The rapid development of spatially resolved transcriptomics (SRT) technologies has provided novel insights for elucidating the spatial architecture of tissue. However, many methods neglecting complementary information among different modalities within SRT datasets, thus limiting the accuracy of spatial domain identification. To address this issue, we propose Landviewer, an end-to-end multi-view graph learning method for spatial domain identification. Landviewer constructs adjacency relationships across gene expression, spatial locations, and histological images, and employs graph convolutional encoders combined with an attention mechanism to achieve automatic fusion of multimodal information. We additionally introduce a self-supervised Kullback-Leibler divergence to enhance the robustness of the Landviewer. This strategy strengthens the influence of high-confidence samples, thereby improving clustering accuracy and stability. Experimental evaluations on multi-platform datasets demonstrate that Landviewer can not only identify functional domains across various tissues, but also decipher region-associated molecular pathways and complex cell-cell interactions.