CN100449282C - Method and device for separating noise signal from infrared spectrum signal by independent vector analysis - Google Patents

Abstract

The present invention relates to an infrared spectral analysis method. The method comprises the following steps: firstly a preprocessor is adopted, centralizing processing and whitening processing are carried out to original signals which a spectrograph primary and secondary collected, and a pretreatment array with two vectors W1, W2 is output; then, an independent component analyzer is arranged for receiving the two vectors W1, W2 which are output by the preprocessor, and two coefficient vectors W1(n), W2 (n) at present are output; a spectral signal selector determines noise signals and useful signals by analyzing the energy of the two vectors W1, W2. The present invention has the advantages that the useful signals and the noise signals can be separated from infrared spectral signals obtained from the infrared spectrograph in a short time through independent component analysis to improve the accuracy of the infrared spectrograph.

Description

Infrared spectrum denoising method and device based on isolated component

Affiliated technical field

The present invention relates to a kind of infrared spectrum analysis.Refer in particular to a kind of infrared spectrum denoising method and device based on isolated component.

Background technology

Both contained useful information in the spectral signal that is obtained by spectrometer, stochastic error (noise) is also superposeing simultaneously.In order to reduce the influence of noise, normally pass through the final spectrum of the mean value of repeatedly sampling as sample to the spectrometer precision.Like this, there is important systematic parameter one sampling number to need to select in each spectrometer.Yet sampling number is selected needs certain experience, if sampling number is very few, influencing at random when spectrum is sampled is big, and spectral accuracy reduces; Sampling number is too much, and the spectrum that obtains after averaging is with the prolongation in sampling time, and the systematic error (as the drift of instrument) that instrument produces also increases thereupon, so not only sacrificed the time, and the precision of spectrum also is difficult to guarantee.

Summary of the invention

Invention provides a kind of infrared spectrum denoising method and device based on two-dimentional neuron isolated component algorithm, isolates useful signal and noise signal in the short period of time from the infrared spectroscopy signals that spectrometer obtains.These apparatus and method are the hybrid processing of the infrared spectrum difference of twice collection of same target being come estimated signal according to spectrometer.

The device of realizing technique scheme comprises: through data-signal continuous successively spectrometer, pretreater, independent component analysis device, spectral signal selector switch.

Infrared and near infrared spectrum during described infrared spectrum comprises, promptly wavelength coverage is: 780nm～50000nm;

Described pretreater take centralization and albefaction to spectrometer for the first time, the original signal of gathering for the second time handles.

Described independent component analysis device uses independent component analysis method to determine current two coefficient vectors of output

Described spectral signal selector switch adopts energy spectrometer to determine noise signal and useful signal.

The method of isolating useful signal and noise signal from infrared spectroscopy signals of the present invention may further comprise the steps:

(1) at first at the pretreater place, to spectrometer for the first time, the original signal of gathering for the second time carries out centralization and albefaction is handled, and exports a pre-service array that two vectors are arranged

(2) the independent component analysis device is set then and is used for accepting pretreater output And export current two coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\};$

(3) the spectral signal selector switch is by right Carry out energy spectrometer and determine noise signal and useful signal.

Described independent component analysis device is to determine described current two coefficient vectors according to following formula $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\}$

w(n)＝w(n-1)+(I-2Bexp(-B ²/2)B ^T)w(n-1)

Wherein, w (n) is included as two current coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\},$ Two coefficient vectors that once calculate before w (n-1) comprises $\{{\stackrel{\→}{w}}_1(n-1),{\stackrel{\→}{w}}_2(n-1)\};$ I is 2 * 1 unit matrixs, and B is for comprising described pre-service array

2 * 1 column matrix, B ^TBe the transposition of B, be row matrix.

Effect of the present invention is can isolate useful signal and noise signal in the short period of time by independent component analysis from the infrared spectroscopy signals that spectrometer obtains, the precision of raising infrared spectrometer.

Description of drawings

Fig. 1 infrared spectrum burbling noise schematic representation of apparatus;

The process flow diagram of Fig. 2 independent component analysis.

The effect of Fig. 3 centering ir data denoising

Fig. 4 is to the effect of near infrared spectrum data denoising

Embodiment

Below in conjunction with accompanying drawing the present invention is described in further detail.

Fig. 1 has shown according to the preferred embodiment of the present invention, has been used for the schematic representation of apparatus of noise in the separate red external spectrum.This device comprises pretreater, independent component analysis device, spectral signal selector switch,

Pretreater to spectrometer for the first time, the original signal of gathering for the second time carries out centralization and albefaction is handled, and exports a pre-service array that two vectors are arranged

The independent component analysis device is used for accepting pretreater output

And export current two coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\},$ These two coefficient vectors use independent component analysis method to calculate, and n represents the current iteration number of times of independent component analysis method.

Be described in detail this isolated component method below and how infrared spectrometer gathered that noise separation in the infrared spectroscopy signals that reaches comes out.Allow useful infrared signal and the independence between the noise signal maximize, promptly useful infrared signal and noise signal return to its mixed virgin state, and this mixed signal is exactly the original spectrum signal that spectrometer collects.

The spectral signal selector switch is by right

Carry out energy spectrometer and determine noise signal and useful signal, obtain the useful infrared spectroscopy signals that noise separation is gone out.

Fig. 2 is the process flow diagram of independent component analysis among the present invention, and this process flow diagram is two-dimentional neuron isolated component algorithm, and this algorithm can be finished by the independent component analysis device among Fig. 1.

Independent component analysis method among Fig. 2 is controlled current two coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\},$ This independent component analysis method is by the signal that comprises pretreater output among Fig. 1

The nonlinear function (Bexp (B of matrix B ²/ 2)) realize, shown in following formula (1).

w(n)＝w(n-1)+(I-2Bexp(-B ²/2)B ^T)w(n-1)

Wherein, w (n) is included as two current coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\},$ Two coefficient vectors that once calculate before w (n-1) comprises $\{{\stackrel{\→}{w}}_1(n-1),{\stackrel{\→}{w}}_2(n-1)\};$ I is 2 * 1 unit matrixs, and B is for comprising described pre-service array

2 * 1 column matrix, B ^TBe the transposition of B, be row matrix.

When opening infrared spectrum burbling noise device, at first at the pretreater place, to spectrometer for the first time, the original signal of gathering for the second time carries out centralization and albefaction is handled, and exports a pre-service array that two vectors are arranged

Then, the independent component analysis device analytic process among Fig. 1 is carried out initialization earlier as shown in Figure 2, produces at random $w\left(0\right)=\{{\stackrel{\→}{w}}_1\left(0\right),{\stackrel{\→}{w}}_2\left(0\right)\},$ N=1 produces I and B in the formula (1) simultaneously, and the independent component analysis device among Fig. 1 calculates above formula (1) then, exports two current coefficient vectors $\{{\stackrel{\→}{w}}_1\left(n\right),{\stackrel{\→}{w}}_2\left(n\right)\};$ Determine whether two coefficient vectors of isolated component output restrain, on same direction, promptly their dot product equals 1 or be approximately equal to 1 to the new value that " convergence " here refers to w with old value.If restrain current coefficient is final isolated component coefficient, if do not restrain then n is added 1, and then carries out formula (1) computing, till convergence.

The independent component analysis method of Fig. 2 can carry out in short convergence time.Therefore, when the infrared spectrum burbling noise device of Fig. 1 is installed on the infrared spectrometer data acquisition system (DAS) and by the estimated useful infrared spectroscopy signals output of this independent component analysis method, the user can obtain low noise, high-precision infrared spectrum at short notice.

Fig. 3 is the effect with apparatus and method centering ir data of the present invention denoising, and Fig. 4 is with the effect of apparatus and method of the present invention to the near infrared spectrum data denoising.The random noise of instrument and ground unrest all have been separated as can be seen among the figure.

Claims (2)

1. based on the infrared spectrum denoising device of isolated component, it is characterized in that comprising spectrometer, pretreater, independent component analysis device, the spectral signal selector switch that links to each other successively through data-signal; Described pretreater take centralization and albefaction to spectrometer for the first time, the original signal of gathering for the second time handles; Described independent component analysis device uses independent component analysis method to determine current two coefficient vectors of output

Described spectral signal selector switch adopts energy spectrometer to determine noise signal and useful signal,

Described independent component analysis device is to determine described current two coefficient vectors according to following formula W (n)=w (n-1)+(I-2B exp (B ²/ ₂) B ^T) w (n-1), wherein, w (n) is included as two current coefficient vectors Two coefficient vectors that once calculate before w (n-1) comprises

N represents the current iteration number of times of independent component analysis method, and I is 2 * 1 unit matrixs, B be comprise to spectrometer for the first time, the original signal of gathering for the second time carries out centralization and albefaction is handled, and exports a pre-service array that two vectors are arranged

2 * 1 column matrix, B ^TBe the transposition of B, be row matrix; Produce at random $w\left(0\right)=\{{\stackrel{\→}{w}}_1\left(0\right),{\stackrel{\→}{w}}_2\left(0\right)\}$ N=1 produces I and B in the above-mentioned formula simultaneously, calculates above-mentioned formula then, exports two current coefficient vectors

Determine whether two coefficient vectors of isolated component output restrain, on same direction, promptly their dot product equals 1 or be approximately equal to 1 to the new value that " convergence " here refers to w (n) with old value; If restrain current coefficient is final isolated component coefficient, if do not restrain then n is added 1, and then carries out above-mentioned formula operation, till convergence.

2. based on the infrared spectrum denoising method of isolated component, it is characterized in that may further comprise the steps:

(1) at first at the pretreater place, to spectrometer for the first time, the original signal of gathering for the second time carries out centralization and albefaction is handled, and exports a pre-service array that two vectors are arranged

(2) the independent component analysis device is set then and is used for accepting pretreater output

And export current two coefficient vectors

(3) the spectral signal selector switch is by right (n) carry out energy spectrometer and determine noise signal and useful signal;

Described independent component analysis device is to determine described current two coefficient vectors according to following formula W (n)=w (n-1)+(I-2B exp (B ²/ 2) B ^T) w (n-1), wherein, w (n) is included as two current coefficient vectors

Two coefficient vectors that once calculate before w (n-1) comprises

N represents the current iteration number of times of independent component analysis method, and I is 2 * 1 unit matrixs, and B is for comprising described pre-service array

2 * 1 column matrix, B ^TBe the transposition of B, be row matrix; Produce at random $w\left(0\right)=\{{\stackrel{\→}{w}}_1\left(0\right),{\stackrel{\→}{w}}_2\left(0\right)\}$ N=1 produces I and B in the above-mentioned formula simultaneously, calculates above-mentioned formula then, exports two current coefficient vectors

Determine whether two coefficient vectors of isolated component output restrain, on same direction, promptly their dot product equals 1 or be approximately equal to 1 to the new value that " convergence " here refers to w (n) with old value; If restrain current coefficient is final isolated component coefficient, if do not restrain then n is added 1, and then carries out above-mentioned formula operation, till convergence.

Images