Lu et al. (2025) Uncertainty Mixture of Experts Model for Long Tail Crop Type Mapping

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-18
• Authors: Q. Richard Lu, Wenzhi Zhao, Jiage Chen, Xuehong Chen, Liqiang Zhang
• DOI: 10.3390/rs17223752

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This paper proposes the Difficulty-based Mixture of Experts Vision Transformer (DMoE-ViT) framework to address challenges in global crop type mapping, specifically intra-class variability and imbalanced training samples, achieving superior classification accuracy and robustness in complex agricultural environments.

Objective

• To develop a robust crop classification framework that mitigates the intra-class long tail distribution problem caused by intra-class variability and imbalanced training samples, thereby enhancing classification accuracy in complex agricultural environments.

Study Configuration

• Spatial Scale: Global (intended application), with evaluation conducted in "Study Area 1" (regional/local scale).
• Temporal Scale: Not explicitly mentioned in the provided text.

Methodology and Data

• Models used: Difficulty-based Mixture of Experts Vision Transformer (DMoE-ViT) framework, which incorporates stratified sample partitioning, a multi-expert mechanism, and uncertainty quantification. Baseline deep learning models were used for comparison.
• Data sources: Not explicitly mentioned in the provided text, but implies a dataset of crop type samples used for training and evaluation.

Main Results

• The DMoE-ViT framework outperforms baseline deep learning models.
• Achieved an accuracy of 96.40% in Study Area 1.
• Achieved a Recall of 0.964 in Study Area 1.
• Achieved an F1-score of 0.964 in Study Area 1.
• Achieved a Kappa Coefficient of 0.960 in Study Area 1.
• Qualitative analysis of sample outputs and uncertainties, alongside quantitative evaluation of sample imbalance effects, demonstrates the framework’s robustness in complex agricultural environments.

Contributions

• Introduces the novel Difficulty-based Mixture of Experts Vision Transformer (DMoE-ViT) framework for crop type mapping.
• Effectively addresses the intra-class long tail distribution problem and intra-class variability by utilizing stratified sample partitioning, a multi-expert mechanism, and uncertainty quantification.
• Mitigates overfitting and enhances robustness in complex agricultural environments.
• Demonstrates superior crop classification performance compared to existing deep learning models.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Lu2025Uncertainty,
  author = {Lu, Q. Richard and Zhao, Wenzhi and Chen, Jiage and Chen, Xuehong and Zhang, Liqiang},
  title = {Uncertainty Mixture of Experts Model for Long Tail Crop Type Mapping},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17223752},
  url = {https://doi.org/10.3390/rs17223752}
}