The Application Of Empirical Mode Decomposition Method In Near Infrared Spectroscopy For Signal Denoising

With the rapid development of economy, the application of near infrared spectroscopyanalysis technology has increasing. However, in addition to information themselves, thespectral signals collected by near infrared spectroscopy contain many noise signals withindependence in measurement due to intervene of the environment. Therefore, whenestablishing a calibration model, the elimination of uncorrelated noises of spectral data inspectral data analysis is becoming extremely important and necessary.There are many ways to denoising NIR signals, such as smooth, Fourier transform, Wavelettransform. Although these methods have gained some effects, they still have their own limitsbecause of the nonlinearity and non-stationary signals. At present, Empirical ModeDecomposition (EMD) is the most popular and common method to eliminate noise in thenonlinearity and non-stationary signals. The paper also studies the application of EMD in NIRfor signal denoising. At first, this paper introduces the basic theory and the method ofelimination of noise of EMD. However, after the disintegration of EMD, the problems ofchoosing useful information in IMF component for elimination of noise and reestablishment ofsignal are not equipped with self-adaption of EMD in preferences. Therefore, on the basis ofcorrelative formula of modal, this paper come up with a method of elimination of noise ofself-adaption of EMD because of the principle of least energy, and conduct a desk study of thismethod in simulation experiment. the next, In order to test and verify the application of themethod in actual signal uses, the paper, take the maize seedling leaves chlorophyll content as anexample, The method is introduced into the maize seedling leaves based on near infraredspectroscopy for signal denoising, denoising the maize seedling leaves spectroscopy, analyzeand compare the method in wavelet denoising and EMD fusion wavelet denoising, and onceagain, establish a calibration model of near infrared spectroscopy on the basis of partial leastsquares regression, calculate the chlorophyll content and forecast the determination coefficientR2between predictive value and measured value is0.984, and Root Mean Square Error is0.075,and these values show the modal of well forecast effect and upper robustness.At last, this paper summarizes the EMD analysis on the basis of former effects, adoptsMatcom, achieves compound programming between Matlab and Visual C++, and compilessoftware for eliminating noise by one-dimensional NIR. And the major function of this softwareis to filtering on the basis of EMD.