Current Development in Theory and Applications of Wavelets
-->
Abstract: We introduce a novel method to analyse near
infrared (NIR) spectra collected in a repeated measures experiment. Using an
adaptive discrete wavelet transform, our method initially extracts features from
the spectra that correlate with the design of the experiment. Then the extracted
features are then mapped onto a five-dimensional hyperplane using penalised
discriminate mapping (PDM) to form PDM scores. The PDM scores are analysed using
a multivariate mixed model (MMM) to determine if the experimental design affects
the NIR spectra. Illustration of the method is given by a case study from the
viticulture industry, where NIR reflectance measurements (400, 402, ..., 2500nm)
were taken from red grape homogenates sampled from a nested repeated measures
experimental design consisting of the following factors: various growing
regions, vineyards, grape varieties and storage durations. Analyses of the
viticulture example using our proposed method identified all main effects and
two-way interactions between regions, grape variety and, most importantly,
storage duration to all be significant By
visualization and univariate analysis of the PDM scores, we identified regions
in the NIR spectrum associated with the storage duration, variety and
interaction effects.
Keywords and phrases: adaptive wavelet, repeated measures, multivariate MANOVA, near infrared, wine grape, penalized discriminate mapping.