TBSCD-Net: A Siamese Multitask Network Integrating Transformers and Boundary Regularization for Semantic Change Detection From VHR Satellite Images

Semantic change detection (SCD) from very high-resolution (VHR) images involves two key challenges: 1) the global features of bitemporal images tend to be extracted insufficiently, leading to imprecise land cover semantic classification results; and 2) the detected changed objects exhibit ambiguous boundaries, resulting in low geometric accuracy. To address these two issues, we propose an SCD method called TBSCD-Net based on a multitask learning framework to simultaneously identify different types of semantic changes and regularize change boundaries. First, we construct a hybrid encoder combining transformer and convolutional neural network (CNN) (TCEncoder) to enhance the extraction of global context information. A bitemporal semantic linkage module (Bi-SLM) is embedded into the TCEncoder to enhance the semantic correlations between bitemporal images. Second, we introduce a boundary-region joint extractor based on Laplacian operators (LOBRE) to regularize the changed objects. We evaluated the effectiveness of the proposed method using the SECOND dataset and a Fuzhou GF-2 SCD dataset (FZ-SCD) and compared it with seven existing methods. The proposed method performed better than the other evaluated methods as it achieved 24.42% separation kappa (Sek) and 20.18% global total classification error (GTC) on the SECOND dataset and 23.10% Sek and 23.15% GTC on the FZ-SCD dataset. The results of ablation studies on the FZ-SCD dataset also verified the effectiveness of the developed modules for SCD.