Classification of pernambuco (Caesalpinia echinata Lam.) wood quality by near infrared spectroscopy and linear discriminant analysis

Near infrared (NIR) spectroscopy, coupled with multivariate data analysis, is proposed as a rapid and effective analytical method for evaluating the quality of pernambuco (Caesalpina echinata Lam.) wood for making bows for stringed instruments. For this purpose, a set of 30 pernambuco sticks were ranked based on their suitability for making high-quality bows and they were assigned to one of the following categories: 0=very poor to poor, 1=good to very good and 2=excellent. Considering the low number of samples in the poor category, the classification study focused on the discrimination between samples of the two higher quality groups. Linear discriminant analysis (LDA) was applied to the NIR data as a classification technique and in order to ensure a more appropriate ratio between the number of objects (samples) and number of variables (absorbance at different wavenumbers), LOA was preceded by feature selection. Based on LDA, 100% of the samples were correctly classified and 92.6% of the samples were correctly predicted by the cross-validation procedure.