Multiplicative signal correction method and apparatus

• US 5,568,400 A
• Filed: 08/28/1990
• Issued: 10/22/1996
• Est. Priority Date: 09/01/1989
• Status: Expired due to Term

• Current Assignee: Harald Martens, Stark, Edward W.
• Original Assignee: Harald Martens, Stark, Edward W.
• Inventors: Martens, Harald, Stark, Edward W.
• Primary Examiner(s): ZANELLI, MICHAEL J
• Application Number:
  US07/572,534
• Time in Patent Office: 2,247 Days
• Field of Search: 364/498, 364/570, 364/571.01, 364/571.02, 364/571.04, 364/572, 364/573, 364/574, 73/1 R, 356/326
• US Class Current: 702/85
• CPC Class Codes: G01J 2003/1828 with order sorter or prefilter
• CPC Class Codes: G01J 3/28 Investigating the spectrum using colour filters G01J3/51
• CPC Class Codes: G01N 2201/129 Using chemometrical methods

Abstract

An improved method and apparatus are disclosed for processing spectral data to remove undesired variations in such data and to remove interfering information present in the data. The method land apparatus corrects multiplicative effects present in the spectral data. Additive and interferent contributions can be corrected as well. In one aspect of the method, coefficients for a selected appropriate model are applied to the input spectral data based on first and second reference spectra. The spectral data are then corrected based on the estimated coefficients at least as to multiplicative errors for producing a linear additive structure for use in calibration, validation and determination by linear multivariate analysis. The method and apparatus will improve the accuracy of spectral data structures derived from measurements Using spectroscopy, chromatography, thermal analysis, mechanical vibration and acoustic analysis, rheology, electrophoresis, image analysis and other analytical technologies producing data of similar multivariate nature.

Claims

1. A method for correcting signals representing input spectral data (X_ki) derived from a sensor during a measurement, at least as to multiplicative errors, said method comprising the steps of:

providing a first and primary reference spectrum signal (P_ko) representing a predetermined standard for such data;

providing at least one second reference spectrum signal (P_ka or P_kj);

estimating coefficients for a selected appropriate model to be applied to said input data based on said first and second reference spectra signals; and

correcting said signals representing said spectral data based on said estimated coefficients at least as to multiplicative errors for producing signals representing a linear additive structure for use in calibration, validation and determination by linear multivariate analysis.

2. The method as in claim 1 wherein said at least one second reference spectrum signal represents expected influence of analytes of interest on input data (P_ka).

3. The method as in claim 1 wherein said at least one second reference spectrum signal represents expected influence of undesired interferences (Pjk).

4. The method as in claim 1 wherein said at least one second reference spectrum signal represents expected influence of analytes of interest on the input data (P_ka) and said method also includes the step of providing at least one third reference spectrum signal (P_kj) representing the expected influence of various undesired interferences, said estimating step being also based on said at least one third reference spectrum.

5. The method as in claim 1 or claim 2 or claim 3 or claim 4 also including the step of correcting said spectral data as to additive errors based on said estimated coefficients.

6. A method of claim 5 wherein said model includes a generalized least squares technique.

7. The method as in claim 1 wherein said model is a linear model.

8. The method of claim 6 wherein said model includes a maximum likelihood regression technique.

9. The method of claim 6 wherein said model includes an estimated best linear predictor technique.

10. The method of claim 6 wherein said model includes a principal component regression technique.

11. The method of claim 6 wherein said model includes a covariance adjustment technique.

12. The method of claim 1 wherein said model is a non-linear model.

13. The method of claim 12 wherein said model includes a Taylor expansion technique.

14. The method of claim 12 wherein said model includes a steepest descent method.

15. The method of claim 12 wherein said model includes a Marquardt'"'"'s compromise technique.

16. The method of claim 12 wherein said model includes a simplex optimization technique.

17. The method as in claim 1 or claim 3 or claim 4 including the steps of using the coefficients of at least one interfering component derived by the modeling to scale the spectra of the component and subtracting the scaled spectra from the data to remove their contribution from the data.

18. The method as in claim 17 including the steps of generating modified reference spectra of interfering components that contain only those portions of original reference spectra of the interfering components that are orthogonal to, and therefore uncorrelated with, at least one reference analyte spectrum, using the coefficients generated from said orthogonal reference spectrum to scale the original reference spectra prior to subtracting the scaled spectra from the data.

19. The method as in claim 18 including the step of further scaling the spectra of the interfering components to control a degree of spectral modification and correction applied to the data.

20. The method as in claim 17 including the step of further scaling the spectra of analyte data.

21. The method as in claim 1 or claim 2 or claim 3 or claim 4 or claim `7 also including the step of updating of standard (P_ka), analyte P_ka, or interference P_kj spectra based on results of later stages of data processing and analysis.

22. The method as in claim 21 wherein said later stages of data processing and analysis include principal components analysis (PCA).

23. The method as in claim 21 wherein said later stages of data processing and analysis include a technique of partial least squares (PLS).

24. The method as in claim 1 or claim 2 or claim 3 or claim 4 or claim 17 including the step of interactively displaying graphical output concerning which analytes and interference spectra, if any, are causing difficulties with respect to estimation of the multiplicative correction and interactive control over which reference spectra are utilized, a spectral range included in estimating the coefficients and weighing of additive corrections employed.

25. The method as in claim 24 including the step of further scaling the spectra of interfering components to control a degree of spectral modification and correction applied to the data.

26. The method as in claim 24 also including the step of further scaling the spectra of analyte data.

27. The method as in claim 1 or claim 2 or claim 3 or claim 4 or claim 17 also including the step of updating of standard P_ka, analyte P_ka, and signal P_kj spectra based on results of later stages of data processing and analysis.

28. Apparatus for correcting input spectral data (X_ki) derived from a sensor during a measurement, at least as to multiplicative errors, said apparatus comprising:

input means for supplying a signal representing spectral data subject to correction for at least multiplicative errors;

means for supplying a signal representing a first and primary reference spectrum (P_ko) as a predetermined standard for such data;

means for supplying a signal representing at least one second reference spectrum (P_ka or P_kj);

means for estimating coefficients for a selected model for application to said input spectral data, said input spectral data signal and first and second reference spectrum signals being supplied to said estimating means; and

means responsive to said estimating means for correcting said spectral data based on estimated coefficients at least as to multiplicative errors for producing a signal representing a linear additive structure for use in calibration, validation and determination by linear multivariate analysis.

29. Apparatus as in claim 28 wherein said means for supplying said signal representing at least one second reference spectrum supplies a signal which represents at least one spectrum representing expected influence of analytes of the interest on input data (P_ka).

30. Apparatus as in claim 28 wherein said means for supplying said signal representing at least one second reference spectrum supplies a signal which represents at least one spectrum representing expected influence of various undesired interferences (P_jk).

31. Apparatus as in claim 28 wherein said means for supplying said signal representing at least one second reference spectrum supplies a signal which represents at least one spectrum representing expected influence of analytes of interest of input data and also including means for supplying a signal of at least one third reference spectrum representing expected influence of various undesired interferences, said third reference spectrum signal also being supplied to said estimating means.

32. Apparatus as in claim 28 or claim 29 or claim 30 or claim 31 including means responsive to said estimating means for correcting said spectral data as to additive errors based on said estimated coefficients.

33. Apparatus as in claim 28 or claim 29 or claim 30 or claim 31 wherein said estimating means estimates coefficients for a linear model.

34. Apparatus as in claim 28 or claim 29 or claim 30 or claim 31 wherein said estimating means estimates coefficients for a non-linear model.

35. In a system for analyzing a medium, said system having a spectrophotometric sensor for providing a signal representing input spectral data (X_ki), the improvement comprising:

apparatus for correcting said spectral data for at least multiplicative errors, said apparatus including;

means for supplying a signal representing a first and primary reference spectrum (P_ko) as a predetermined standard for said spectral data;

means for supplying a signal representing at least one second reference spectrum (P_ka or P_kj);

means for estimating coefficients for a selected model for application to said input spectral data, said input spectral signals and first and second reference spectral signals being supplied to said estimating means; and

means responsive to said estimating means for correcting said spectral data based on estimated coefficients at least as to multiplicative errors for producing a signal representing a linear additive structure for use in calibration, validation and determination based on linear multivariate analysis.

36. The method for correcting input spectral data (X_ki) derived from a measurement, by applying an additive model and multiplicative model for the data in a sequential fashion, comprising the steps of:

1) obtaining a set of spectral data Z from original input spectral data corrected for multiplicative effects by using a standard multiplicative scatter correction technique;

2) estimating an additive model which takes into account spectral variations of analytes and interferences;

3) reconstructing spectral data Y without the estimated additive effects;

4) estimating the multiplicative effects on Y;

5) constructing a new matrix of corrected spectra Z from X; and

repeating steps 2, 3 and 4 until convergence occurs.

37. The method of claim 36 wherein the additive effects are obtained from a different model than the multiplicative effects.

38. The method of claim 37 wherein one model is a physical model and the other model is a chemical model.

39. The method of claim 36 wherein step 4 includes the steps of:

providing a first and primary reference spectrum;

providing at least one second reference spectrum;

estimating coefficients for a selected appropriate model to be applied to the input data based on said first and second reference spectra; and

correcting said spectral data Y based on said estimated coefficients.

44. The method of claim 36 or claim 39 or claim 41 including the further step of using said corrected input spectral data or said corrected scores for calibration of measuring equipment, prediction and determination procedures.

45. The method of claim 44 wherein the prediction procedure, using corrected spectral data, includes the steps of:

using an independent set of data and applying factor loadings P to find a new set of spectral scores T; and

determining the additive and multiplicative effects from the new spectral scores and using the mean quantities of additive and multiplicative effects and those portions of the interferents and analytes which had been used in a prior calibration to obtain a corrected spectra; and

using the corrected data and the calibration model coefficients from the linear model coefficients from the linear model to predict properties of interest.

46. The method of claim 44 wherein the prediction procedure, using corrected spectral scores, includes the steps of:

using an independent set of data and applying factor loadings P to find a new set of spectral scores T; and

determining an S^-1 factor related to a matrix containing T*d for multiplicative effects from the new spectral scores T and using a factor U from calibration relating to additive and interferent effects to find

W=S^-1 *T*U;

where W is the matrix of the additive, multiplicative and interferent corrected scores which can be used as regressors in additive mixture models;

using the corrected data and calibration model coefficients from the linear model to predict properties of interest.

40. The method for correcting input spectral data (X_ki) derived from a measurement or for correcting the scores obtained by bilinear modeling of such data, said method comprising the steps of:

examining the measured spectral data as a combined additive and multiplicative structure such that

space="preserve" listing-type="equation">X=T*P'"'"'+E

where X=(x_ik) is the matrix of spectral ordinates for i=1,2 . . . N objects, k=1,2 . . . , K wavelengths, T=(t_il) is the matrix of scores for objects i, l=1,2 . . . L representing bilinear factors Obtained from a bilinear model, P=(P_ki) are the loadings for objects i on bilinear factors l, and E=(e_ik) are the residuals between data X and model T*P'"'"';

decomposing the loadings into a function of a reference spectral and, optionally, a matrix of spectral components for analyte and interference phenomena;

obtaining the additive and multiplicative effects for the input spectra from the coefficients from the loading decomposition and scores; and

correcting said input spectra based on said obtained additive and multiplicative effects.

41. The method of claim 40 including the step of correcting said input spectral data simultaneously for interferent contributions and additive and multiplicative effects.

42. The method for correcting input spectral data (X_ki) derived from a measurement or for correcting the scores obtained by bilinear modeling of such data, said method comprising the steps of:

examining the measured spectral data as a combined additive and multiplicative structure such that

space="preserve" listing-type="equation">X=T*P'"'"'+E

where X=(x_ik) is the matrix of spectral ordinates for i=1,2 . . . N objects, k=1,2 . . . , K wavelengths, T=(t_il) is the matrix of scores for objects i, l=1,2 . . . L representing bilinear factors obtained from a bilinear model, P=(P_ki) are the loadings for objects i on bilinear factors l, and E=(e_ik) are the residuals between data X and model T*P'"'"';

decomposing the loadings into a function of a reference spectrum and, optionally, a matrix of spectral components for analyte and interference phenomena;

obtaining the additive effects for the scores T from the offset corrected loadings and the scores T; and

applying the obtained additive effects to the scores;

obtaining the multiplicative effect from the coefficient d from the reference spectrum; and

obtaining scores corrected for additive and multiplicative effects using the multiplicative effect, or scores T and, optionally, the additive effect.

43. The method of claim 42 including the steps of additionally correcting the input spectral data for multiplicative effects.

47. Apparatus for correcting an input spectral data (X_ki) signal derived from a measurement, particularly as to multiplicative errors, by the fitting of an additive model and multiplicative model in sequential fashion, comprising:

first means for obtaining a signal representing a set of spectral data Z from the original input spectral data signal corrected for multiplicative effect by using a standard multiplicative scatter correction technique;

second means for providing a signal representing estimate of an additive model which takes into account spectral variations of analytes and interferences;

third means for providing a signal representing the reconstructing of spectral data Y without the estimated additive effects;

fourth means for providing a signal representing an estimate of the multiplicative effects on Y;

fifth means for providing a signal representing the construction of a new matrix of corrected spectra Z from X;

means for respectively providing said signal from said fifth means to said second, third and fourth means; and

means responsive to the output signals of said second, third and fourth means for determining when convergence occurs.

48. Apparatus for correcting an input spectral data (X_ki) signal derived from a measurement or for correcting a signal representing the scores obtained by bilinear modeling of such data, comprising:

means for providing a signal responsive to said input signal representing the measured spectral data as a combined additive and multiplicative structure such that

space="preserve" listing-type="equation">X=T*P'"'"'+E

where X=(x_ik) is the matrix of spectral ordinates for i=1,z . . . N objects, k=1,2 . . . , k wavelengths, T=(t_il) is the matrix of scores for objects i, l=1,2 . . . L representing bilinear factors obtained from a bilinear model, P=(P_ki) are the loadings for objects i on bilinear factors l and E=(e_ik) are the residuals between data X and model T*P'"'"';

means for providing a signal representing decomposed loadings which have been decomposed into a function of a reference spectra and, optionally, a matrix of spectral components for analyte and interference phenomena;

means for providing a signal representing the additive and multiplicative effects for the input spectra obtained from the coefficients from the loading decomposition and scores; and

means for providing a signal representing the correction of said input spectra from said signal representing the additive and multiplicative effects.

49. Apparatus for correcting an input spectral data (X_ki) signal derived from a measurement or for correcting a signal representing the scores obtained by linear modeling of such data, comprising:

means for providing a signal responsive to said input signal representing the measured spectral data as a combined additive and multiplicative structure such that

space="preserve" listing-type="equation">X=T*P'"'"'+E

where X=(x_ik) is the matrix of spectral ordinates for i=1,z . . . N objects, k=1,2 . . . , k wavelengths, T=(t_il) is the matrix of scores for objects i, l=1,2 . . . L representing bilinear factors obtained from a bilinear model, P=(P_ki) are the loadings for objects i on bilinear factors l and E=(e_ik) are the residuals between data X and model T*P'"'"';

means for providing a signal representing decomposed loadings which have been decomposed into a function of a reference spectra and, optionally, a matrix of spectral components for analyte and interference phenomena;

means for providing a signal representing the additive effects for the scores T obtained from the offset corrected loadings and the scores T;

means responsive to said signal representing the additive effects for providing a signal representing the application of the additive effects to the multivariate scores;

means for providing a signal representing the multiplicative effect from the coefficient d from the reference spectrum; and

means for providing a signal representing scores corrected for additive and multiplicative effects using the multiplicative effect, or scores T and, optionally the additive effect.