A Dual-Branch Framework for Semantic Change Detection with Boundary and Temporal Awareness
Yun-Cheng Li, Sen Lei, Heng-Chao Li, Ke Li
TL;DR
Semantic Change Detection (SCD) in bi-temporal remote sensing suffers from blurred boundaries and weak temporal modeling. The authors propose DBTANet, a dual-branch Siamese architecture that fuses a frozen Segment Anything Model (SAM) for global semantics and boundary priors with a lightweight ResNet34 for local detail. Two key innovations are introduced: the Gaussian-smoothed Projection Module (GSPM) to purify shallow SAM features and the Bidirectional Temporal Awareness Module (BTAM) to capture multi-scale temporal dependencies, with feature fusion governed by learnable gates $F_{shallow} = (1-α) F^{res}_{shallow} + α GSPM(F^{SAM}_{shallow})$ and $F_{deep} = (1-β) F^{res}_{deep} + β F^{SAM}_{deep}$. Empirical results on Landsat-SCD and SECOND show state-of-the-art performance, validating the effectiveness of combining global priors, boundary cues, temporal reasoning, and boundary-aware constraints for robust SCD.
Abstract
Semantic Change Detection (SCD) aims to detect and categorize land-cover changes from bi-temporal remote sensing images. Existing methods often suffer from blurred boundaries and inadequate temporal modeling, limiting segmentation accuracy. To address these issues, we propose a Dual-Branch Framework for Semantic Change Detection with Boundary and Temporal Awareness, termed DBTANet. Specifically, we utilize a dual-branch Siamese encoder where a frozen SAM branch captures global semantic context and boundary priors, while a ResNet34 branch provides local spatial details, ensuring complementary feature representations. On this basis, we design a Bidirectional Temporal Awareness Module (BTAM) to aggregate multi-scale features and capture temporal dependencies in a symmetric manner. Furthermore, a Gaussian-smoothed Projection Module (GSPM) refines shallow SAM features, suppressing noise while enhancing edge information for boundary-aware constraints. Extensive experiments on two public benchmarks demonstrate that DBTANet effectively integrates global semantics, local details, temporal reasoning, and boundary awareness, achieving state-of-the-art performance.