Neural, fuzzy and statistical models for unsupervised pattern recognition of Fourier transform Raman spectra

The use of computational neural networks, fuzzy logic and statistical models for the supervised and unsupervised pattern recognition of Fourier-transform (FT) Raman spectra is described. Supervised feed-forward neural networks are used for the classification, interpretation and feature selection of FT-Raman spectra of hard and soft woods. Sensitivity analysis of fully-trained networks provides a method for the interpretation of spectral differences between the two types of woods. These spectral differences arise from differences in the degree of conjugation and in the cellulose content of the woods. The classification ability of the computational neural networks is shown to be excellent, even for poor quality or low-resolution FT-Raman spectra of woods.;Neural, fuzzy and statistical unsupervised pattern recognition models are shown to complement the conventional (i.e., group-frequency based) methods for the interpretation of FT-Raman spectra of NO...