Glaucoma is a chronic neurological disease in the human eye where there is damage to the nerves which causes vision loss to blindness. Glaucoma can be detected by classifying retinal images. Several previous studies that classified glaucoma did not perform segmentation beforehand. Segmentation is needed to extract the features of the optic disc and optic cup from retinal images that are used to detect glaucoma. This study proposes two stages in the detection of glaucoma, namely the segmentation and classification stages. Segmentation is carried out using the U-Net architecture. Classification is done using a new architecture, namely Xcep-Dense. The Xcep-Dense architecture is a new architecture which is the result of a combination of the Xception and DenseNet architectures. At the segmentation stage, accuracy, recall, precision, and F1-score values are obtained above 90%. The Cohen’s kappa value has a value above 85% and loss below 20%. At the classification stage, accuracy and specification values were obtained above 85%, sensitivity and F1-score above 80%, and Cohen’s kappa above 70%. The predicted image obtained at the segmentation stage has a very similar appearance to the ground truth. Based on the results of the performance evaluation obtained, it shows that the method proposed in this study is feasible in detecting glaucoma.Glaucoma,