遥感影像语义变化检测在生态环境、土地利用、土地覆盖监测等方面发挥着重要作用。近年来基于深度学习的变化检测方法是遥感智能解译关注的热点，然而现有的三分支语义变化检测方法缺少对变化分支和语义分支一致性建模，导致双时相语义变化检测的自相矛盾。针对该问题，本文提出一种基于孪生CNN与Transformer的遥感影像语义变化检测算法。在编码阶段，首先设计孪生ResNet34网络提取影像的多尺度特征，并嵌入差异增强模块来提高变化信息的关注度；然后利用语义标记器将特征图映射为紧凑的语义Token，并通过Transformer编码器联合双时相语义和变化信息来建模“语义-变化”一致性。在解码阶段，通过Transformer解码器利用跳跃连接，将不同细粒度的语义信息融合，生成细化的语义特征图；经过上采样恢复、掩码相乘得到双时相语义变化结果。在遥感语义变化检测公开数据集SECOND上的实验结果表明，本文提出的算法能够有效地关注变化区域，保持变化结果和语义结果的一致性，且达到了优异的评价指标和视觉效果。

Abstract: Semantic change detection of remote sensing images plays an important role in ecological environment, land use, land cover monitoring and so on. In recent years, deep learning-based change detection methods are the hotspot of remote sensing intelligent interpretation concern, however, the existing three-branch semantic change detection methods lack of modeling the consistency of change branch and semantic branches, which leads to the self-contradiction of the bi-temporal semantic change detection. Method To address this problem, this paper proposes a remote sensing image semantic change detection algorithm based on Siamese CNN and Transformer. In the encoding stage, the Siamese ResNet34 network is firstly designed to extract the multi-scale features of the image, and the Difference Enhancement Module is embedded to increase the attention of the change information; then the semantic tokenizer is used to map the feature map into compact semantic tokens, and the Transformer encoder is used to combine the bi-temporal semantics and the change information to model the "semantics-change" consistency. In the decoding stage, different fine-grained semantic information is fused by Transformer decoder using hopping connection to generate a refined semantic feature map. Finally, the result of bi-temporal semantic change is obtained after up-sampling recovery and mask multiplication. Result The experimental results on the remote sensing semantic change detection public dataset SECOND and LandSat-SCD show that the algorithm proposed in this paper can effectively focus on the change region, maintain the consistency between the change results and the semantic results, and achieve excellent evaluation indexes and visual effects. Conclusion We can draw the following conclusions: (1) The proposed Difference Enhancement Module can enhance the difference characteristics of bi-temporal remote sensing images and improve the network"s focus on change information. (2) The proposed bi-temporal Transformer module maps the difference information and bi-temporal semantic information into semantic tokens and fuses them to jointly model the "semantic-change" information of the whole spatio-temporal domain in the token space, effectively modeling the long-range dependencies in the images and modeling the bi-temporal contextual correlations. The long-range dependency in the image is effectively modeled, and the bi-temporal contextual relevance is modeled. The ReTNet network designed accordingly pays more attention to the change area, and can accurately detect the change location and recognize the change element type of the bi-temporal remote sensing image.