Retro-regression residual remediation for spectral/signal identification

• US 20060190216A1
• Filed: 02/24/2005
• Published: 08/24/2006
• Est. Priority Date: 02/24/2005
• Status: Active Grant

• Current Assignee: Peraton, Inc.
• Original Assignee: ITT Manufacturing Enterprises, Inc. (ITT, Inc.)
• Inventors: Boysworth, Marc Kenneth
• Granted Patent: US 7,127,372 B2
• Time in Patent Office: Days
• US Class Current: 702/179
• CPC Class Codes: G01J 3/28 Investigating the spectrum using colour filters G01J3/51
• CPC Class Codes: G01N 2021/6417 Spectrofluorimetric devices
• CPC Class Codes: G01N 21/31 Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry G01N21/72 takes precedence
• CPC Class Codes: G01N 21/33 using ultraviolet light G01N21/39 takes precedence
• CPC Class Codes: G01N 21/359 using near infrared light
• CPC Class Codes: G01N 21/65 Raman scattering
• CPC Class Codes: G01N 2201/1293 resolving multicomponent spectra

Abstract

An improved regression-based qualitative analysis algorithm useful when the mixture to be analyzed contains a compound not in the library spectra, a so-called unknown. A regression of a measured spectrum is computed against the library spectra. This regression is referred to as a “master” regression. Estimated mixing coefficients for an estimated spectrum are computed from the regression. Next, a residual error is computed between the estimated spectrum and the measured spectrum. Peaks in the residual error are identified that extend in a direction opposite to that of peaks in the measured spectrum. These peaks are referred to as “negative” peaks. A regression is performed on the peaks. This is referred to as a “retro-regression” to be distinguished from the master regression performed on the measured spectrum. Using information from the retro-regression, corrected mixing coefficients are computed and the process repeats.

Claims

1. A method for improving regression-based spectroscopic analysis, comprising:

a. computing a residual error between an estimated spectrum and a measured spectrum taken of a sample, wherein the estimated spectrum is derived from mixing coefficients for members of a library of spectra that are produced by computing a regression on the measured spectrum with the library;

b. identifying peaks in the residual error that extend in a direction opposite to that of peaks in the measured spectrum;

c. performing a regression on the peaks to produce retro-regression coefficients;

d. computing corrected mixing coefficients based on the retro-regression coefficients; and

e. repeating (a) through (d) with the corrected mixing coefficients computed in (d) being used to generate a new estimated spectrum for computing the residual error in (a) at the next iteration.

2. The method of claim 1, and further comprising (f)(1) removing a member of the library whose mixing coefficient in the corrected mixing coefficients is less than a threshold.

3. The method of claim 2, wherein when a member of the library spectra is removed, further comprising (f)(2) performing a regression of the measured spectrum with the library without the member that was removed to produce new mixing coefficients, and wherein (e) of repeating comprises repeating (a) through (f)(2) the new mixing coefficients being used to generate a new estimated spectrum for computing the residual error at the next iteration.

4. The method of claim 3, wherein (e) repeating comprises repeating (a) through (f)(2) until no members of the library remain.

5. The method of claim 1, wherein (h) repeating comprises repeating (a) through (d) until there are no more peaks in the residual error.

6. The method of claim 1, wherein (b) identifying comprises identifying a peak as at least a predetermined number of contiguous points that are on the opposite side, with respect to a baseline, to peaks in the measured spectrum.

7. The method of claim 6, wherein (e) repeating comprises repeating (a) through (d) until there are no more regions in the residual error having at least the predetermined number of contiguous points.

8. The method of claim 1, wherein (d) computing corrected mixing coefficients comprises subtracting the retro-regression coefficients from the mixing coefficients for the estimated spectrum used in computing the residual error in the current iteration.

9. A processor readable medium storing instructions that, when executed by a processor, cause the processor to:

a. compute a residual error between an estimated spectrum and a measured spectrum taken of a sample, wherein the estimated spectrum is derived from a mixing coefficients for members of a library of spectra that are produced by computing a regression on the measured spectrum with the library;

b. identify peaks in the residual error that extend in a direction opposite to that of peaks in the measured spectrum;

c. perform a regression on the peaks to produce retro-regression coefficients;

d. compute corrected mixing coefficients based on the retro-regression coefficients; and

e. repeat (a) through (d) with the corrected mixing coefficients computed in (d) being used to generate a new estimated spectrum for computing the residual error in (a) at the next iteration.

10. The processor readable medium of claim 9, and further comprising instructions stored on the medium that, when executed, cause the processor to (f)(1) remove a member of the library whose mixing coefficient in the corrected mixing coefficients is less than a threshold.

11. The processor readable medium of claim 10, and further comprising instructions stored on the medium that, when executed, cause the processor to (f)(2) perform a regression of the measured spectrum with the library without the member that was removed to produce new mixing coefficients, and that cause the processor to (e) repeat (a) through (f)(2) are repeated with the new mixing coefficients being used to generate a new estimated spectrum for computing the residual error at the next iteration.

12. The processor readable medium of claim 11, and further comprising instructions stored on the medium that, when executed, cause the processor to repeat (a) through (f)(2) until no members of the library remain.

13. The processor readable medium of claim 9, and further comprising instructions stored on the medium that, when executed, cause the processor to repeat (a) through (d) until there are no more peaks in the residual error.

14. The processor readable medium of claim 9, and further comprising instructions stored on the medium that, when executed, cause the processor to identify a peak as at least a predetermined number of contiguous points that are on the opposite side, with respect to a baseline, to peaks in the measured spectrum.

15. The processor readable medium of claim 9, and further comprising instructions stored on the medium that, when executed, cause the processor to repeat (a) through (d) until there are no more regions in the residual error having at least the predetermined number of contiguous points.

16. The processor readable medium of claim 9, and further comprising instructions stored on the medium that, when executed, cause the processor to (d) compute corrected mixing coefficients by subtracting the retro-regression coefficients from the mixing coefficients for the estimated spectrum used in computing the residual error in the current iteration.

17. A system for spectroscopic analysis comprising:

a. a sensor that produces data from a mixture to be analyzed; and

b. a processor coupled to the sensor, wherein the processor is programmed to;

i. generate a measured spectrum from the data produced by the sensor;

ii. perform a regression of the measured spectrum with a library of spectra;

iii. generate estimated mixing coefficients from the regression;

iv. compute a residual error between an estimated spectrum generated from the estimated mixing coefficients and the measured spectrum;

v. identify peaks in the residual error that extend in a direction opposite to that of peaks in the measured spectrum;

vi. perform a regression on the peaks to produce retro-regression coefficients;

vii. compute corrected mixing coefficients based on the retro-regression coefficients; and

viii. repeat (iv) through (vii) with the corrected mixing coefficients computed in (vii) being used to generate a new estimated spectrum for computing the residual error in (iv) at the next iteration.

18. The system of claim 17, wherein the processor is programmed to remove a member of the library whose mixing coefficient in the corrected mixing coefficients is less than a threshold.

19. The system of claim 18, wherein the processor is programmed to perform a regression of the measured spectrum with the library without the member that was removed to produce new mixing coefficients, and repeating (iv) through (vii) with the new mixing coefficients being used to generate a new estimated spectrum for computing the residual error at the next iteration.

20. The system of claim 17, wherein the processor is programmed to compute the corrected mixing coefficients by subtracting the retro-regression coefficients from the mixing coefficients for the estimated spectrum used in computing the residual error in the current iteration.

21. A method for processing spectroscopic measured data of a sample, comprising:

a. identifying peaks in a residual error between measured data and an estimated data computed from a regression performed on the measured data, wherein the peaks extend in a direction opposite to that of peaks in the measured data;

b. performing a regression on the peaks;

c. computing a corrected data based on the regression of the peaks;

d. computing a new residual error between the measured data and the corrected data; and

e. repeating (a) through (d) using the new residual error.

22. The method of claim 21, wherein (a) identifying comprises identifying a peak as at least a predetermined number of contiguous points that are on the opposite side, with respect to a baseline, to peaks in the measured signal.

23. The method of claim 22, wherein (e) repeating comprises repeating (a) through (d) until there are no more regions in the residual error having at least the predetermined number of contiguous points.

24. The method of claim 21, wherein (c) computing comprises subtracting the retro-regression coefficients from mixing coefficients for the estimated data.