Application of multilayer feed forward neural networks to automated compound identification in low-resolution open-path FT-IR spectrometry

A drawback of current open-path Fourier transform infrared (OP/FT-IR) syste ms is that they need a human expert to determine those compounds that may b e quantified from a given spectrum. In this work, multilayer feedforward ne ural networks with one hidden layer were used to automatically recognize co mpounds in an OP/FT-IR spectrum without compensation of absorption lines du e to atmospheric H2O and CO2. The networks were trained by fast-back-propag ation. The training set comprised spectra that were synthesized by digitall y adding randomly scaled reference spectra to actual open-path background s pectra measured over a variety of path lengths and temperatures; The refere nce spectra of 109 compounds were used to synthesize the training spectra. Each neural network was trained to recognize only one compound in the prese nce of up to 10 other interferences in an OP/FP-IR spectrum. Every compound in a database of vaporphase reference spectra can be encoded in an indepen dent neural network so that a neural network library can be established. Wh en these networks are used for the identification of compounds, the process is analogous to spectral library searching, The effect of learning rate an d band intensities on the convergence of network training was examined. The networks were successfully used to recognize five alcohols and two chlorin ated compounds in field-measured controlled-release OP/ET-IR spectra of mix tures of these compounds.