Derivative hyperspectral image analysis for land use classification

As hyperspectral remote sensing data become commonly available, researchers need an effective tool specifically designed for analyzing this new type of data. Derivative analysis has been proved a useful tool capable of detecting subtle information from hyperspectral data sets. However, there are no systematic procedures for effectively applying derivative analysis to remote sensing hyperspectral images yet.; This research developed a systematic procedure for using derivative analysis to help improve supervised classification of hyperspectral images. The algorithm allows investigators to identify derivative features better separating target classes according to the Jeffries-Matusita distances between classes. These features can be added into the classification image in order to improve the classification result.; A maximum likelihood classification for vegetation was used as an example in this research. It demonstrated the effectiveness of using derivatives to detect useful information that might be lost during feature reduction operations. Classification accuracies of classes that were poorly classified in a 10-band principal component image gradually improved as more appropriate derivative features were extracted from the original image and appended to the base image.; With the data set used in this study, derivative analysis did not generally provide a better performance than principal component analysis, but it may be suitable for some data sets and applications. The procedure developed in this research can be used as a starting point for subsequently designing an advanced system to systematically analyze hyperspectral images for remote sensing applications.