Method and arrangement for calibration of input data

• US 6,754,543 B1
• Filed: 03/14/2001
• Issued: 06/22/2004
• Est. Priority Date: 06/22/1998
• Status: Expired due to Fees

• Current Assignee: Umetri AB
• Original Assignee: Umetri AB
• Inventors: Wold, Svante
• Primary Examiner(s): Patel, Ramesh
• Assistant Examiner(s): PEREZ DAPLE, AARON C
• Application Number:
  US09/720,425
• Time in Patent Office: 1,196 Days
• Field of Search: 356/451, 356/300, 702/90, 702/190, 703/11, 700/55, 700/54, 700/52, 700/73, 700/29, 708/801, 708/814
• US Class Current: 700/52
• CPC Class Codes: G01N 21/359 using near infrared light
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04

Abstract

NIR spectra are often pre-processed in order to remove systematic noise such as base-line varation and multiplicative scauer effects. This is done by differentiating the spectra to first or second derivatives, by Mtltiplicative Siga Correction (MSC), or by sirnilar filtering methods. This pre-processinp may, however, also remove infosmaion fror The spectra regarding Y (the measured response variable in multivarate calibration applications). Provided is a variant ofPLS that can be used to achieve a signal correction that is as close to orthogonal as possible to a given Y-vector or Y-matrix (430). Thus, ensung that the signal correction removes as little information as possible regarding Y. In the case when the number of X-variables, in an X matrix, K exceeds the number of observations N, strict orthogonality is obtined. A filter (470) according to the present invention is named Orthogonal Signal Correction (OSC).

Claims

1. A computer-implementcd method for concentration or property calibration of input data samples from samples of substances or matter to determine a filter model for further samples of the same substance or matter, comprising:

optionally trnsforming, centermg and scaling the input data;

arranging said input data in an input set matrix X;

determining a concentration or property set matrix Y;

determining a loading set

determining with a computer a score set matrix T with a non-standard PLS algorithm having weights w, wherein weights w are defined by the formula;

2. A method according to claim 1, wherein said input data sets are arranged as two matrixes or one matrix and one concentration of property vector.

3. A method according to claim 1 or 2, wherein said filtering is bilinear and combined with a second linear filtering method.

4. A method according claim 3, wherein said second filtering method is one of wavelet-filtering or Fourier-filteg.

5. A method according to claim 1 wherein said filter model is improved by applying multiple sets of input data as training sets and repeating said steps with a better concentraion or property, thus tuning the filter model.

6. An arragement for concentration or property calibration of spectroscopic input data from samples of substance or matter, to determine a filter means for further spectra of samples of the same type, comprising:

a transforming means, a centering meas, and a scale means to optionally operate on the input data in order to provide two start sets,

arrangement means for arranging said input data in an input set matrix X;

determining means for determining, from an input set and a concentration or property calibration set matrix Y, a score set matrix T and a loading set;

calculation means for determining the product between the score set matrix and the loading set, said product resembling the input set matrix as much as possible under the constraint that the score set matrix is orthogonal to the concentration or property set matrix;

filter means for filtering said spectroscopic input data by subtracting said product from the input set matrix in order to remove variations relating to properties other than present calibration properties thereby determining said filter model;

whereby

said filter model provides that further samples of the same type can be filtered with the filter model and wherein the score set matrix T is determined with a non-standard PLS algorithm having weights w, wherein weigts w are defined by the formula;

7. An arrangement according to claim 6, wherein said input data sets are arranged as two matrixes or one matrix and one concentration or property vector.

8. An arrangement according to claim 6 or 7, where said filtering means performs bilinear filtering and is combined with a second linear filtering means.

9. An arrangement according to claim 8, wherein said second filtering means is one that provides wavelet-filtering or Fourier-filtering.

10. An arrangement according to claim 6 wherein said filter model is improved by applying multiple sets of input data as training sets, thus providing better determined properties by tuning the filter model.