Glaucoma is a prominent contributor to vision loss, and the timely identification of this condition is vital for effective intervention. Retinal fundus images have gained popularity in glaucoma detection due to their non-invasive nature. Nevertheless, the manual scrutiny of these images is labour-intensive and susceptible to human errors. Recent years have witnessed remarkable advancements in deep learning, particularly in various medical imaging tasks, including glaucoma detection. We have introduced an innovative, tailored approach involving convolutional neural networks (CNNs) enriched with deep-learned features to advance the current technological landscape. This paper introduces five customized CNN models combined with different pre-trained deep learning architectures. Our experimental work was validated using a publicly accessible retinal fundus image dataset. Among these five models, InceptionV3+VGG16 outperformed the rest, achieving an impressive accuracy 98.67%, recall 98.22%, precision 99.10% and F1-score of 98.66%. Additionally, we employed the Adam optimizer and assessed the model’s performance using a range of evaluation metrics, including accuracy, precision, recall, and F1 score. This study underscores the potential of deep learning-based methodologies in the early detection of glaucoma, opening doors for more precise and efficient diagnosis and treatment.