A Review on PolSAR Decompositions for Feature Extraction

Abstract

Feature extraction plays a pivotal role in processing remote sensing datasets, especially in the realm of fully polarimetric data. This review investigates a variety of polarimetric decomposition techniques aimed at extracting comprehensive information from polarimetric imagery. These techniques are categorized as coherent and non-coherent methods, depending on their assumptions about the distribution of information among polarimetric cells. The review explores well-established and innovative approaches in polarimetric decomposition within both categories. It begins with a thorough examination of the foundational Pauli decomposition, a key algorithm in this field. Within the coherent category, the Cameron target decomposition is extensively explored, shedding light on its underlying principles. Transitioning to the non-coherent domain, the review investigates the Freeman–Durden decomposition and its extension, the Yamaguchi’s approach. Additionally, the widely recognized eigenvector–eigenvalue decomposition introduced by Cloude and Pottier is scrutinized. Furthermore, each method undergoes experimental testing on the benchmark dataset of the broader Vancouver area, offering a robust analysis of their efficacy. The primary objective of this review is to systematically present well-established polarimetric decomposition algorithms, elucidating the underlying mathematical foundations of each. The aim is to facilitate a profound understanding of these approaches, coupled with insights into potential combinations for diverse applications.

Correlated Decision Fusion Accompanied with Quality Information on a Multi-Band Pixel Basis for Land Cover Classification

“…Our methodology involves a pixel-level correlated decision fusion, which enhances the accuracy and robustness of land cover classification. In reference [31], a multitude of decomposition methods are analyzed and presented that have been used to extract the biophysical scattering behavior of SAR data. However, in this work we utilized Pauli's decomposition components and land surface temperature (LST) as features to extract local decisions for each pixel, considering the unique information provided by each sensor.…”

Inversion of Forest Aboveground Biomass in Regions with Complex Terrain Based on PolSAR Data and a Machine Learning Model: Radiometric Terrain Correction Assessment

“…Feature selection is essential to achieve robust and high-precision estimation of forest AGB based on PolSAR data [43][44][45]. The data and features determine the upper limit of machine learning, while models and algorithms only approach this upper limit [46].…”