SIGNet: A Siamese Graph Convolutional Network for Multi-Class Urban Change Detection

Detecting changes in urban areas presents many challenges, including complex features, fast-changing rates, and human-induced interference. At present, most of the research on change detection has focused traditional binary (BCD), which becomes increasingly unsuitable for diverse tasks as cities grow. Previous networks often rely convolutional operations, struggle to capture global contextual information underutilize category semantic information. In this paper, we propose SIGNet, a Siamese graph network, solve above problems improve accuracy multi-class (MCD) tasks. After maximizing fusion differences at different scales using joint pyramidal upsampling (JPU), SIGNet uses convolution-based reasoning (GR) method construct static connections features space cross-attention couple dynamic types during process. Experimental results show that achieves state-of-the-art MCD datasets when capturing relationships between regions correlations categories. There are currently few pixel-level domain. We introduce new well-labeled dataset, CNAM-CD, is large dataset containing 2508 pairs high-resolution images.