Glaucoma, a progressive neuropathy of the eye, can lead to permanent blindness if not identified earlier. Applying deep learning (DL) models for automated glaucoma identification typically requires hundreds of images. However, in practice, it’s not possible to have such a large number of well-defined glaucoma images, thus requiring a small set of images to be augmented to generate a much larger set. In this work, a GAN model, namely, Deep Convolutional Generative Adversarial Network (DCGAN) has been used to produce a sufficiently large number of fundus images to train a Deep Convolutional Neural Network (DCNN) for the detection of glaucoma. Further, an additional augmented dataset has also been created using conventional image transformation techniques such as flipping, rotation, etc. Both datasets have been trained using a deep CNN model, visualizing the features extracted in the intermediate layers. Experimental results prove that the DCGAN approach for image augmentation and training by deep CNN model produced an improved accuracy of 94.34% in glaucoma detection with interpretable feature visualization.