[Automated recognition of VOCs using artificial neural networks]

Quantitative analysis of FTIR spectra, which are seriously overlapped in the spectral bands, was studied by artificial neural networks. The optimum network was chosen by a new criterion, i. e. the degree of approximation. After the network was established, two kinds of spectra were resolved. It was demonstrated that accurate results could be obtained when two components were both included. In addition, the unknown spectrum could be identified and quantified. It was showed that the artificial neural network has excellent non-linear ability of solution. Meanwhile, the method provides an efficient approach to the identification and quantification of the unknown samples.