Cotton (test 2) Cotton 0,73 Linen 1,28
In discriminant analysis the distance from a class shows how similar is the test sample spectrum to the spectra of the calibration samples of that class. Shorter distance to a class means higher probability of belonging to that class. Two most probable classes (Class 1, Class 2) and their distances were calculated for each sample. As it can be seen from Table 2, the method distinguishes between the different fibre classes well. Bold font indicates samples, in which case fibres were identified correctly, and italic font shows wrong identifications. The only failure is seen when linen was incorrectly assigned to cotton class. This is not unexpected because both fibres are composed of cellulose and their IR spectra are essentially identical. At the same time the assignments of cotton and viscose – very similar to linen – were correct. For viscose, distance to class difference for class 1 and class 2 is big enough to consider the result reliable.
3.3.2. Classification of Mixed Two-component Fibres
Textiles of 11 pure fibres and 15 different binary fibres, 316 spectra in total, were used for creating the model for this analysis. The variances described by the principal components were: PC1 46%, PC2 26% and PC3 18%. In total 10 principal components were calculated. Fig. 3 presents the results of PCA analysis of two-component fibres.