathy (DR). A Convolutional Neural Network (CNN) model called Augmented Efficient Net is used to detect diabetic retinopathy from fundus images in this work. The classification of diabetic retinopathy was performed using four different Efficient Net models, namely Efficient Net B0, Efficient Net B1, EfficientNetB3 and EfficientNet-B4 and tested on publicly available fundus image datasets APTOS 2019. Each model performance was assessed, and the best model was found among the four models. We use these models to solve the difficulties we have in transfer learning along with preprocessing. Data augmentation and hyper parameter optimization to make our models stronger and more accurate. Experiments showed that on the APTOS 2019 datasets, The training accuracy for EfficientNet-B4 was 99.03%, while the validation and testing accuracy were 87% respectively.