# Improvement of PolSAR Decomposition Scattering Powers Using a Relative Decorrelation Measure

## TL;DR

This paper introduces a novel method to enhance PolSAR scattering power decomposition by leveraging complex correlation coefficients, leading to more physically meaningful results and reduced negative power pixels.

## Contribution

The approach uniquely utilizes off-diagonal coherency matrix elements and a degree of polarization optimization to improve scattering power estimates in PolSAR data.

## Key findings

- Improved scattering power estimates compared to existing methods.
- Reduction in negative power pixels.
- Enhanced physical relevance of the scattering powers.

## Abstract

In this letter, a methodology is proposed to improve the scattering powers obtained from model-based decomposition using Polarimetric Synthetic Aperture Radar (PolSAR) data. The novelty of this approach lies in utilizing the intrinsic information in the off-diagonal elements of the 3$\times$3 coherency matrix $\mathbf{T}$ represented in the form of complex correlation coefficients. Two complex correlation coefficients are computed between co-polarization and cross-polarization components of the Pauli scattering vector. The difference between modulus of complex correlation coefficients corresponding to $\mathbf{T}^{\mathrm{opt}}$ (i.e. the degree of polarization (DOP) optimized coherency matrix), and $\mathbf{T}$ (original) matrices is obtained. Then a suitable scaling is performed using fractions \emph{i.e.,} $(T_{ii}^{\mathrm{opt}}/\sum\limits_{i=1}^{3}T_{ii}^{\mathrm{opt}})$ obtained from the diagonal elements of the $\mathbf{T}^{\mathrm{opt}}$ matrix. Thereafter, these new quantities are used in modifying the Yamaguchi 4-component scattering powers obtained from $\mathbf{T}^{\mathrm{opt}}$. To corroborate the fact that these quantities have physical relevance, a quantitative analysis of these for the L-band AIRSAR San Francisco and the L-band Kyoto images is illustrated. Finally, the scattering powers obtained from the proposed methodology are compared with the corresponding powers obtained from the Yamaguchi \emph{et. al.,} 4-component (Y4O) decomposition and the Yamaguchi \emph{et. al.,} 4-component Rotated (Y4R) decomposition for the same data sets. The proportion of negative power pixels is also computed. The results show an improvement on all these attributes by using the proposed methodology.