Investigation of the mechanism and algorithm of principal component regression based on daubechies wavelet

In this paper, a new algorithm for multivariate calibration named principle component regression based on wavelet (PCRW) is proposed. The algorithm is constructed by integrating wavelet transform with principal component regr ession. The results of theoretical analysis and simulated experiments demon strate that the new algorithm can more effectively filter off noise anti ex tract useful information from the actual spectral data. Applying this metho d to the practical analysis of Chloramphenicolum and Metronidazok leads to a decrease in the average of mean relative error (MRE) from 1.70% obtained by PCR to 0.90% obtained by PCRW. Furthermore, by combining statistical cri teria and multiscale analysis of wavelet, a new method for determining the number of principle components is developed. The theoretical and experiment al investigations show that the new method is more reliable than convention al ones.