13 August 2025 in IEEE Transactions on Geoscience and Remote Sensing
Spatiotemporal fusion (STF) plays a critical role in remote sensing and Earth observation by integrating high spatial resolution details with high temporal consistency, serving as an essential tool for environmental monitoring and postdisaster assessment. Recently, numerous STF methods, especially deep learning-based approaches, have been proposed and shown promising results. However, most of them struggle with effectively capturing complex spatiotemporal interactions as well as balancing computational efficiency and interpretability. To address these challenges, we propose an innovative STF framework, frequency-selected differential fusion Transformer (FSDFormer), which comprehensively processes multiscale spatiotemporal data while maintaining linear computational complexity. FSDFormer integrates three key components: the spatial enhanced agent Transformer (SEAT) block, the frequency-adaptive filtering network (FAFN), and the spatiotemporal differential fusion (SDF) block. The SEAT block employs a novel attention mechanism to reduce computational complexity, while the FAFN enhances feature extraction by introducing frequency-domain filtering in the forward network. Grounded in a spatiotemporal dependency-aware fusion strategy, the SDF block leverages a two-stage mechanism to excavate and fuse temporal and spatial information sequentially. Extensive experiments show that FSDFormer delivers superior performance on three STF datasets, outperforming state-of-the-art (SOTA) methods in accuracy, robustness, and generalizability. The code is available at https://github.com/Sichen-Lu/FSDFormer
21 September 2026 in International Journal of Applied Earth Observation and Geoinformation
2 March 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Spatiotemporal fusion (STF) methods are pivotal for generating satellite imagery with high spatial and temporal resolution, enabling continuous, gap-free monitoring of land surface dynamics. While deep learning-based STF has shown remarkable potential, it often struggles to balance computational efficiency with effective long-range dependency modeling and adaptive feature extraction across heterogeneous landscapes. To address these issues, we introduce MambaSTF, a novel STF framework based on visual state space models, capable of capturing long-range contexts with linear complexity. Its core includes a Quad-directional 2D-Selective-Scan (QD-SS2D) mechanism that systematically models anisotropic spatial structures via multidirectional scanning. The dedicated State Space Fusion (SSF) module employs dual orthogonal streams: Spatial Selective-Scan Fusion (S-SSF) interleaves multi-resolution features to enhance context, while Temporal Selective-Scan Fusion (T-SSF) aligns multi-temporal sequences to capture dynamics. These are adaptively fused through a gating mechanism, enabling context-aware spatiotemporal integration. Extensive experiments across diverse datasets confirm its superiority in reconstructing complex landscapes under varying atmospheric conditions. Rigorous uncertainty quantification further demonstrates operational stability and robustness, even amid abrupt land cover changes. The code is available at https://github.com/Sichen-Lu/MambaSTF.