Abstract : Retinal fundus imaging plays a vital role in the early diagnosis of retinal diseases such as glaucoma, diabetic retinopathy, and cataracts. However, image quality is often compromised by specular reflections caused by flash photography, leading to reduced diagnostic accuracy. This study proposes a hybrid model for glaucoma detection that combines a preprocessing stage to remove specular reflections with an advanced segmentation phase utilizing a modified Convolutional Neural Network (CNN) architecture, specifically based on AlexNet. The preprocessing phase extracts diffuse and specular components using seven methods, and the best quality output is passed through a high-emphasis filter. The segmented images are then processed in the feature extraction phase, followed by classification using a Support Vector Machine (SVM) with multiple kernel options. This model aims to improve the accuracy of glaucoma detection, overcoming limitations posed by image artifacts and providing a robust solution for retinal disease screening. The hybrid CNN-AlexNet model enhances detection capabilities, ensuring better image quality and higher classification accuracy, making it an effective tool for early glaucoma diagnosis. Experimental results demonstrate that the proposed model achieves significant improvements in preprocessing, the performance parameters are 0.961, with the glaucoma detection phase shows an accuracy of 94.83%, sensitivity of 94.19%, specificity of 96.37%, and an area under the receiver operating characteristic curve (AUROC) of 0.99.1.