Machine Learning-Assisted Analysis of Polarimetric Scattering From Cylindrical Components of Vegetation

Abstract

Reliable and efficient analysis of electromagnetic scattering by cylindrical components of vegetation is important for microwave remote sensing of vegetated terrain. In this paper, we proposed a machine learning (ML) scheme for the analysis of polarimetric bistatic scattering from a finite dielectric cylinder. A deep neural network architecture is adopted in the hope that with increased depth of the neural network, hence increased abstraction capability, it may be able to handle the highly oscillatory scattering patterns to an adequately acceptable degree. The scheme has demonstrated the capability of modifying and adapting itself to capture the complicated polarimetric bistatic scattering patterns of a finite dielectric cylinder. The physical consideration of reciprocity relation is largely fulfilled except for the scattered directions close to the cylinder axis. Moreover and more importantly, for cases where interpolation is expected, the scheme has unambiguously demonstrated the capability of learning the bistatic scattering cross section and phase patterns. The performance is also robust against the number of parameters to be interpolated, be it single or multiple. In summary, the proposed ML scheme bodes well for the design of the future physically based algorithms where the conventional datacube was used as the base for interpolation.