Graph Neural Networks (GNNs) excel at semi-supervised learning by iteratively aggregating node features with neighborhood information. However, the majority of existing GNNs concentrate on either local or global augmentation, while effectively integrating both local and global node information remains a significant challenge. To address the challenge, this paper proposes a Decoupled Graph Neural Network with Hybrid Data Augmentation (HDANET). First, HDANET proposes a hybrid data augmentation strategy that combines both local and global augmentations. This strategy learns the feature distribution of neighboring nodes based on the globally augmented node features, thereby generating locally augmented node features and improving the expressive power of GNNs. Second, a multi-scale information mixed order propagation is proposed. It decouples linear transformations and propagation operations in graph convolution, and integrates multiple order node representations to enable enhanced graph node representations to be learned in a larger receptive field. Finally, a regularization term is incorporated to enhance the model’s generalization ability. Experimental results demonstrate that HDANET achieves superior performance in semi-supervised node classification tasks, exhibiting both robustness and excellent generalization.