Graph Transformer (GT) models have demonstrated superior performance in various graph-related tasks due to their ability to model long-range node dependencies. However, current GT-based methods are constrained by low-dimensional node attributes and insufficient exploitation of global graph structure, leaving their feature embedding space underexplored. In this study, we propose a Feature Expansion based Graph Transformer (FEGT) for effective graph data processing, aiming to improve node classification accuracy. FEGT integrates node, edge, and structural embeddings to extend the original feature space of Graph Transformers, employing a fusion architecture that combines message-passing mechanisms and Transformer modules for node and edge feature extraction. Importantly, we capture structural information by decomposing the graph adjacency matrix and extracting principal component features, which are incorporated into the feature space as structural embeddings. We present numerical experiments on benchmark graph datasets (Cora, Citeseer, Pubmed, PATTERN, CLUSTER) to verify the effectiveness of FEGT. The results demonstrate that FEGT achieves superior performance in node classification compared to seven competitive models (GCN, GIN, GAT, GatedGCN, SAN, GT, GPS). Ablation studies further validate the effectiveness of the feature expansion with structural embeddings.