Method and system for contaminant detection during food processing

• US 6,587,575 B1
• Filed: 02/09/2001
• Issued: 07/01/2003
• Est. Priority Date: 02/09/2001
• Status: Expired due to Fees

• Current Assignee: Provision Technologies Division Institute For Technology Development, The University of Georgia Research Foundation, Inc.
• Original Assignee: Provision Technologies Division Institute For Technology Development, The University of Georgia Research Foundation, Inc., United States Department Of Agriculture
• Inventors: Smith, David A., Lawrence, Kurt C., Park, Bosoon, Martinez, Luis A., Lanoue, Mark A., Windham, William R., Heitschmidt, Jerry, Poole, Gavin H.
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): Choobin, Barry
• Application Number:
  US09/779,832
• Time in Patent Office: 872 Days
• Field of Search: 382/110, 250/458.1, 250/461.2, 250/339.09, 435/34
• US Class Current: 382/110
• CPC Class Codes: A22B 5/007 Non-invasive scanning of carcasses, e.g. using image recognition, tomography, X-rays, ultrasound analysing meat, e.g. search for foreign embedded objects G01N33/12; image analysis G06T7/00
• CPC Class Codes: G01N 2021/3155 Measuring in two spectral ranges, e.g. UV and visible
• CPC Class Codes: G01N 21/274 Calibration, base line adjustment, drift correction
• 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 21/94 Investigating contamination, e.g. dust G01N21/85 takes precedence
• CPC Class Codes: G01N 2201/1293 resolving multicomponent spectra
• CPC Class Codes: G01N 33/12 Meat; Fish

Abstract

Imaging systems, containing at least one charge-coupled device detector, are used for determining contamination of foodstuffs, such as for example, animal carcasses. Image processing algorithms allow for the identification of contaminants.

Claims

1. An imaging system for determination of contamination on food comprising:

at least one charge-coupled device detector with an optical filter capable of collecting at least two discrete narrow-band images,

a lighting system,

a data processing unit operatively connected to said detectors for receiving images for analysis of the spectral properties of an image created by said detector, and

a computer readable memory encoded with a computer program containing a detection algorithm based on mathematical analysis of selected key wavelengths of radiation detected by said detector wherein said selected key wavelengths are derived by using a calibration process including;

(a) collecting spectra with a visible/near infrared monochromator by irradiating samples of uncontaminated food and pure contaminants representative of the types of contamination to be determined with visible/near infrared radiation and digitally recording reflectance intensity from about 400 nm to about 2500 nm in about 2-nm intervals,

(b) transforming said spectra recorded in step (a) for each sample to log₁₀, spectra in absorbence units,

(c) transforming said log₁₀ spectra with standard normal variate and detrending procedures to remove interferences of scatter, particle size, and variations in baseline shift and curvilinearity,

(d) processing said transformed spectra in step (c) with at least one of Principal Component Analysis and Partial Least Squares regression for formation of scores and loadings,

(e) comparing said scores with variations in Principal Components for selecting discrete Principal Components at which scores correlate with uncontaminated foods and contaminants,

(f) evaluating loadings of said discrete Principal Components for extreme variations in absolute value to identify key wavelengths,

(g) selecting images at key wavelengths identified in step f, and

(h) calculating algorithm to detect contaminants.

2. The imaging system of claim 1 wherein said optical filter is selected from the group consisting of a line-scan spectrograph, a liquid crystal tunable filter, an acousto-optic tunable filter, and a narrow band-pass filter;

wherein said filters are capable of collecting at least two discrete spectral images each taken at a different wavelength.

3. A method for identifying contamination on food comprising:

(a) identifying key wavelengths by performing the following steps;

(i) preparing samples of uncontaminated and pure contaminants representative of the types of contamination to be determined,

(ii) collecting spectra of said samples,

(iii) transforming spectra of said samples to log₁₀ in absorbence units,

(iv) transforming said log₁₀ spectra with standard normal variate and detrending procedures to remove interferences of scatter, particle size, and variations in baseline shift and curvilinearity,

(v) processing said transformed spectra in step (iv) with at least one of Principle Component Analysis and Partial Least Squares regression for formation of scores and loadings,

(vi) comparing said scores with variations in Principal Components for selecting discrete Principal Components at which scores correlate with uncontaminated foods and contaminants,

(vii) evaluating loadings of said discrete principal components for extreme variations in absolute value for identifying key wavelengths,

(viii) identifying said key wavelengths based on the results of step (vii),

(b) calibrating image wavelengths wherein said calibration includes selecting sensor binning to determine band numbers, imaging known wavelength standards to identify wavelength peaks and band numbers, performing a non-linear regression on said wavelengths against said band numbers, and applying said regression to subsequent images,

(c) creating hyperspectral or multispectral images of said samples,

(d) selecting said images at said key wavelengths based on the results of step (viii),

(e) applying algorithms using key wavelengths identified in step (viii) to form an image dataset for the identification of contamination.

4. A method for identifying contamination on food comprising:

(a) identifying key wavelengths by performing the following steps;

(i) preparing samples of uncontaminated and pure contaminants representative of the types of contamination to be determined,

(ii) collecting spectra of said samples,

(iii) transforming spectra of said samples to log₁₀ in absorbence units,

(iv) transforming said log₁₀ spectra in absorbence units,

(v) processing said transformed spectra in step (iv) with at least one of Principle Component Analysis and Partial Least Squares regression for formation of scores and loadings,

(vi) comparing said scores with variations in Principal Components for selecting discrete Principal Components at which scores correlate with uncontaminated foods and contaminants,

(vii) evaluating loadings of said discrete principal components for extreme variations in absolute value for selecting key wavelengths,

(viii) identifying key wavelengths based on the results of step (vii),

(b) calibrating image wavelengths wherein said calibration includes selecting sensor binning to determine band numbers, imaging known wavelength standards to identify wavelength peaks and band numbers, performing a non-linear regression on said wavelengths against said band numbers, and applying said regression to subsequent images,

(c) creating hyperspectral or multispectral images of said samples,

(d) selecting said images at said key wavelengths based on the results of step (viii),

(e) calculating a ratio of two images at said two wavelengths to form a ratio image,

(f) performing a masking procedure on said ratio image to reduce background noise,

(g) applying histogram stretching to said ratio image to qualitatively identify contaminants in real-time, and/or

(h) applying thresholding to said ratio image from step f to quantitatively identify contaminants in real-time.

5. The method of claim 4 further including transforming said log10 spectra with standard normal variate and detrending procedures to remove interference of scatter, particle size, and variations in baseline shift and curvilinearity.

6. The method of claim 4 wherein hyperspectral images are collected by a line-scan spectrograph with a charge-coupled detector.

7. The method of claim 4 wherein multispectral images are collected using an imaging device selected from the group consisting of a common aperture camera with at least two charge-coupled device detectors, at least one charge-coupled device detector with a liquid crystal tunable filter, at least one charge-coupled device with an acousto-optic tunable filter, at least one charge-coupled device with a line-scan spectrograph, and multiple charge-coupled device detectors with narrow band-pass filters.

8. The method of claim 4 wherein said ratio image is determined by dividing an image at a first key wavelength by an image at a second key wavelength on a pixel by pixel basis of said collected images from step (4d).

9. A method comprising:

(a) preparing samples of uncontaminated food and pure contaminants representative of the types of contamination to be determined,

(b) collecting spectra of said samples,

(c) transforming spectra of said samples to log₁₀ spectra in absorbence units,

(d) processing said transformed spectra in step (c) with at least one of Principal Component Analysis and Partial Least Squares regression for formation of scores and loadings,

(e) comparing said scores with variations in Principal Components for selecting discrete Principal Components at which scores correlate with uncontaminated foods and contaminants,

(f) evaluating loadings of said discrete principal components for extreme variations in absolute value for selecting key wavelengths, and

(g) identifying key wavelengths for identification of contaminants based on the results of step (f).

10. A process for detecting contamination on food comprising:

(a) illuminating said food with a source of electromagnetic radiation having a predetermined spectral content,

(b) detecting radiation from said source reflected by said food item in each of four predetermined wavelengths λ

₁, λ

₂, λ

₃, λ

₄, and generating a first data set comprising digital values indicative of reflected radiation intensity in each of said wavelengths;

(c) processing said digital values according to an algorithm as follows $I=\frac{\left({\mathrm{\lambda <br><br>}}_1-n\right)<br><br>\text{\hspace{1em}<br><br>}<br><br>\left({\mathrm{\lambda <br><br>}}_3+{\mathrm{\lambda <br><br>}}_4\right)}{{\mathrm{\lambda <br><br>}}_3<br><br>\left({\mathrm{\lambda <br><br>}}_1+{\mathrm{\lambda <br><br>}}_2\right)<br><br>\text{\hspace{1em}<br><br>}}$

 

wherein n is a constant integer; and

I is an indication of fecal contamination.

11. The process of claim 10 wherein

λ ₁ is from about 750 to about 830 nm

λ

₂ is from about 450 to about 500 nm

λ

₃ is from about 500 to about 535 nm

λ

₄ is from about 550 to about 585 nm.

12. The process of claim 10 further comprising

calculating, for each digital value in said first data set, mean and variance values, based on a set of proximate digital values, thereby creating a mean value data set and a variance value data set;

adding said mean value and variance value data sets to create a final data set;

determining presence or absence of contamination based on values in said final data set.

13. The process of claim 12 wherein said determining step comprises:

comparing data values in said final data set to a predetermined threshold value; and

determining presence of contamination based on results of said contamination.

14. A process for detecting contamination on food comprising:

(a) illuminating said food with a source of electromagnetic radiation having a predetermined spectral content;

(b) detecting radiation from said source reflected by said food in a plurality of predetermined wavelengths and generating a data set comprising signals indicative of reflected radiation intensity in each of said wavelengths;

(c) processing said data set according to a predetermined mathematical function to generate a plurality of rule files comprising respective image files;

(d) combining said rule files to generate a combined data set;

(e) performing a texture analysis on said combined data set to generate spatially distributed mean and variance data;

(f) summing said mean and variance data to yield output data; and

(g) detecting contamination based on said output data.

15. The process of claim 14 wherein said predetermined mathematic function is defined by:

${\cos }^{-1}<br><br>\left(\frac{\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i<br><br>r_i}{{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i^2\right)}^{1/2}<br><br>{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>r_i^2\right)}^{1/2}}\right)$

wherein

nb=number of said predetermined wavelengths;

t=detected reflected radiation value from said food for a defined wavelength; and

<

r=reflectance value of a contaminant at said defined wavelength.

16. A method for determining contamination on poultry or livestock carcasses comprising:

(a) obtaining poultry or livestock carcasses for which contaminants are to be determined,

(b) creating hyperspectral or multispectral images of said carcasses,

(c) selecting images at key wavelength,

(d) applying an algorithm to detect contaminants from images selected in step (c),

(e) applying masking to reduce background noise in images of step (d),

(f) applying histogram stretching to images of step (e) to qualitatively identify contaminants, and

(g) applying thresholding to images of step (e) or (f) to quantitatively identify contaminants in real-time.

17. The method of claim 16 wherein said algorithms are selected from the group consisting of a ratio of key wavelengths and a linear combination of key wavelengths.

18. A computer readable medium encoded with a computer program for detecting contamination on food by causing a computer to process image signals indicative of intensity of radiation reflected from said food in four wavelengths λ

₁-λ

₄, using an algorithm;

$I=\frac{\left({\mathrm{\lambda <br><br>}}_1-n\right)<br><br>\text{\hspace{1em}<br><br>}<br><br>\left({\mathrm{\lambda <br><br>}}_3+{\mathrm{\lambda <br><br>}}_4\right)}{{\mathrm{\lambda <br><br>}}_3<br><br>\left({\mathrm{\lambda <br><br>}}_1+{\mathrm{\lambda <br><br>}}_2\right)<br><br>\text{\hspace{1em}<br><br>}}$

a wherein I is an indication of contamination.

19. An apparatus for detecting contamination on food comprising:

(a) a plurality of sensors for detecting spatially distributed radiation reflected from a food at four wavelengths λ

₁, λ

₂, λ

₃, and λ

₄, and generating image signals indicative thereof;

a computer; and

a computer readable medium coupled to said computer for causing said computer to process said image signals using an algorithm;

$I=\frac{\left({\mathrm{\lambda <br><br>}}_1-n\right)<br><br>\text{\hspace{1em}<br><br>}<br><br>\left({\mathrm{\lambda <br><br>}}_3+{\mathrm{\lambda <br><br>}}_4\right)}{{\mathrm{\lambda <br><br>}}_3<br><br>\left({\mathrm{\lambda <br><br>}}_1+{\mathrm{\lambda <br><br>}}_2\right)<br><br>\text{\hspace{1em}<br><br>}}$

 

wherein I is an indication of contamination.

20. A computer readable medium encoded with a computer program for detecting contamination on a food by causing a computer to process spectrally resolved image information indicative of intensity of radiation reflected from said food in a plurality of wavelength by performing the following steps:

(a) calculating data sets in the form of rule files, using said spectrally resolved image information, comprising values of α

wherein $\mathrm{\alpha <br><br>}={\cos }^{-1}<br><br>\left(\frac{\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i<br><br>r_i}{{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i^2\right)}^{1/2}<br><br>{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>r_i^2\right)}^{1/2}}\right)$

 

wherein

nb=number of bands of spectrally resolved image information;

t=detected spectrally resolved image information values for an i_th band; and

r=spectrally resolved image information value for an i_th band for a contaminant whose presence is to be detected.

(b) combining said rule files to generate a combined data set;

(c) performing a texture analysis on said combined data set to generate spatially distributed mean and variance data; and

(d) summing said mean and variance data to yield output data indicative of contamination.

21. An apparatus for detecting contamination on food comprising:

(a) a plurality of sensors for detecting spatially distributed spectrally resolved image information indicative of intensity radiation reflected from a food in a plurality of wavelength bands;

(b) a computer readable medium coupled to said computer for causing said computer to process said image signals by calculating data sets in the form of Rule Files, using said spectrally resolved image information, using a formula $\mathrm{\alpha <br><br>}={\cos }^{-1}<br><br>\left(\frac{\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i<br><br>r_i}{{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>t_i^2\right)}^{1/2}<br><br>{\left(\underset{i=1}{\overset{\mathrm{nb}}{\sum <br><br>}}<br><br>r_i^2\right)}^{1/2}}\right)$

 

wherein

nb=number of bands of spectrally resolved image information;

t=detected spectrally resolved image information values for an i_th band, and

r=spectrally resolved image information value for an i^th band for a contaminant whose presence is to be detected;

(c) combining said Rule Files to generate a combined data set;

(d) performing a texture analysis on said combined data set to generate spatially distributed mean and variance data; and

(e) summing said mean and variance data to yield output indicative of contamination.