Open AccessUnsupervised classification based on non‐negative eigenvalue decomposition and Wishart classifier
Author(s) -
Wang Chunle,
Yu Weidong,
Wang Robert,
Deng Yunkai,
Zhao Fengjun,
Lu Youchun
Publication year - 2014
Publication title -
iet radar, sonar and navigation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.489
H-Index - 82
eISSN - 1751-8792
pISSN - 1751-8784
DOI - 10.1049/iet-rsn.2014.0076
Subject(s) - wishart distribution , classifier (uml) , pattern recognition (psychology) , artificial intelligence , eigenvalues and eigenvectors , decomposition , eigendecomposition of a matrix , mathematics , computer science , machine learning , physics , multivariate statistics , chemistry , quantum mechanics , organic chemistry
In this study, the authors propose an unsupervised terrain and land‐use classification algorithm for polarimetric synthetic aperture radar (SAR) image analysis. Under the non‐reflection symmetry condition, the non‐negative eigenvalue decomposition (NNED) employing Arii volume scattering model is derived. They first apply NNED to divide pixels into three categories of surface, volume and double bounce scatterings. Then the pixels in each category are further divided into several classes based on the scattering characteristic parameter of the dominant scattering component. Utilising the initial classification result as training sets, the complex Wishart classifier can then be performed within each category or beyond categories to refine the final classification result. The effectiveness of this algorithm is demonstrated using the German Aerospace Center's E‐SAR polarimetric data acquired over the Oberpfaffenhofen area in Germany.