Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Deep Learning Approach for SAR Tomographic Imaging of Forested Areas

Zoé Berenger, Loïc Denis, Florence Tupin, Laurent Ferro-Famil, Yue Huang

TL;DR

It is shown that lightweight neural networks can be trained to perform a regularized inversion implemented in the form of iterative minimization algorithms with a single feed-forward pass, leading to fast reconstructions that could better scale to the amount of data provided by the future BIOMASS mission.

Abstract

Synthetic aperture radar tomographic imaging reconstructs the three-dimensional reflectivity of a scene from a set of coherent acquisitions performed in an interferometric configuration. In forest areas, a large number of elements backscatter the radar signal within each resolution cell. To reconstruct the vertical reflectivity profile, state-of-the-art techniques perform a regularized inversion implemented in the form of iterative minimization algorithms. We show that light-weight neural networks can be trained to perform the tomographic inversion with a single feed-forward pass, leading to fast reconstructions that could better scale to the amount of data provided by the future BIOMASS mission. We train our encoder-decoder network using simulated data and validate our technique on real L-band and P-band data.

A Deep Learning Approach for SAR Tomographic Imaging of Forested Areas

TL;DR

It is shown that lightweight neural networks can be trained to perform a regularized inversion implemented in the form of iterative minimization algorithms with a single feed-forward pass, leading to fast reconstructions that could better scale to the amount of data provided by the future BIOMASS mission.

Abstract

Synthetic aperture radar tomographic imaging reconstructs the three-dimensional reflectivity of a scene from a set of coherent acquisitions performed in an interferometric configuration. In forest areas, a large number of elements backscatter the radar signal within each resolution cell. To reconstruct the vertical reflectivity profile, state-of-the-art techniques perform a regularized inversion implemented in the form of iterative minimization algorithms. We show that light-weight neural networks can be trained to perform the tomographic inversion with a single feed-forward pass, leading to fast reconstructions that could better scale to the amount of data provided by the future BIOMASS mission. We train our encoder-decoder network using simulated data and validate our technique on real L-band and P-band data.
Paper Structure (14 sections, 5 equations, 4 figures, 2 tables)

This paper contains 14 sections, 5 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem formulation
  4. From fixed dictionaries to learned representations: strategies for vertical profile reconstruction
  5. Proposed method
  6. Generative model
  7. Experiments
  8. Simulated data
  9. Analysis of the reconstructed profiles
  10. Architecture choice
  11. Boreal forest at L band
  12. Tropical forest at P band
  13. Discussion
  14. Conclusion

Figures (4)

  • Figure 1: Pipeline of the data simulation, training and testing process of the proposed method.
  • Figure 2: Reconstruction of a profile consisting of wide (left) and narrow (right) Gaussians with Beamforming, Capon, Wavelet-based CS and our method. For each approach, the average profile over 100 measurements with various speckle realizations and its interquartile range have been plotted, with the reference profile shown in black.
  • Figure 3: Tomogram for a specific azimuth value in a boreal forest at L band derived with the method: (a) Beamforming; (b) Capon; (c) Wavelet-based CS; (d) Predicted by the proposed neural network.
  • Figure 4: Tomogram for a specific azimuth value in a tropical forest at P band derived with the method: (a) Beamforming; (b) Capon; (c) Wavelet-based CS; (d) Predicted by the proposed neural network.