Li et al. (2025) A Class-Aware Unsupervised Domain Adaptation Framework for Cross-Continental Crop Classification with Sentinel-2 Time Series

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-19
• Authors: Shuang Li, Li Liu, J. Huo, Shengyang Li, Yue Yin, Yonggang Ma
• DOI: 10.3390/rs17223762

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study proposes PLCM, an unsupervised domain adaptation framework, to overcome domain shift challenges in cross-continental crop classification using Sentinel-2 satellite time series, achieving robust and balanced high-accuracy mapping, particularly for difficult-to-adapt crop categories.

Objective

• To develop an unsupervised domain adaptation (UDA) framework for robust and accurate large-scale crop classification across diverse geographical regions, mitigating domain shift challenges in cross-continental scenarios using Sentinel-2 satellite image time series.

Study Configuration

• Spatial Scale: Regional to continental scale, specifically transferring a model trained in the United States to Wensu County, Xinjiang, China.
• Temporal Scale: Multi-temporal observations from Sentinel-2 satellite image time series, spanning data from 2022 (source domain) and 2024 (target domain).

Methodology and Data

• Models used: PLCM (PSAE-LTAE + Class-aware MMD), which includes a Pixel-Set Attention Encoder (PSAE) and a class-aware Maximum Mean Discrepancy (MMD) loss function.
• Data sources: Sentinel-2 satellite image time series.

Main Results

• The PLCM framework demonstrated robust performance, achieving a competitive overall Macro F1-score of 96.56%.
• It provided a more balanced performance across crop categories, particularly excelling at identifying difficult-to-adapt categories (e.g., Cotton).
• Ablation studies confirmed that both the PSAE module and the class-aware MMD strategy were critical for the observed performance gains.
• The framework effectively learned domain-invariant and class-discriminative features.

Contributions

• Proposed PLCM, an unsupervised domain adaptation (UDA) framework specifically designed for robust crop classification using satellite image time series across diverse geographical regions.
• Introduced a Pixel-Set Attention Encoder (PSAE) to intelligently aggregate spatial features within parcels, enhancing robustness against noise and intra-parcel heterogeneity.
• Developed a class-aware Maximum Mean Discrepancy (MMD) loss function for fine-grained feature alignment within each crop category, mitigating negative transfer while preserving class-discriminative information.
• Demonstrated a practical and effective solution for high-accuracy, large-scale crop mapping in challenging cross-continental, cross-year scenarios.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Li2025ClassAware,
  author = {Li, Shuang and Liu, Li and Huo, J. and Li, Shengyang and Yin, Yue and Ma, Yonggang},
  title = {A Class-Aware Unsupervised Domain Adaptation Framework for Cross-Continental Crop Classification with Sentinel-2 Time Series},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17223762},
  url = {https://doi.org/10.3390/rs17223762}
}