Yu Zhang

Multi-temporal optical remote sensing imagery provides complementary observations for cloud-free reconstruction, but effective cloud-awareness fusion remains challenging due to spatially heterogeneous and time-varying cloud contamination. In the typical three-temporal setting, naive fusion strategies such as concatenation or averaging may introduce cloud-corrupted information into the reconstruction, hindering robust temporal integration. To address this, we propose TemporalMambaFusionNet (TMFNet), an implicitly cloud-aware framework that performs pixel-wise selective fusion for three-temporal cloud removal. Specifically, a Mamba-inspired selective scan with a learnable gated update is designed to implement a “write/forget” mechanism, which prioritizes cloud-free signals within the temporal sequence while suppressing cloud-contaminated ones. Furthermore, a cloud-aware head is introduced at each timestamp to guide feature selection and confidence-normalized aggregation, ensuring robustness against diverse cloud and shadow conditions. Experimental results on Sen2 MTC Old and Sen2 MTC New demonstrate that TMFNet consistently outperforms representative baselines in terms of structural fidelity and spectral consistency, while also providing task-oriented suppression maps that reflect per-timestamp contribution during temporal fusion. Code and pretrained models are publicly available at https: //github.com/Zhangyu-student/TMFNet.

With advances in aerospace technology, high-resolution optical imagery from satellites like China's Gaofen series is vital for climate monitoring and agriculture. However, cloud cover significantly impedes the utility of optical remote sensing imagery, making effective cloud detection essential. Currently, Gaofen optical data cloud detection faces two difficulties: one is that the variable shapes and scales of clouds and their shadows pose difficulties for segmentation tasks, and the other is the scarcity of sufficient and well-annotated datasets. This work addresses these challenges through two main contributions. First, we propose the MFCSD(MaskFormer-based Cloud and Shadow Detection) model, an extended MaskFormer architecture that innovatively integrates an ASPP module for multi-scale feature extraction and a channel attention mechanism to enhance feature representation. More distinctively, based on the geometric properties of clouds and its shadow, we introduce a novel Shadow Shift Loss that explicitly quantifies and corrects the positional shift between clouds and their shadows, thereby significantly improving the classification accuracy of cloud shadows in remote sensing imagery. Second, we construct a manually annotated cloud detection dataset from Gaofen imagery, covering diverse geographic and climatic conditions across China to serve as a comprehensive benchmark. Extensive experiments demonstrate that our model achieves excellent performance, outperforming previous leading methods including SwinTransformer by 0.06 in mIoU and 0.05 in fwIoU while maintaining lower computational complexity. By integrating these innovations, this work provides an effective solution and valuable data resource for advancing cloud detection in remote sensing.

Cloud contamination in multi-temporal remote sensing images, particularly due to thick clouds, significantly impacts the accuracy of environmental monitoring tasks. While existing Low-Rank Tensor Completion (LRTC) methods address cloud removal, they struggle with strict multi-temporal correlation requirements and poor generalization due to parameter sensitivity. To overcome these limitations, we propose WLRT-DR, a novel two-stage framework combining Wavelet-domain Low-Rank Tensor Completion with Deep Regularization. First, we introduce discrete wavelet transform (DWT) to decompose multi-temporal imagery into frequency-specific subbands (LL, LH, HL, HH). Empirical analysis reveals distinct low-rank properties in each subband, enabling targeted low-rank processing that better preserves spatial textures and structural continuity compared to unified spatial-domain LRTC. Second, we develop a dual-phase architecture where wavelet-domain LRTC (Stage 1) is synergistically combined with deep network regularization (Stage 2). This hybrid strategy leverages the complementary strengths: wavelet-domain low-rank recovery efficiently reconstructs fundamental spatial features, while pretrained deep networks impose non-linear constraints to enhance spectral fidelity. Comprehensive validation on simulated (Landsat data covering diverse Beijing landscapes) and real-world data demonstrates that WLRT-DR delivers enhanced reconstruction accuracy, robustness to complex spatio-temporal patterns, and improved processing efficiency. The framework provides an effective alternative for cloud reconstruction by integrating traditional tensor completion and deep learning in a novel manner.

Optical satellite imagery has become indispensable for Earth observation, yet thick cloud contamination severely impedes continuous monitoring capabilities. Current temporal-based methodologies present dual limitations: conventional approaches suffer from inefficient processing speeds and poor robustness, while existing deep learning-based solutions exhibit constrained temporal adaptability by relying on limited temporal information from single or fixed-period images. These inherent shortcomings consequently hinder their scalability in large-scale real-world applications. To address this challenge, we propose a Multi-temporal Adaptive Cloud Removal Transformer (MACRT), which uses a temporal aggregation strategy to integrate information from multiple reference images, reduce the constraint of reference image selection, and provide a more practical cloud removal solution. Since neighboring pixels of remote sensing images have strong correlations, the proposed network architecture introduces a novel spatio-spectral multi-head attention (SSMHA) module. This module decomposes the attention computation into two branches: coarse-resolution spatial and spectral, which work together to capture long-range dependencies across both dimensions while maintaining a compact parameter count. Extensive experiments on multispectral datasets (Landsat and Sentinel-2) demonstrate the superior performance of MACRT. Quantitative evaluations on Landsat imagery reveal improvements of 1.7 % in SSIM and 2.1° in SAM over state-of-the-art methods, achieving enhanced structural preservation and spectral accuracy. The source code, pretrained models, and all experimental results will be publicly available at https://github.com/Zhangyu-student/ACRT upon the paper’s acceptance.