Classification of hyperspectral data for land cover mapping : is there any significant improvement?

Abstract

This paper highlights the results of classification of an airborne MASTER hyperspectral data for land cover mapping in Redang Island, Malaysia. Two addressed issues in the study are: (1) whether or not hyperspectral would increase classification accuracy over medium spatial resolution (10m) of MASTER data for land cover mapping, and (2) radiometric normalization still required in hyperspectral data Three classification algorithms examined in this study, namely Binary encoding, Spectral Angle Mapper and Linear spectral Unmixing. The topographic-effect normalization was applied to the test site prior data classification. Results of study indicated that Linear Spectral Unmixing classification technique gives the best overall classification accuracy of the hyperspectral data for land cover in the study area. The result of this study also clearly indicated that hyperspectral data could not improve classification accuracy significantly especially when the .mixed pixels are abundant