Diabetic retinopathy (DR) is a serious complication of diabetes that can lead to blindness if not detected and treated early. Conventional screening methods involve fundus examination by trained medical personnel, which is time-consuming and costly. This study proposes an automated detection approach for diabetic retinopathy using digital fundus images and Convolutional Neural Network (CNN) methods. CNN, a deep learning architecture, is utilized to automatically learn and extract features from retinal fundus images. The dataset used for detection and classification consists of 5 classes: mild, moderate, no_DR, proliferative, and severe. The image training process employs the VGG-19 model trained for 100 epochs, achieving a commendable accuracy of 72% with a dataset of 3000 fundus images split into a 70:30 ratio for training and validation (70% for training, 30% for validation). The diagnosis results include 2160 images classified as DR and 840 images classified as NDR. Training with an 80:20 data split (80% for training, 20% for validation) yielded an accuracy of 69%, with 2070 images diagnosed as DR and 930 images as NDR.