 This paper proposes a novel approach called Signet to address the limitations of existing change detection methods. It combines the advantages of graph convolution and graph cross-attention to capture contextual relationships between different regions and semantic correlations between different categories. The proposed model outperforms other state-of-the-art models on various MCD datasets, demonstrating its effectiveness in detecting changes in urban areas. Additionally, it is also applied to a large MCD dataset with high-resolution images, achieving promising results. This article was authored by Yen Pingzho, Jin Jiewang, Jin Li Ding, and others.