An automatic optimization method of forest type classification using PolSAR image based on genetic algorithm

Abstract

In order to improve the performance of nonparametric classifier on high dimensional polarization features set, an automatic optimization method based on genetic algorithm is proposed, and is used for polarimetric synthetic aperture radar (PolSAR) image forest land type classification. The method focusing on two main aspects that affect the classification performance, which are features combination and model hyperparameter. Different from conventional process which optimize those two aspects respectively. Our proposed method using genetic algorithm as searching engine, by regarding features combination and hyperparameters as a set of model settings. We can optimize those two aspects simultaneously, so that the synergy affect between those two aspects can be considered. A PolSAR image (C-band, quad polarization) data were used to verify the proposed optimization method using support vector machine (SVM) as classifier.