Glaucoma is an optic neuropathy that progressively develops, and it is one of the major causes of irreversible blindness all over the world. Diagnosis at its early stage is vital in avoiding vision blindness, but the traditional way of diagnosing the disease through an examination of the fundus of the retina is usually random and arbitrary. This paper presents a deep learning framework for the early-stage glaucoma (Retinal Glaucoma (RG) versus Non-Retinal Glaucoma (NRG) classification. An optic disc and optic cup: CNN model, which extracts the discriminative feature with emphasis on optic disc, optic cup, and the peripapillary region, is developed with retinal fundus images. To enhance image quality and bring out clinically relevant structures, preprocessing measures, such as enhancement of contrast, noise removal and vessel segmentation, are made. Data augmentation methods are utilized to check the unbalance in classes and improve the generalization of a model. On a curated dataset, a model is trained and validated, and model performance is measured using accuracy, sensitivity, specificity, F1-score. Experimental findings present a high accuracy in the classification of the early-stage RG with increased sensitivity rates of RG detection, indicating that the proposed approach may become an efficient decision-support system for ophthalmologists. The results denote that automated deep learning-based glaucoma screening can be useful in scalable and low-cost deployment in clinical practice and early identification of high-risk groups.