An application of wavelet transforms and neural networks for decomposition of millimeter-wave spectroscopic signals

This paper reports on wavelet-based decomposition methods and neural networks for remote monitoring of airborne chemicals using millimeter-wave spectroscopy. Because of instrumentation noise and the presence of untargeted chemicals, direct decomposition of the spectra requires a large number of data to train a neural network and yields low accuracy. We have demonstrated that a neural network trained with features obtained from a discrete wavelet transform provides better decomposition with faster training time. Results based on synthesized and experimental spectra are presented to show the efficacy of the wavelet-based methods.