Deng et al. (2026) Application of Deep Learning Semantic Segmentation Models in Remote Sensing-Based Cropland Non-Grain and Non-Agriculturalization Monitoring: A Comparative Study

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

Identification

• Journal: Land
• Year: 2026
• Date: 2026-08-09
• Authors: Zhao Deng, Ming Cheng, Junde Xie, Tianyong Wan, Pengzhi Yang, Jianbo Tan, Sixue Xia, Jia Zhang, Xin Wu
• DOI: 10.3390/land15081437

Research Groups

Not specified

Short Summary

This study evaluates the performance, robustness, and generalization of seven mainstream semantic segmentation models for cropland non-grain and non-agriculturalization monitoring (CNNM) using UAV-derived imagery.

Objective

• To investigate the robustness and generalization ability of deep learning-based semantic segmentation models for CNNM in complex scenarios and identify the factors influencing their performance.

Study Configuration

• Spatial Scale: Local/Field scale (UAV-based high-resolution imagery)
• Temporal Scale: Not specified

Methodology and Data

• Models used: Seven mainstream semantic segmentation models (comprising both generic visual models and remote sensing-specific models).
• Data sources: Two self-constructed unmanned aerial vehicle (UAV)-based cropland monitoring datasets.

Main Results

• Remote sensing-specific semantic segmentation models demonstrated superior performance compared to generic visual models.
• Increasing the depth of the model backbone does not consistently guarantee significant improvements in performance.
• Specific categories, particularly agricultural facility land, pose significant recognition challenges, highlighting the need for tailored algorithms.

Contributions

• Provides a comparative benchmark of mainstream deep learning models for CNNM.
• Identifies the limitations of generic visual models and the lack of correlation between backbone depth and performance in this specific application.
• Offers practical guidance for the development of future tailored semantic segmentation algorithms for cropland monitoring.

Funding

Not specified

Citation

@article{Deng2026Application,
  author = {Deng, Zhao and Cheng, Ming and Xie, Junde and Wan, Tianyong and Yang, Pengzhi and Tan, Jianbo and Xia, Sixue and Zhang, Jia and Wu, Xin},
  title = {Application of Deep Learning Semantic Segmentation Models in Remote Sensing-Based Cropland Non-Grain and Non-Agriculturalization Monitoring: A Comparative Study},
  journal = {Land},
  year = {2026},
  doi = {10.3390/land15081437},
  url = {https://doi.org/10.3390/land15081437}
}