Glaucoma is a serious eye disease that can cause irreversible vision loss, making early diagnosis and treatment essential. In this study, we propose a novel hybrid deep learning-based method for identifying glaucoma in retinal images. Our approach involves training a hybrid neural network using a modified VGG16 architecture based on a CNN-modified DenseNet architecture, optimizing with an improved Adam optimization algorithm, and applying image denoising with both adaptive and non-adaptive thresholding. To smooth out the retinal images, we use anisotropic diffusion filtering instead of the conventional median filter method. For segmentation of retinal images, we employ a cascading UNet architecture, where each iteration of segmentation builds upon the results of the previous iteration, leading to more precise glaucoma diagnosis, even at the early stages of the disease. To evaluate the performance of our suggested technique, we conducted extensive trials using a publicly available collection of retinal images. Our results indicate that our proposed approach outperforms the state-of-the-art in glaucoma detection, achieving higher accuracy, sensitivity, and specificity. Our approach has the potential to significantly improve early glaucoma detection and treatment, thereby reducing the risk of blindness.