RTNet: Remote sensing image change detection combining RepVGG and transformer

During the process of partitioning the input image into blocks, existing visual transformers damage the internal structural information of the image blocks, and the long-range attention mechanism easily ignores the local unique properties of the image, resulting in inferior performance to traditional convolutional networks. This study proposes a new neural network RTnet, which combines RepVGG network with Transformer to fully utilize the advantages of RepVGG network, the advantages of RepVGG network in feature extraction, while also introducing Transformer's self-attention mechanism to further capture cross spatial dependencies and improve the accuracy of change detection to a certain extent. Through experiments on the LEVRO-CD and DSIFN-CD datasets, it was through experiments on the LEVRO-CD and DSIFN-CD datasets, it was shown that the RTnet network achieved better Change Detection (CD) performance compared to previous similar architectures, and the accuracy of change detection increased by 1.41%. detection increased by 1.41%.