Jia et al. (2026) TCSNet: A Thin-Cloud-Sensitive Network for Hyperspectral Remote Sensing Images via Spectral-Spatial Feature Fusion

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-04-26
• Authors: Yuanyuan Jia, Siwei Zhao, Xuanbin Liu, Yinnian Liu
• DOI: 10.3390/rs18091326

Research Groups

Not specified in the provided text.

Short Summary

The paper introduces the Thin-Cloud-Sensitive Network (TCSNet), a dual-branch deep learning architecture designed to improve the detection of thin clouds in hyperspectral imagery by balancing spatial and spectral feature extraction.

Objective

• To overcome the limitations of existing cloud detection methods that prioritize spatial features over spectral features, thereby improving the detection accuracy of thin clouds and reducing spectral confusion with the underlying surface.

Study Configuration

• Spatial Scale: Pixel and image level (Hyperspectral imagery).
• Temporal Scale: Not specified.

Methodology and Data

• Models used: Thin-Cloud-Sensitive Network (TCSNet), featuring:
  • Dual-branch encoder: Convolutional Neural Network (CNN) for local features and PVTv2-B2 Transformer for long-range spectral dependencies.
  • Cross-Modal Fusion (CMF) module with a single-channel gate.
  • Squeeze-and-Excitation (SE) channel attention.
  • Multi-Scale Fusion (MSF) module for top-down feature integration.
  • Combined Attention Mechanism (CAM) in the decoder.
• Data sources: Gaofen-5 01 hyperspectral data.

Main Results

• TCSNet demonstrated superior performance in cloud detection, achieving:
  • Overall Recall: 92.98%
  • Thin-cloud Recall ($\text{Recall}{\text{thin}}$): 85.59%
  • Thick-cloud Recall ($\text{Recall}{\text{thick}}$): 99.75%

Contributions

• Developed a novel dual-branch architecture that effectively integrates multi-scale local spatial features with long-range spectral dependencies.
• Introduced specialized modules (CMF, MSF, and CAM) to refine feature recalibration and boundary detection, specifically addressing the challenge of thin-cloud sensitivity in hyperspectral remote sensing.

Funding

Not specified in the provided text.

Citation

@article{Jia2026TCSNet,
  author = {Jia, Yuanyuan and Zhao, Siwei and Liu, Xuanbin and Liu, Yinnian},
  title = {TCSNet: A Thin-Cloud-Sensitive Network for Hyperspectral Remote Sensing Images via Spectral-Spatial Feature Fusion},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18091326},
  url = {https://doi.org/10.3390/rs18091326}
}