Land classification from LiDAR full-waveforms based on multi-class support vector machines

Abstract

In this study, a multi-class support vector machines (SVM) based land classification method is presented to predict the land types of Beijing area. The returned full-waveforms were collected from the Ice, Cloud and land Elevation Satellite (ICESat) mission and the Full Width at Half Maximum (FWHM) of the full-waveforms were used to be the attributes of test data for generating the SVM prediction model. FWHM were obtained from waveforms filtered by Empirical Mode Decomposition (EMD). The SVM prediction model with high cross validation accuracy was selected to predict the land types of Beijing area. GLAS full-waveforms, which were used to predict and validate the land classification, were acquired when ICESat was passing over Beijing urban and rural areas from 1st Jan 2003 to 31st Dec 2005. Besides of terrace and building, the main land types of Beijing area are plain and stone Mountain that lacks of trees. Thus the received waveforms of ICESat/GLAS were divided into five kinds, `invalid', `plain', `terrace', `building' and `mountain' waveforms. Over this test site, the algorithm achieved an overall classification accuracy of 91.5%. This method can be developed to be an on-line automation algorithm to classify the land type.