A Highly Efficient Method for Training Sample Selection in Remote Sensing Classification

Abstract

Remote sensing classification is an important way to obtain land cover information, and the selection of classification training samples for most of the classification method is an expensive and time-consuming task. However, the traditional training samples selection method is a direct selection based on two-dimensional (2D) images, therefore, training sample selection efficiency is always low in the regions with complex terrain and landscape fragmentation, and the ROI (region of interest) separability is unsatisfactory for classification. This study aims at the low efficiency and low ROI separability for traditional training sample selection method put forward a new training sample selection method using a three-dimensional (3D) terrain model that was created by OLI image fusion digital elevation model (DEM) to select ROIs, which departs from the traditional method based on a two-dimensional image. A Landsat-8 OLI image of the Yunlong Reservoir Basin in Kunming was used to test this proposed method. Study results showed that the proposed method obtained ROI separability that was greater than 1.9, and with most reaching 2.0; while the ROI separability of traditional method still had unqualified situation, which showed the new method was more effective.