Comparison of multiple Chinese earth observation data for urban land use classification with different classification methods

Abstract

It still remains unsolved that to what extent the Chinese satellite data sources can be used to substitute each other in land use and land cover mapping and change analysis. We compared three Chinese land observation data sources (CBERS-02B, BEIJING-1 and HJ-1B) for urban land use classification using five classification algorithms including minimum distance classification (MDC), maximum likelihood classification (MLC), neural network classification (NN), support vector machines (VCMs) and decision trees (DTs). Our results show: the MLC and SVMs can achieve good results for CBERS-02B data; DTs and SVMs can achieve good results for BEIJING-1 data; The SVMs and MLC can achieve good results for HJ-1B data. BEIJING-1 data led to superior results to the other two data sources when classified just based on spectral information.