Glaucoma, a major cause of blindness, results from damage to the optic nerve due to increased intraocular pressure. Traditionally classified into normal or glaucoma categories, recent advancements have refined severity levels into mild (CDR 0.6–0.8), moderate (CDR 0.8–0.9), and severe (CDR > 0.9). However, manual Cup-to-Disc Ratio (CDR) calculations are expensive, require experienced clinicians, and often identify optic nerve damage only at advanced stages. This study aims to develop an automated system using Convolutional Neural Networks (CNN) to classify glaucoma severity into four detailed levels: normal, mild, moderate, and severe. The proposed method overcomes the limitations of manual CDR calculation by introducing a machine learning-based approach that provides a more detailed and accurate classification of glaucoma severity. Specifically, the study demonstrates the application of CNNs, with a focus on the VGG16 model architecture, to enhance glaucoma diagnosis. Utilizing the ORIGA dataset, the study involves pre-processing, optic disc and cup segmentation, feature extraction, and classification using the VGG16 model. The model’s performance was assessed through key metrics, including accuracy, precision, recall, specificity, and F1 Score. The developed model achieved an accuracy of 88.36%, precision of 91.23%, recall of 87.82%, specificity of 97.41%, and an F1 Score of 87.41%, indicating its effectiveness in detecting and classifying glaucoma based on fundus retinal images. This automated classification system presents a reliable and efficient alternative to traditional diagnostic methods, potentially enabling earlier detection and more effective treatment planning for glaucoma patients. Implementing this automated system in clinical settings could significantly enhance the early diagnosis and management of glaucoma, thereby reducing the risk of vision loss in patients.