Zhu et al. (2025) KADL: Knowledge-Aided Deep Learning Method for Radar Backscatter Prediction in Large-Scale Scenarios

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-05
• Authors: Dong Zhu, Zhao Peng, Qiang Zhao, Qingliang Li, Jinpeng Zhang, Lixia Yang
• DOI: 10.3390/rs17243933

Research Groups

Not specified in the provided text.

Short Summary

This paper proposes a novel knowledge-aided deep learning (KADL) method for predicting large-scale radar backscatter, demonstrating superior accuracy (root mean square error of 4.74 dB), robustness, and generalization compared to existing empirical and purely data-driven models by integrating physical knowledge.

Objective

• To develop a novel knowledge-aided deep learning method that accurately and robustly predicts radar backscatter from large-scale scenarios, overcoming the limitations of traditional empirical and purely data-driven deep learning methods regarding physical consistency and generalization.

Study Configuration

• Spatial Scale: Large-scale scenarios
• Temporal Scale: Not specified in the provided text.

Methodology and Data

• Models used:
  • Knowledge-Aided Deep Learning (KADL) method, comprising:
    • Knowledge Perception Module (KPM)
    • Knowledge-Weighted Fusion (KWF) strategy
    • Cascaded Deep Neural Network (DNN) solver
  • Specific empirical models (refitted for comparison)
  • Advanced Nonhomogeneous Terrain Clutter Model (ANTCM) (for comparison)
• Data sources:
  • Multi-source remote sensing data (used to derive dielectric properties)
  • Measured data (for experiments and comparison)
  • Priori physical knowledge: soil moisture, leaf area index (LAI)

Main Results

• KADL achieved a root mean square error (RMSE) of 4.74 dB on independent test data.
• KADL achieved a mean absolute percentage error (MAPE) of 8.7% on independent test data.
• KADL demonstrated superior robustness, with a standard deviation of RMSE as low as 0.18 dB across multiple trials.
• The proposed KADL method validated superior accuracy, robustness, and generalization ability for large-scale backscatter prediction compared to traditional empirical and advanced nonhomogeneous terrain clutter models.

Contributions

• Proposes a novel knowledge-aided deep learning (KADL) method that effectively integrates physical knowledge into a deep learning framework for radar backscatter prediction.
• Introduces a Knowledge Perception Module (KPM) and an efficient Knowledge-Weighted Fusion (KWF) strategy to enhance feature exploitation and suppression of non-informative features.
• Addresses the limitations of traditional empirical methods (simplified physics, limited parameters) and purely data-driven deep learning methods (lack of physical consistency, poor generalization) in large-scale radar backscatter modeling.
• Demonstrates superior accuracy, robustness, and generalization ability of KADL through extensive experiments on measured data.

Funding

Not specified in the provided text.

Citation

@article{Zhu2025KADL,
  author = {Zhu, Dong and Peng, Zhao and Zhao, Qiang and Li, Qingliang and Zhang, Jinpeng and Yang, Lixia},
  title = {KADL: Knowledge-Aided Deep Learning Method for Radar Backscatter Prediction in Large-Scale Scenarios},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17243933},
  url = {https://doi.org/10.3390/rs17243933}
}