Research And Application On Adaptive Spectra Preprocessing Methods Based On EMD And EEMD

Raman and Near Infrared(NIR) spectral analysis technology, with its nondestructive and rapid measure performance, have been one of the most important analysis and testing measures in modern time over decades. Raman and NIR signal have weak energy and are susceptible to interference, which will affect directly the quality and applied range of calibration model, so it is very important to preprocess spectra. General correction method to denoising and baseline, such as wavelet, derivation, interpolation, and polynomial fitting, need to adjust basis function, decomposition levels and signal reconstruction of algorithm according to character of different spectrum. Because of parameters setting with on apriority, it is difficult to realize optimal preprocessing for Raman spectrum, which has sharp peaks and changeable waveform. In order to solve weak adaptability caused by general spectral preprocessing, improve the denoising effect of Raman and NIR spectra, enhance the correction accuracy of spectral baseline, and increase detection accuracy of spectral model, this paper mainly discusses on several aspects as follows:Firstly, modes overlapping occurs easily between Raman spectrum’s high frequency signal and noise, so the Ensemble Empirical Mode Decomposition(EEMD) method was used to remove the noise. According to the approximate completeness of Raman spectrum, an EEMD denoising method based on normalized index optimization. This method views the global optimal point of normalized denoising index as stop condition, and completes itself adaptively with no parameters setting during the decomposion. It can be concluded by Raman spectrum denoising data that this method improves the denoising accuracy and decreases the difficulty in operation in an easy and efficient way.Secondly, in order to improve the denoising accuracy of NIR spectrum based on EEMD of normalized index optimization in low signal-noise ratio circumstance, the “3R” denoising method on NIR spectrum is proposed. For near infrared transmission spectrum and absorption spectrum of the same sample, correlations between suspicious noise IMF and transmission spectrum, absorption spectrum have been calculated, and correlation between IMFs of transmission spectrum and absorption spectrum and the self-correlation of IMF component are also computed, then three sets of correlations would show that which IMF is noise component, and noise IMFs would be removed. This method is applied to modeling of measuring the leaf chlorophyll content of corn and the result shows: it is superior to Empirical Mode Decomposition(EMD) low pass filtering, wavelet decomposition and EEMD method on denoising accuracyon.Thirdly, to solve spetral slant caused by baseline drift, a kind of residue related baseline correction algorithm on EEMD has been proposed, which realizes adaptive baseline correction of Raman spectrum and NIR spectrum. The simulations of NIR spectrum show that the method is better than the wavelet correction but worse than the global fitting of polynomial, and familiar with nonlinear least square fitting method. The experimental study of Raman spectral baseline correction indicates, the model corrected by the method has the best index of correction set and prediction set, and it affects characteristic peaks to the slightest degree. All above show that residue related baseline correction algorithm on EEMD has the strong adaptability and it has an optimized result of baseline correction.Finally, a method based on EMD time-frequency analysis of NIR spectrum is proposed in order to detect alkaline element qualitatively.This method, according to proportion and modulation way, explains the reason that modes overlapping occurs in Raman spectrum decomposed on EMD and NIR spectrum of corn leaves denoisedb by EMD. NIR spectra alkali stressed are processed by Hilbert Transform(HT) to get spectrum of wavelegth and frenquence, and then to obtain the total spectrum. It realizes the qualitative detection of alkali component by the characteristic frequency of the total spectrum; hence, the feasibility of NIR spectrum by EMD time-frequency qualitative analysis is verified.