Answers are grounded in this paper and cite their sources.
Remote sensing image change detection aims to identify surface changes across bi-temporal images. However, existing methods still suffer from insufficient bi-temporal feature interaction, inadequate multi-scale fusion, and lost boundary details in complex scenes, which lead to false alarms and missed detections. To address these issues, we propose GCT-Net, an end-to-end change detection network comprising three key modules. First, a Spatial-Temporal Refinement Module (STRM) models global bi-temporal interactions using global context aggregation and Transformer-driven attention. Second, a Gated Bidirectional Feature Pyramid Network (G-BiFPN) performs adaptive multi-scale fusion and suppresses background noise via attention gating. Third, a Coordinated Difference Enhancement Module (CDEM) refines difference features and boundaries through nonlinear difference learning and coordinate attention. We evaluate GCT-Net on three public datasets: LEVIR-CD, DSIFN-CD, and WHU-CD. GCT-Net achieves highly competitive performance and a favorable accuracy-efficiency trade-off. Specifically, it achieves an F1-score of 93.78% on DSIFN-CD (surpassing ChangeFormer by 7.11 percentage points) and the highest IoU of 85.32% on WHU-CD, while maintaining competitive results on LEVIR-CD. Ablation studies verify the complementary contributions of STRM, G-BiFPN, and CDEM. Meanwhile, GCT-Net requires only 21.25 M parameters and 22.14 GFLOPs. These results confirm the effectiveness and robustness of GCT-Net in complex remote sensing scenarios.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: