Search prefilters for library matching of infrared spectra in the PDQ database using the autocorrelation transformation

摘要

The feasibility of using the autocorrelation transformation in combination with pattern recognition methods to develop search prefilters to facilitate infrared spectral library searching of the PDQdatabase has been investigated. The use of the autocorrelation transformation addressesmany problems encountered when transferring classificationmodels between spectrometers. The autocorrelation transformation is also sensitive at distinguishing subtle but significant features in spectral data such as minor peaks, shoulders, and peakswith unique shapes. Previous workers have shown that peak shifts and related spectral alignment problems can be obviated by using the autocorrelation transformation. The autocorrelation transformation produces a histogramfor each IR spectrum,which can be a more useful representation of spectra for pattern recognition involving data collected on different spectrometers. Instead of directly sampling points from the histograms using classification techniques such as SIMCA to identify the informative region, each histogramwas first deconvolved to better capture the signal using Coiflet wavelets with the informative wavelet coefficients identified using a genetic algorithm for pattern recognition and feature selection.