Share Email Print

Journal of Applied Remote Sensing • Open Access • new

Terrain classification of polarimetric synthetic aperture radar imagery based on polarimetric features and ensemble learning
Author(s): Chuanbo Huang

Paper Abstract

An evolutionary classification system for terrain classification of polarimetric synthetic aperture radar (PolSAR) imagery based on ensemble learning with polarimetric and texture features is proposed. Polarimetric measurements cannot produce sufficient identification information for PolSAR terrain classification in some complex areas. To address this issue, texture features have been successfully used in image segmentation. The system classification feature has been adopted using a combination of Pauli features and the last principal component of Gabor texture-feature dimensionality reduction. The resulting feature combination assigned through experimental analysis is very suitable for describing structural and spatial information. To obtain a good integration effect, the basic classifier should be as precise as possible and the differences among the features should be as distinct as possible. We therefore examine and construct an ensemble-weighted voting classifier, including two support vector machine models that are constructed using kernel functions of the radial basis and sigmoid, extreme learning machine, k -nearest neighbor, and discriminant analysis classifier, which can avoid redundancy and bias because of different theoretical backgrounds. An experiment was performed to estimate the proposed algorithm’s performance. The results verified that the algorithm can obtain better accuracy than the four classifiers mentioned in this paper.

Paper Details

Date Published: 3 April 2017
PDF: 13 pages
J. Appl. Remote Sens. 11(2) 026002 doi: 10.1117/1.JRS.11.026002
Published in: Journal of Applied Remote Sensing Volume 11, Issue 2
Show Author Affiliations
Chuanbo Huang, Southwest Univ. of Science and Technology (China)

© SPIE. Terms of Use
Back to Top