Use of regularizing methods in modeling and prediction of wood weathering as a response to UV radiation

Abstract

Samples of norway spruce (Picea abies) were exposed to various amounts of natural or arti cial UV radiation and recorded with a NIR hyperspectral camera. Multivariate statistical analysis using Tikhonov regularization, Principal Component Regression and Partial Least squares were applied to the resulting data. Results were mixed, but with consistently better models for the samples exposed to arti cial radiation, able to predict exposure amounts with a reasonable degree of accuracy and generalizability. Tikhonov regularization performed better or on par with PCR or PLS in most cases, and showed a tendency to better predict new data. Samples exposed to natural weathering yielded signi cantly worse results, presumably related to the myriad of interfering, unquanti ed phenomena present in the environment. It was thus concluded that the weathering e ect on wooden surfaces is well reflected in the NIR spectrum. NIR hyperspectral imaging holds promise as a useful tool in the further investigations into the subject, with the caveat that advancements need to be made in the statistical modeling of the data, to make it more robust and applicable outside of a purely research oriented environment.