From measurement to material - preparing hyperspectral signatures for classification

Due to the possibility of classifying unknown materials fast and accurately the industries interest in spectroscopy is growing. However, reliable classification is a matter of suitable preprocessing. Existing solutions found in the literature are often very specific a particular combination of materials. In this paper we present a method to preprocesses hyperspectral data in order to enables general classification of many materials. The system is divided into five modules: selection, transformation, reduction, decorrelation and classification. We demonstrate our method in a demonstrator system that is available as both web- and standalone application.