Classification of Urban Hyperspectral Remote Sensing Imagery Based on Optimized Spectral Angle Mapping

Abstract

Hyperspectral remote sensing imagery provides highly precise spectral information. Thus, it is suitable for the land use classification of urban areas that are composed of complicated structures. In this study, a new spectral angle and vector mapping (SAVM) classification method, which adds a factor based on “the differences in the spectral vector lengths” among image pixels to the spectral angle mapping (SAM) classification method, is proposed. The SAM and SAVM methods were applied to classify the aerial hyperspectral digital imagery collection experiment imagery acquired from the business district of Washington, DC, USA. The results demonstrated that the overall classification accuracy of the SAM was 64.29%, with a Kappa coefficient of 0.57, while the overall classification accuracy of the SAVM was 81.06%, with a Kappa coefficient of 0.76. The overall classification accuracy was improved by 16.77% by the SAVM, indicating that the use of a SAVM classification method that considers both the spectral angle between the reference spectrum and the test spectrum and the differences in the spectral vector lengths among image pixels can improve the classification accuracy of urban area with hyperspectral remote sensing imagery.