Method for remote sensing analysis be decorrelation statistical analysis and hardware therefor

• US 6,018,587 A
• Filed: 07/01/1997
• Issued: 01/25/2000
• Est. Priority Date: 02/22/1991
• Status: Expired due to Fees

• Current Assignee: Applied Spectral Imaging Ltd.
• Original Assignee: Applied Spectral Imaging Ltd.
• Inventors: Cabib, Dario
• Primary Examiner(s): Johns, Andrew W.
• Application Number:
  US08/886,293
• Time in Patent Office: 938 Days
• Field of Search: 435/6, 364/724.11, 708/314, 382/165, 382/191, 382/209, 382/217, 382/218, 382/224, 382/129
• US Class Current: 382/165
• CPC Class Codes: G01J 2003/2866 Markers; Calibrating of scan
• CPC Class Codes: G01J 3/02 Details
• CPC Class Codes: G01J 3/12 Generating the spectrum; Monochromators
• CPC Class Codes: G01J 3/1256 using acousto-optic tunable filter; acousto-optic elements or systems G02F1/11, G02F1/33
• CPC Class Codes: G01J 3/26 using multiple reflection, e.g. Fabry-Perot interferometer, variable interference filters
• CPC Class Codes: G01J 3/2823 Imaging spectrometer
• CPC Class Codes: G01J 3/4406 Fluorescence spectrometry
• CPC Class Codes: G01J 3/453 by correlation of the amplitudes
• CPC Class Codes: G01N 21/6428 Measuring fluorescence of fluorescent products of reactions or of fluorochrome labelled reactive substances, e.g. measuring quenching effects, using measuring "optrodes" in vivo A61B5/00; immunoassay G01N33/53
• CPC Class Codes: G01N 21/6458 Fluorescence microscopy fluorescence microscopes per se G02B21/0076 and G02B21/16
• CPC Class Codes: G01N 33/5005 involving human or animal cells immunoassay G01N33/56966; immunoassays of protozoa G01N33/56905; protozoa in screening assays C12Q1/025
• CPC Class Codes: G01N 33/56966 Animal cells
• CPC Class Codes: G01N 33/582 with fluorescent label
• CPC Class Codes: G06V 10/88 Image or video recognition using optical means, e.g. reference filters, holographic masks, frequency domain filters or spatial domain filters
• CPC Class Codes: G06V 20/13 Satellite images
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: G06V 20/69 Microscopic objects, e.g. biological cells or cellular parts

Abstract

A method for remote scenes classification comprising the steps of (a) preparing a reference template for classification of the remote scenes via (i) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes; (ii) using a first spectral imager for measuring a spectral cube of the preclassified reference scenes; (iii) employing a principal component analysis for extracting the spectral cube for decorrelated spectral data characterizing the reference scenes; and (vi) using at least a part of the decorrelated spectral data for the preparation of the reference template for remote scenes classification; (b) using a second spectral imager for measuring a spectral cube of analyzed remote scenes, such that a spectrum of each pixel in the remote scenes is obtained; (c) employing a decorrelation statistical method for extracting decorrelated spectral data characterizing the pixels; and (d) comparing at least a part of the decorrelated spectral data extracted from the pixels of the remote scenes with the reference template.

Claims

1. A method for preparing a reference template for classification of remote scenes comprising the steps of:

(a) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified scenes;

(b) using a spectral imager for measuring a spectral cube of said preclassified referenced scenes;

(c) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(d) using at least a part of said decorrelated spectral data for the preparation of the reference template for remote scenes classification.

2. The method for preparing a reference template for classification of remote scenes as in claim 1, wherein said spectral imager includes an element selected from the group consisting of a dispersion element, a filter, a tunable filter and an interferometer.

3. The method for preparing a reference template for classification of remote scenes as in claim 1, wherein said decorrelation statistical method is selected from the group consisting of principal component analysis, canonical variable analysis and singular value decomposition.

4. The method for preparing a reference template for classification of remote scenes as in claim 3, wherein said principal component analysis includes expressing each of said scenes as linear combinations of N eigenvectors.

5. The method for preparing a reference template for classification of remote scenes as in claim 4, wherein N is an integer greater than two.

6. The method for preparing a reference template for classification of remote scenes as in claim 4, wherein N is an integer greater than two and smaller than eight.

7. The method for preparing a reference template for classification of remote scenes as in claim 3, wherein said principal component analysis includes the steps of:

(a) selecting k spectral slices for said spectral cube of said reference scenes;

(b) calculating an average spectrum for each of said reference scenes;

(c) stretching each of said average spectra for obtaining a stretched average spectrum for each of said reference scenes;

(d) averaging said stretched average spectra for each of said reference scenes, for obtaining an ensemble average spectrum for each of said reference scenes;

(e) calculating a k dimension eigen system for said ensemble average spectra and extracting N eigenvectors;

(f) using said N eigenvectors for defining an N-dimension vector for each of said reference scenes; and

(g) using said N-dimension vectors for preparing the reference template for the remote scenes classification.

8. The method for preparing a reference template for classification of remote scenes as in claim 7, wherein k is an integer greater than nine.

9. The method for preparing a reference template for classification of remote scenes as in claim 7, wherein N is an integer greater than two.

10. The method for preparing a reference template for classification of remote scenes as in claim 7, wherein said principal component analysis further includes the step of:

(h) performing a spatial averaging procedure over all spectral slices.

11. A method for remote scenes classification comprising the steps of:

(a) preparing a reference template for classification of the remote scenes via;

(i) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes;

(iii) employing a principal component analysis for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(vi) using at least a part of said decorrelated spectral data for the preparation of the reference template for remote scenes classification;

(b) using a second spectral imager for measuring a spectral cube of analyzed remote scenes, such that a spectrum of each pixel in said remote scenes is obtained;

(c) employing a decorrelation statistical method for extracting decorrelated spectral data characterizing said pixels; and

(d) comparing at least a part of said decorrelated spectral data extracted from said pixels of the remote scenes with said reference template.

13. The method for remote scenes classification as in claim 11, further comprising the step of:

(f) according to said comparison, attributing each pixel an artificial color.

14. The method for remote scenes classification as in claim 11, wherein said spectral imager includes an element selected from the group consisting of a dispersion element, a filter and an interfcrometer.

15. The method for remote scenes classification as in claim 11, wherein said decorrelation statistical method is selected from the group consisting of principal component analysis, canonical variable analysis and singular value decomposition.

16. The method for remote scenes classification as in claim 15, wherein said principal component analysis includes expressing each of said analyzed scenes as linear combinations of N eigenvectors.

12. A method for remote scenes classification comprising the steps of:

(a) preparing a reference template for classification of remote scenes via;

(i) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassifled reference scenes;

(iii) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(iv) using at least a part of said decorrelated spectral data for the preparation of the reference template for the remote scenes classification;

(b) using a second spectral imager for measuring a spectral cube of analyzed remote scenes, such that a spectrum of each pixel in said remote scenes is obtained;

(c) projecting said spectrum of each of said pixels onto said decorrelated spectral data for obtaining a projected spectrum for each of said pixels; and

(d) comparing said projected spectra with said reference template.

17. A method for remote scenes classification comprising the steps of:

(a) preparing a reference template for classification of remote scenes via;

(i) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes;

(iii) employing a principal component analysis for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes via;

(a) selecting k spectral slices for said spectral cube of said reference scenes;

(b) calculating an average spectrum for each of said reference scenes;

(c) stretching said average spectra for obtaining a stretched average spectrum for each of said reference scenes;

(d) averaging said stretched average spectra for each of said reference scenes for obtaining an ensemble average spectrum for each of said reference scenes;

(e) calculating a k dimension eigen system for said ensemble average spectra and extracting N eigenvectors;

(f) using said N eigenvectors for defining an N-dimension vector for each of said reference scenes; and

(g) using said N-dimension vectors for preparing said reference template for the remote scenes classification;

(b) using a second spectral imager for measuring a spectral cube of the remote scenes, such that a spectrum of each pixel in the remote scenes is obtained;

(c) projecting said spectrum of each of said pixels into said N eigenvectors for obtaining a projected N dimension vector for each of said pixels; and

(d) correlating each of said projected N dimension vectors with said reference template.

18. The method for remote scenes classification as in claim 17, the method further comprising the step of performing a spatial averaging procedure on all spectral slices.

19. A method of calculating decorrelation matched filters for remote scenes classification, the decorrelation matched filters being for extracting decorrelated spectral data from the remote scenes, the method comprising the step of:

(a) obtaining decorrelated spectral data characterizing a set of reference scenes via;

(i) classifying said set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes; and

(iii) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(b) mathematically manipulating at least a part of said decorrelated spectral data for obtaining a mathematical description of the decorrelation matched filters.

20. The method of calculating decorrelation matched filters for remote scenes classification as in claim 19, wherein said decorrelated spectral data is obtained using a principal component analysis, which includes expressing each of said reference scenes by a linear combination of N eigenvectors.

21. A set of decorrelation matched filters for remote scenes classification, the decorrelation matched filters being for extracting decorrelated spectral data from the remote scenes, the set comprising physical filters having shapes, said shapes following a mathematical description, said mathematical description being obtainable by:

(a) obtaining decorrelated spectral data characterizing a set of reference scenes via;

(i) classifying said set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes; and

(iii) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(b) mathematically manipulating at least a part of said decorrelated spectral data for obtaining said mathematical description of the decorrelation matched filters.

22. A method of tuning a tunable filter for remote scenes classification, the method renders the tunable filter to mimic a set of decorrelation matched filters, and is for extracting decorrelated spectral data from the remote scenes, the method comprising the steps of:

(a) obtaining decorrelated spectral data characterizing a set of reference scenes via;

(i) classifying said set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube ol said preclassified reference scenes; and

(iii) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes;

(b) mathematically manipulating at least a part of said decorrelated spectral data for obtaining a mathematical description describing the set of decorrelation matched filters; and

(c) sequentially tuning the tunable filter according to said mathematical description.

23. The method of tuning a tunable filter for remote scenes classification as in claim 22, wherein said tunable filter is selected from the group consisting of AOTF and LCTF.

24. A method for remote scenes classification comprising the steps of:

(a) preparing a reference template for classification of the remote scenes via;

(i) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(ii) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes;

(iii) employing a decorrelation statistical method lor extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(iv) using at least a part of said decorrelated spectral data for the preparation of the reference template for the remote scenes classification;

(b) calculating a mathematical description of decorrelation matched filters for classification of the remote scenes employing said reference template, said calculation being by mathematically manipulating at least a part of said decorrelated spectral data;

(c) using said mathematical description of said decorrelation matched filters for manufacturing said decorrelation matched filters;

(d) using said decorrelation matched filters for extracting decorrelated spectral data from each pixel of the remote scenes; and

(e) comparing said decorrelated spectral data extracted from each pixel of the remote scenes with the reference template.

25. The method for remote scenes classification as in claim 24, further comprising the step of:

(f) attributing each pixel an artificial color according to said comparison.

26. The method for remote scenes classification as in claim 24, wherein said decorrelation statistical method is selected from the group consisting of principal component analysis, canonical variable analysis and singular value decomposition.

27. The method for remote scenes classification as in claim 26, wherein said principal component analysis includes expressing each of said reference scenes as linear combinations of N eigenvectors.

28. A method for remote scenes classification comprising the steps of:

(a) providing a set of decorrelation matched filters for the remote scenes classification, the decorrelation matched filters being for extracting decorrelated spectral data from the remote scenes, the set including physical filters having shapes, said shapes following a mathematical description, said mathematical description being achieved by;

(i) preparing a reference template for classification of remote scenes via;

(a) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(b) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes;

(c) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(d) using at least a part of said decorrelated spectral data for the preparation of the reference template for the remote scenes classification; and

(ii) mathematically manipulating at least a part of said decorrelated spectral data for obtaining said mathematical description of said decorrelation matched filters;

(b) using said decorrelation matched filters for extracting decorrelated spectral data from each pixel of the remote scenes; and

(c) comparing said decorrelated spectral data extracted from each pixel of the remote scenes with the reference template.

29. The method for remote scenes classification as in claim 28, further comprising the step of:

(d) according to said comparison, attributing each pixel an artificial color.

30. The method for remote scenes classification as in claim 28, wherein said decorrelation statistical method is selected from the group consisting of principal component analysis, canonical variable analysis and singular value decomposition.

31. The method for remote scenes classification as in claim 30, wherein said principal component analysis includes expressing each of said reference scenes as linear combinations of N eigenvectors.

32. A method for remote scenes classification comprising the steps of:

(a) providing a tunable filter and tuning information for tuning said tunable filter so as to mimic a set of decorrelation matched filters, said tunable filter being for extracting decorrelated spectral data from the remote scenes, said tuning information being achieved by;

(i) preparing a reference template for classification of remote scenes via;

(a) classifying a set of reference scenes via a conventional classification technique for obtaining a set of preclassificd reference scenes;

(b) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes;

(c) employing a decorrelation statistical method for extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(d) using at least a part of said decorrelated spectral data for the preparation of the reference template for the remote scenes classification;

(ii) mathematically manipulating at least a part of said decorrelated spectral data for obtaining said tuning information, which is a mathematical description describing said set of decorrelation matched filters; and

(b) using said information for sequentially tuning said tunable filter for extracting decorrelated spectral data from each pixel of the remote scenes; and

(c) comparing said decorrelated spectral data extracted from each pixel of the remote scenes with the reference template.

33. A spectral decorrelation measurement apparatus for remote scenes classification by extracting decorrelated spectral data from the remote scenes, the apparatus is connected to a telescope used to view the remote scenes, said apparatus comprising:

(a) a detector; and

(b) an optical system for transmitting electromagnetic radiation from the remote scenes onto said detector, said optical system including a set of decorrelating matched filters, said decorrelation matched filters being for extracting decorrelated spectral data from the remote scenes, said filters of said set of decorrelatio n matched filters having shapes, said shapes following a mathematical description, said mathematical description being calculated by;

(i) obtaining decorrelated spectral data characterizing a set of reference scenes via;

(a) classifying said set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(b) using a first spectral imager for measuring a spectral cube of said preclassified reference scenes; and

(c) employing a decorrelation statistical method flor extracting said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(ii) mathematically manipulating at least a part of said decorrelated spectral data for obtaining said mathematical description of the decorrelation matched filters.

34. The spectral decorrelation measurement apparatus for remote scenes classification as in claim 33, wherein said optical system further includes a collimating lens for collimating radiation reaching any of said decorrelating matched filters.

35. The spectral decorrelation measurement apparatus for remote scenes classification as in claim 33, wherein said decorrelation matched filters are arranged on a rotatable filter carrying element.

36. A spectral decorrelation measurement apparatus for remote scenes classification by extracting decorrelated spectral data from the remote scenes, the apparatus is connected to a telescope used to view the remote scenes, said apparatus comprising:

(a) a detector; and

(b) an optical system for transmitting electromagnetic radiation from the remote scenes onto said detector, said optical system including a tunable filter and a tuning device, said tuning device being for tuning said tunable filter, so that said tunable filter sequentially mimics a set of decorrelating matched filters, said decorrelation matched filters mimicked by said tunable filter being for extracting decorrelated spectral data from the remote scenes, said tuning of said tunable filter being calculated according to a mathematical description, said mathematical description being calculated by;

(i) obtaining decorrelated spectral data characterizing a set of reference scenes via;

(a) classifying said set of reference scenes via a conventional classification technique for obtaining a set of preclassified reference scenes;

(b) using a first spectral imager for measuring a spectral cube of said preclassificd reference scenes; and

(c) employing a decorrelation statistical method for extracting said spectral cube for said spectral cube for decorrelated spectral data characterizing said reference scenes; and

(ii) mathematically manipulating at least a part of said decorrelated spectral data for obtaining said mathematical description of said mimicked decorrelation matched filters.