Hyperspectral Image Classification for Mineralogical Identification

Classification is broadly used in a range of exploration technologies. This paper will summarize standard classification methods and provide a new hybrid method for classifying spectra used in mapping mineralogy with hyperspectral remote sensing data. Classification methods commonly employed do not allow the user specifically to identify key features that give confidence as to the quality of the classification. Methods that do allow feature selection typically use the position of specific absorption features only. Sometimes these positions are not in themselves diagnostic and other broad feature shapes are often useful in expert interpretation. A new hybrid method is presented in this paper that uses both the position and character of the absorption feature and longer wavelength characteristics such as broad feature shape and asymmetry. Further, this method develops the concept of a “verification image” which, when used with the classification, can give confidence that those image pixels identified contain specific feature characteristics, apart from the classification itself, that gives further support for the classification result.