Delaunay Triangulation Local Method for Analysis of Near-Infrared Spectra of Plant Samples

Abstract

The proposed delaunay triangulation local method is applied as a local calibration method in the analysis of near infrared data. The method forms mesh of simplexes in principal component space and then carries out calibration subsets selection based on the situation of unknown sample in the mesh. It obtains higher prediction accuracy than the local method, in which calibration subsets are the closest points chosen through immediate Euclidean distances between samples in principal component space. On comparing with global PLS method, delaunay triangulation local method have better results with few numbers of principal components. Furthermore, it is no need to establish the calibration model and it is fast and simple. As a result, the proposed delaunay triangulation method can be used as a valid local method for distinguishing similarity of complex samples in near infrared spectral data analysis.