Glaucoma is a leading cause of irreversible vision loss worldwide, emphasizing the need for early and accurate detection. This study proposes a deep learning-based framework to classify retinal images as normal or glaucomatous. Nine state-of-the-art convolutional neural network (CNN) architectures, including EfficientNet variants, DenseNet201, ResNet50, XceptionV3, and MobileNetV3 Large, were evaluated. The dataset was preprocessed with resizing, normalization, and structured for efficient data loading. Hyperparameter optimization and fine-tuning techniques were employed to maximize performance. EfficientNetV2L demonstrated the best results, achieving a sensitivity of 98.78%, an F1 score of 92.05%, and an accuracy of 91.95%. Lightweight models, such as MobileNetV3 Large, also exhibited competitive performance, making them suitable for resource-constrained settings. Evaluation metrics, including sensitivity, specificity, precision, accuracy, and F1 score, validated the robustness of the proposed framework. The results highlight the potential of advanced CNNs for real-time and scalable glaucoma detection. Future work will focus on expanding datasets, integrating explainable AI (XAI) techniques, and optimizing models for clinical deployment. This framework aims to contribute to improving early diagnosis and reducing the global burden of glaucoma-induced blindness.