Abstract
This paper describes the evaluation process of a calibration method, which utilises a structure of multiple layers of partial least squares (PLS) calibration models to obtain accurate constituent data. This method is hereafter referred to as multi-layer partial least squares (ML-PLS). In ML-PLS, the fundamental idea is that the PLS model prediction on one layer selects which PLS model to use on the next layer in the structure. The general rule for the structure is that the models should become more and more “local” from layer to layer. Thus, the method can be seen as a stepwise procedure that selects which local model to use for the measurement. The ML-PLS method was applied to the measurement of moisture content in timber using near infrared (NIR) spectroscopy. The spectra were collected from the surface of Norway spruce (Picea abies) wood samples. The samples were soaked in water and spectra collected as the samples dried. Initial calibration attempts showed that significant improvements in prediction accuracy could be obtained by using local models. Therefore, this became the subject of further investigations leading to the ML-PLS method. The highest accuracy obtained for moisture content with a global model, expressed as a root mean square error of prediction (RMSEP), was 2.11%. In comparison, by utilising the ML-PLS method, the RMSEP was reduced to 1.16%.
© 2010 IM Publications LLP
PDF Article
More Like This
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Contact your librarian or system administrator
or
Login to access Optica Member Subscription