Simultaneous Multicomponent Organic Compound Analysis Using Two Intelligent Signal Processing Methods

Ling Gao,Shouxin Ren

DOI: https://doi.org/10.1109/bmei.2012.6512969

2012-01-01

Abstract:Two intelligent signal processing methods, independent component analysis-based latent variable regression (ICA-LVR) and wavelet packet transform based latent variable regression method (WPT-LVR) methods were developed to perform simultaneous spectrophotometric determination of methy1 salicylate (MSA), dibuty1 phthalate (DBP) and potassium hydrogenphthalate (PHP). Independent component analysis is a newly developed signal processing technique aiming at solving related blind source separation (BSS) problem. ICA can be used to extract independent source variables and their corresponding concentration data from the observed mixture spectra without using any prior knowledge about the components. Wavelet packet representations of signals provided a local time-frequency description and separation ability between information and noise. The quality of the noise removal can be improved by using best-basis algorithm and thresholding operation. In this case, ICA and WPT were used to reduce dimensionality and to extract information from the observed mixture spectra. Latent variables were made by projecting the ICA-processed or WPT-processed signals onto orthogonal basis eigenvectors. The model of the independent source matrix or WPT-processed signals with concentration matrix was build by latent variable regression (LVR). Two programs, GICALVR and GWPTLVR, were designed to perform simultaneous multicomponent determination. Experimental results showed the ICA-LVR and WPT-LVR methods to be successful even where there was severe overlap of spectra and had the clear superiority over the LVR method.