An InSAR Interferogram Filtering Method Based on Multi-Level Feature Fusion CNN.

Area of Science:

• Geophysics
• Remote Sensing
• Computer Science

Background:

• Interferogram filtering is crucial for accurate InSAR data processing, including terrain reconstruction and deformation monitoring.
• Traditional methods struggle with complex interferogram distributions and simultaneous noise suppression and detail preservation.
• Existing filters often fail to adapt to diverse noise conditions and spatial complexities.

Purpose of the Study:

• To develop an advanced interferogram filtering model using a convolutional neural network (CNN).
• To improve upon traditional methods by better preserving phase details and suppressing noise.
• To enhance the accuracy and efficiency of InSAR data processing for deformation monitoring.

Main Methods:

• A novel CNN-based multi-level feature fusion model was designed, utilizing a multi-depth, multi-path convolution strategy.
• The model was evaluated using simulated data with qualitative and quantitative analyses to assess performance and generalization.
• Real-world InSAR data were used for filtering and unwrapping experiments, and time-series InSAR (TS-InSAR) assessed deformation monitoring accuracy.

Main Results:

• The proposed CNN model demonstrated superior performance in preserving phase details while effectively suppressing noise compared to traditional methods.
• Experiments validated the model's strong generalization capabilities on diverse interferogram datasets.
• The filtering method showed significant advantages in both performance and efficiency, improving deformation monitoring accuracy.

Conclusions:

• The developed CNN-based interferogram filtering model offers a significant advancement over traditional techniques.
• This method provides enhanced accuracy and efficiency for InSAR data processing, particularly for deformation monitoring.
• The findings open new research avenues for high-precision InSAR data analysis and practical applications.