Zhang et al. (2026) Fine-Grained Classification of Lakeshore Wetland–Cropland Mosaics via Multimodal RS Data Fusion and Weakly Supervised Learning: A Case Study of Bosten Lake, China

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

Identification

• Journal: Land
• Year: 2026
• Date: 2026-01-01
• Authors: Jing Zhang, Alim Samat, ErZhu Li, Enzhao Zhu, Wenbo Li
• DOI: 10.3390/land15010092

Research Groups

Not explicitly stated in the provided text.

Short Summary

This study evaluates deep learning models for high-precision classification of complex wetland-cropland mosaics in arid regions, demonstrating that multimodal remote sensing data fusion combined with weakly supervised learning can achieve high accuracy while significantly reducing labeling costs.

Objective

• To systematically compare the performance of fully supervised and weakly supervised deep learning models, enhanced by multimodal remote sensing data fusion, for classifying 19 wetland–cropland mosaic types in arid wetlands, aiming to balance high accuracy and low labeling costs.

Study Configuration

• Spatial Scale: Bosten Lake wetland, Xinjiang, China (case area).
• Temporal Scale: Not explicitly stated, but refers to a specific period of remote sensing data acquisition.

Methodology and Data

• Models used: Fully supervised models (FCN, U-Net, DeepLabV3+, SegFormer) and a weakly supervised learning model (One Model Is Enough - OME).
• Data sources: Pleiades and PlanetScope-3 multimodal remote sensing data, fused using the Gram–Schmidt method.

Main Results

• SegFormer achieved the best overall performance among fully supervised models with 98.75% accuracy and 95.33% mean Intersection over Union (mIoU).
• The weakly supervised OME model, using only image-level labels, matched fully supervised performance with 98.76% accuracy and 92.82% F1-score, significantly reducing labeling effort.
• Multimodal data fusion consistently boosted the accuracy of all models, most notably increasing U-Net’s mIoU by 63.39%.
• Models exhibited complementary strengths: U-Net excelled in wetland vegetation segmentation, DeepLabV3+ in crop classification, and OME in preserving spatial details.

Contributions

• Validates a novel pathway integrating multimodal remote sensing data fusion with weakly supervised learning to achieve high accuracy in arid wetland mapping while drastically reducing labeling costs.
• Provides a systematic comparison of state-of-the-art fully supervised and weakly supervised deep learning models for complex wetland-cropland mosaic classification.

Funding

Not explicitly stated in the provided text.

Citation

@article{Zhang2026FineGrained,
  author = {Zhang, Jing and Samat, Alim and Li, ErZhu and Zhu, Enzhao and Li, Wenbo},
  title = {Fine-Grained Classification of Lakeshore Wetland–Cropland Mosaics via Multimodal RS Data Fusion and Weakly Supervised Learning: A Case Study of Bosten Lake, China},
  journal = {Land},
  year = {2026},
  doi = {10.3390/land15010092},
  url = {https://doi.org/10.3390/land15010092}
}