Abstract: The rapid development of Spatial Transcriptomics (ST) has enabled the simultaneous acquisition of gene expression profiles and spatial locations from tissue sections, facilitating the characterization of tissue microenvironments and heterogeneity. Spatial domain identification is a core task in ST analysis. However, existing methods still underutilize the joint exploitation of spatial structure and biological functional signals when handling high-dimensional and sparse data, which limits their interpretable ability in delineating complex tissue regions. To address this, we propose TF-GCN, a multi-view graph convolutional network that integrates transcriptional regulatory information. Specifically, we construct a regulatory activity view using SCENIC and jointly model it with a spatial adjacency view, achieving adaptive multi-view integration via co-convolution alignment and attention-based fusion. Experimental results on public datasets such as DLPFC and BRCA demonstrate that TF-GCN outperforms baseline methods in clustering accuracy and robustness. Moreover, we further validate its generalizability on high-resolution STARmap data and an unlabeled breast cancer dataset. Ablation studies and attention-weight analyses indicate that co-convolution, attention fusion, and regularization each contribute substantially to performance improvements, and that the learned weights adapt to tissue heterogeneity. Overall, TF-GCN more effectively fuses spatial information with transcriptional regulatory signals, improving the characterization of complex tissue architectures.