Bi-Temporal Semantic Reasoning for the Semantic Change Detection in HR Remote Sensing Images

Semantic change detection (SCD) extends the multiclass (MCD) task to provide not only locations but also detailed land-cover/land-use (LCLU) categories before and after observation intervals. This fine-grained semantic information is very useful in many applications. Recent studies indicate that SCD can be modeled through a triple-branch convolutional neural network (CNN), which contains two temporal branches branch. However, this architecture, communications between branch are insufficient. To overcome limitations existing methods, we propose novel CNN architecture for SCD, where features merged deep CD unit. Furthermore, elaborate on reason bi-temporal correlations. The resulting reasoning (Bi-SRNet) types of blocks both single-temporal cross-temporal correlations, as well loss function improve consistency results. Experimental results benchmark dataset show proposed obtains significant accuracy improvements over approaches, while added designs Bi-SRNet further segmentation changed areas. codes article accessible at https://github.com/ggsDing/Bi-SRNet.