Classification of Near-Infrared Spectra Using Wavelength Distances: Comparison to the Mahalanobis Distance and Residual Variance Methods

A simple and easy to understand method for classification of near-infrared spectra is reported. The method uses a sample's normalized distance from a library of mean spectra. The probability distribution of the test is described, and its ability to discriminate between similar materials was tested and is reported. Its ability to detect samples that fail to meet product specifications and samples adulterated with minor levels of impurities was also tested and is reported. The performance of the method is compared to methods based on principal component analysis, Mahalanobis distances, and SIMCA residual variance distances. Overall, the wavelength distance method gave better classification results than the Mahalanobis and SIMCA methods when small training sets were used, but poor results were obtained in the detection of samples that do not meet product specifications and samples adulterated with low levels of contamination