The accurate extraction of pertinent information from fundus images is of paramount importance for the diagnosis of glaucoma. For a considerable period, researchers engaged in this field have typically employed the convolutional neural network approach for detection. Despite notable advancements, the inability of convolutional neural networks to capture long-range dependencies in order to make a judgement in the fundus images as a whole represents a current challenge. Conversely, MLP-based models are gaining widespread attention due to their simple network structure and performance that is not weaker than CNNs and transformers. In this paper, we propose a novel MLP-based approach called Dynamixing Gate and Depthwise Group Norm Network (DG2Net), which uses a convolutional neural network to extract local information while using an MLP-like model to obtain global information. Specifically, we propose the Dynamixing Gate MLP for enhancing the extraction of spatial information and the Depthwise Group Norm MLP to enhance the model’s ability to extract channel information. Extensive experiments were conducted on two publicly available glaucoma datasets, EyePACS AIROGS-Light and EyePACS-AIROGS-light-V2. The results demonstrated that our method outperforms other existing methods, and that DG2Net is highly competitive for glaucoma detection.