Method for identifying known materials within a mixture of unknowns

• US 6,035,246 A
• Filed: 05/15/1997
• Issued: 03/07/2000
• Est. Priority Date: 11/04/1994
• Status: Expired due to Term

• Current Assignee: Sandia Corporation (Martin Marietta Corporation)
• Original Assignee: Sandia Corporation (Martin Marietta Corporation)
• Inventors: Wagner, John S.
• Primary Examiner(s): Cosimano, Edward R.
• Application Number:
  US08/773,748
• Time in Patent Office: 1,027 Days
• Field of Search: 250/339.05, 250/339.11, 250/339.12, 250/373, 356/317, 356/318, 356/320, 364/497, 364/498, 395/13, 395/20, 395/21, 395/23, 422/82.05, 436/171, 700/266, 706/15, 706/45
• US Class Current: 700/266
• CPC Class Codes: G01J 3/28 Investigating the spectrum using colour filters G01J3/51
• CPC Class Codes: G01N 21/25 Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
• CPC Class Codes: G01N 21/3577 for analysing liquids, e.g. polluted water
• CPC Class Codes: G01N 21/359 using near infrared light
• CPC Class Codes: G01N 33/0034 comprising neural networks or related mathematical techniques

Abstract

One or both of two methods and systems are used to determine concentration of a known material in an unknown mixture on the basis of the measured interaction of electromagnetic waves upon the mixture. One technique is to utilize a multivariate analysis patch technique to develop a library of optimized patches of spectral signatures of known materials containing only those pixels most descriptive of the known materials by an evolutionary algorithm. Identity and concentration of the known materials within the unknown mixture is then determined by minimizing the residuals between the measurements from the library of optimized patches and the measurements from the same pixels from the unknown mixture. Another technique is to train a neural network by the genetic algorithm to determine the identity and concentration of known materials in the unknown mixture. The two techniques may be combined into an expert system providing cross checks for accuracy.

Claims

1. A method for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

(a) irradiating the purest available sample of at least one known constituent with a reference electromagnetic wave excitation source that sequentially emits a plurality of individual bands of wavelengths of radiation, with the energy from the excitation source interacting with the sample at a first set of wavelengths;

(b) recording at least one parameter as a measurement of the response of the sample of the at least one known constituent to the excitation source at a second set of wavelengths to form an n-dimensional array consisting of a multiplicity of elements, where n≧

2;

(c) creating an optimized patch subset in the array by an evolutionary algorithm, the optimized patch subset containing only those array elements which indicate the presence of the sample of the at least one known constituent;

(d) irradiating the unknown mixture at the first set of wavelengths;

(f) recording the at least one parameter as a measurement of the response of the unknown mixture to the excitation source at the second set of wavelengths to form the n-dimensional array of step (b); and

(g) determining the amount of the at least one known constituent in the unknown mixture by minimizing a function of the residuals between measurements of the at least one parameter associated with the elements in the optimized patch subset for the sample of the at least one known constituent and the measurements of the at least one parameter associated with the elements in a patch subset from the unknown mixture which corresponds with the optimized patch set.

2. The method of claim 1 wherein the excitation source is a flash lamp equipped with a plurality of filters to provide the first set of wavelengths.

3. The method of claim 1 wherein the excitation source is a laser tuned to the first set of wavelengths.

4. The method of claim 1 wherein the determining of the amount of the at least one known constituent is done by multivariate analysis.

5. The method of claim 1 wherein the at least one known constituent is at least one chemical.

6. The method of claim 1 wherein steps (a)-(c) are conducted separately from and prior to the succeeding steps and the optimized patch subset from step (c) is stored for later use in the determining step (g).

7. The method of claim 1 wherein the at least one parameter is the magnitude of the each of the second set of wavelengths.

8. The method of claim 7 wherein the magnitude measured is of uv fluorescence.

9. The method of claim 1 wherein the at least one known constituent is at least one biologic material.

10. The method of claim 9 wherein the at least one biologic material is at least one material selected from the group consisting of viruses, bacteria, DNA, RNA, and abnormal cells including cancer cells.

11. The method of claim 1 wherein the creating of the optimized patch subset in the array by the evolutionary algorithm is done by a method comprising:

(a) creating a first generation of unoptimized patch subsets with z members, each patch subset being described by initial patch boundaries as part of a genetic sequence describing each patch subset,

(b) assigning a fitness level to each member of the first generation as a function of the magnitude of a function of the residuals remaining when the amounts of the at least one known constituent determined from the member of the first generation of patch sets are subtracted from the actual amounts,

(c) dropping those members of the first generation below a certain fitness level, allowing at least some of the remaining members to reproduce by exchanging corresponding portions of their genetic sequences, and introducing random mutations into some of the sequences of the members to form a next generation of members,

(d) assigning a fitness level to each member of the next generation of step (c) as a function of the residuals remaining when the amounts of the at least one known constituent determined from the member of the next generation of patch sets of step (c) are subtracted from the actual amounts, and

(e) repeating steps (c) and (d) until a member of a succeeding generation is identified as having sufficient fitness to qualify as the optimized patch subset.

12. The method of claim 11 wherein the member of a succeeding generation identified as having sufficient fitness is a member representing the average patch subset for that generation.

13. The method of claim 1 wherein the element is a pixel representing one of the second set of wavelengths.

14. The method of claim 13 wherein the pixel is associated with other pixels to form a two dimensional array.

15. The method of claim 14 wherein the optimized patch subset includes only those associated pixels which contain information indicative of the at least one known constituent.

16. A method for identifying the amount of at least one known chemical contained within an unknown mixture comprising:

(a) irradiating the purest available sample of at least one chemical with a reference electromagnetic wave excitation source that sequentially emits a plurality of individual bands of wavelengths of radiation, with the energy from the excitation source interacting with the sample at a first set of wavelengths;

(b) recording the uv fluourescence response of the sample of the at least one chemical to the excitation source at a second set of wavelengths to form an two-dimensional array;

(c) creating an optimized patch subset in the array by an evolutionary algorithm, the optimized patch subset containing only those array elements which indicate the presence of the sample of the at least one chemical;

(d) irradiating the unknown mixture at the first set of wavelengths;

(f) recording the at least one parameter as a measurement of the response of the unknown mixture to the excitation source at the second set of wavelengths to form the n-dimensional array of step (b); and

(g) determining the amount of the at least one known constituent in the unknown mixture by minimizing a function of the residuals between measurements of the at least parameter associated with the elements in the optimized patch subset for the sample of the at least one chemical and the measurements of the at least one parameter associated with the elements in a patch subset from the unknown mixture which corresponds with the optimized patch set.

17. The method of claim 16 wherein the excitation source is a flash lamp equipped with a plurality of filters to provide the first set of wavelengths.

18. The method of claim 16 wherein the excitation source is a laser tuned to the first set of wavelengths.

19. The method of claim 16 wherein the determining of the amount of the at least one known constituent is done by multivariate analysis.

20. The method of claim 16 wherein steps (a)-(c) are conducted separately from and prior to the succeeding steps and the optimized patch subset from step (c) is stored for later use in the determining step (g).

21. A method for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

(a) irradiating a set of at least one of the purest available samples of the known constituents with a reference excitation source, the excitation source emitting at a first set of wavelengths;

(b) recording at least one parameter as a measurement of the response of the at least one samples of the constituents to the excitation source at a second set of wavelengths;

(c) creating, by an evolutionary algorithm, an optimized neural network comprising at least three layers of neurons comprising an input layer, at least one middle layer and an output layer, the input layer including neurons having a single input connection exclusively to an individual band of wavelengths in the second set of wavelengths, the output layer of neurons having a single output connection, and each neuron in a layer being at least initially connected to each of the neurons in the layers above it, the input from each such connection to a neuron from the layer above being multiplied by a weight in each neuron, the optimized neural network being created by a process comprising;

(i) creating a first generation of a neural network having at least three layers with z members in the generation each member being described by a genetic sequence containing initial weights assigned to the input connections to the neurons in the layers,

(ii) assigning a fitness level to each member of the first generation as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b),

(iii) dropping those members of the first generation below a certain fitness level, allowing at least some of the remaining members to reproduce by exchanging corresponding portions of their genetic sequences, and introducing random mutations into some of the sequences of the members to form a next generation of members,

(iv) assigning a fitness level to each member of the next generation of step (iii) as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b) modified by a change in the relative concentrations of the at least one sample in the set, and

(v) repeating steps (iii) and (iv) with repeated changes in the relative concentrations of the samples in the set until a member of a succeeding generation is identified as having sufficient accuracy of output information to qualify as the optimized neural network

(d) irradiating the unknown mixture with the excitation source at the first set of wavelengths;

(e) recording the at least one parameter from the unknown mixture at the second set of wavelengths; and

(f) identifying the at least one known constituent in the unknown mixture by observing the output of the optimized neural network after it is supplied with the parameters measured at the second set of wavelengths from the unknown mixture at the inputs of the optimized neural network.

22. The method of claim 21 wherein the evolutionary algorithm further removes at least some of the interconnections from the network with zero value weights associated therewith.

23. The method of claim 21 wherein the evolutionary algorithm further removes at least some of the neurons for which the summed inputs to that neuron total zero.

24. The method of claim 21 wherein the evolutionary algorithm introduces random mutations in part by adding new neurons to the network.

25. The method of claim 21 wherein the evolutionary algorithm introduces random mutations in part by adding new interconnections to the neurons in the network.

26. The method of claim 21 wherein the excitation source is a flash lamp equipped with a plurality of filters to provide the first set of wavelengths.

27. The method of claim 21 wherein the excitation source is a laser tuned to the first set of wavelengths.

28. The method of claim 21 wherein the parameter measured is the magnitude of uv fluorescence.

29. The method of claim 21 wherein the at least one known constituent is at least one chemical.

30. The method of claim 21 wherein at least some of the input layer neurons are connected exclusively to inputs providing information about the operational characteristics of the apparatus performing the irradiating and recording.

31. The method of claim 21 wherein the identification includes a determination of concentration.

32. The method of claim 21 wherein steps (a)-(c) are conducted separately from and prior to the succeeding steps and the optimized neural network from step (c) is stored for later use in the identifying step (f).

33. The method of claim 21 wherein the at least one known constituent is at least one biologic material.

34. The method of claim 33 wherein the at least one biologic material is at least one material selected from the group consisting of viruses, bacteria, DNA, RNA, and abnormal cells including cancer cells.

35. A method for identifying the amount of at least one known chemical contained within an unknown mixture comprising:

(a) irradiating a set of at least one purest available sample of the known chemicals with a reference excitation source, the excitation source emitting at a first set of wavelengths;

(b) recording at least one parameter as a measurement of the response of the at least one sample of the chemical to the excitation source at a second set of wavelengths;

(c) creating, by an evolutionary algorithm, an optimized neural network comprising at least three layers of neurons comprising an input layer, at least one middle layer and an output layer, the input layer including neurons having a single input connection exclusively to an individual band of wavelengths in the second set of wavelengths, the output layer of neurons having a single output connection, and each neuron in a layer being at least initially connected to each of the neurons in the layers above it, the input from each such connection to a neuron from the layer above being multiplied by a weight in each neuron, the optimized neural network being created by a process comprising;

(i) creating a first generation of a neural network having at least three layers with z members in the generation each member being described by a genetic sequence containing initial weights assigned to the input connections to the neurons in the layers,

(ii) assigning a fitness level to each member of the first generation as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b),

(iii) dropping those members of the first generation below a certain fitness level, allowing at least some of the remaining members to reproduce by exchanging corresponding portions of their genetic sequences, and introducing random mutations into some of the sequences of the members to form a next generation of members,

(iv) assigning a fitness level to each member of the next generation of step (iii) as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b) modified by a change in the relative concentrations of the at least one sampler in the set, and

(v) repeating steps (iii) and (iv) with repeated changes in the relative concentrations of the samples in the set until a member of a succeeding generation is identified as having sufficient accuracy of output information to qualify as the optimized neural network

(d) irradiating the unknown mixture at the first set of wavelengths;

(e) recording the at least one parameter from the unknown mixture at the second set of wavelengths; and

(f) identifying the at least one known chemical in the unknown mixture by observing the output of the optimized neural network after it is supplied with the parameters measured at the second set of wavelengths from the unknown mixture at the inputs of the optimized neural network.

36. The method of claim 35 wherein the evolutionary algorithm further removes at least some of the interconnections from the network with zero value weights associated therewith.

37. The method of claim 35 wherein the evolutionary algorithm further removes at least some of the neurons for which the summed inputs to that neuron total zero.

38. The method of claim 35 wherein the evolutionary algorithm introduces random mutations in part by adding new neurons to the network.

39. The method of claim 35 wherein the evolutionary algorithm introduces random mutations in part by adding new interconnections to the neurons in the network.

40. The method of claim 35 wherein the identification includes a determination of concentration.

41. The method of claim 35 wherein steps (a)-(c) are conducted separately from and prior to the succeeding steps and the optimized neural network from step (c) is stored for later use in the identifying step (f).

42. A system for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

(a) a reference electromagnetic wave excitation source that sequentially emits a plurality of discrete bands of wavelengths of radiation for irradiating a purest available sample of at least one constituent at a first set of wavelengths;

(b) means for recording at least one parameter as a measurement of the response of the sample of the at least one constituent to the excitation source at a second set of wavelengths to form an n-dimensional array consisting of a multiplicity of elements, where n≧

2;

(c) means for creating an optimized patch subset in the array by an evolutionary algorithm, the optimized patch subset containing only those array elements which indicate the presence of the pure sample of the at least one constituent;

(d) means for irradiating the unknown mixture at the first set of wavelengths;

(f) means for recording the at least one parameter as a measurement of the response of the unknown mixture to the excitation source at the second set of wavelengths to form the n-dimensional array of step (b); and

(g) means for determining the amount of the at least one known constituent in the unknown mixture by minimizing a function of the residuals between measurements of the at least parameter associated with the elements in the optimized patch subset for the sample of the at least one constituent and the measurements of the at least one parameter associated with the elements in a patch subset from the unknown mixture which corresponds with the optimized patch set.

43. The system of claim 42 wherein the excitation source is a flash lamp equipped with a plurality of filters to provide the first set of wavelengths.

44. The system of claim 42 wherein the excitation source is a laser tuned to the first set of wavelengths.

45. The system of claim 42 wherein the means for determining of the amount of the at least one known constituent comprises multivariate analysis means.

46. The system of claim 42 wherein the at least one known constituent is at least one chemical.

47. The system of claim 42 wherein the means for creating the optimized patch subset in the array by the evolutionary algorithm comprises:

(a) means for creating a first generation of unoptimized patch subsets with z members in the generation each member being described by initial patch boundaries as part of a genetic sequence describing each,

(b) means for assigning a fitness level to each member of the first generation as a function of the magnitude of a function of the residuals remaining when the amounts of the at least one known constituent determined from the member of the first generation of patch sets are subtracted from the actual amounts,

(c) means for dropping those members of the first generation below a certain fitness level, allowing at least some of the remaining members to reproduce by exchanging corresponding portions of their genetic sequences, and introducing random mutations into some of the sequences of the members to form a next generation of members,

(d) means for assigning a fitness level to each member of the next generation of step (c) as a function of the residuals remaining when the amounts of the at least one known constituent determined from the member of the next generation of patch sets of step (c) are subtracted from the actual amounts, and

(e) means for repeating steps (c) and (d) until a member of a succeeding generation is identified as having sufficient fitness to qualify as the optimized patch subset.

48. The system of claim 47 wherein the member of a succeeding generation identified as having sufficient fitness is a member representing the average patch subset for that generation.

49. The system of claim 42 wherein the at least one parameter is the magnitude of the each of the second set of wavelengths.

50. The system of claim 49 wherein the magnitude measured is of uv fluorescence.

51. The system of claim 42 wherein the at least one known constituent is at least one biologic material.

52. The system of claim 51 wherein the at least one biologic material is at least one material selected from the group consisting of viruses, bacteria, DNA, RNA, and abnormal cells including cancer cells.

53. The system of claim 42 wherein the element is a pixel representing one of the second set of wavelengths.

54. The system of claim 53 wherein the pixel is associated with other pixels to form a two dimensional array.

55. The system of claim 54 wherein the optimized patch subset includes only those associated pixels which contain information indicative of the at least one known constituent.

63. The system of claim 53 wherein the at least one known constituent is at least one biologic material.

64. The system of claim 63 wherein the at least one biologic material is at least one material selected from the group consisting of viruses, bacteria, DNA, RNA, and abnormal cells including cancer cells.

56. A system for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

(a) a reference excitation source for irradiating a set of at least one purest available sample of the known constituents at a first set of wavelengths;

(b) means for recording at least one parameter as a measurement of the response of the at least one sample of the constituent to the excitation source at a second set of wavelengths;

(c) means for creating, by an evolutionary algorithm, an optimized neural network comprising at least three layers of neurons comprising an input layer, at least one middle layer and an output layer, the input layer including neurons having a single input connection exclusively to an individual band of wavelengths in the second set of wavelengths, the output layer of neurons having a single output connection, and each neuron in a layer being at least initially connected to each of the neurons in the layers above it, the input from each such connection to a neuron from the layer above being multiplied by a weight in each neuron, the means for creating the optimized neural network comprising;

(i) means for creating a first generation of a neural network having at least three layers with z members in the generation each having being described by a genetic sequence containing initial weights assigned to the input connections to the neurons in the layers,

(ii) means for assigning a fitness level to each member of the first generation as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b),

(iii) means for dropping those members of the first generation below a certain fitness level, allowing at least some of the remaining members to reproduce by exchanging corresponding portions of their genetic sequences, and introducing random mutations into some of the sequences of the members to form a next generation of members,

(iv) means for assigning a fitness level to each member of the next generation of step (iii) as a function of the accuracy of the information provided from the outputs of the output neurons compared against the measurements of step (b) modified by a change in the relative concentrations of the at least one samples in the set, and

(v) means for repeating steps (iii) and (iv) with repeated changes in the relative concentrations of the samples in the set until a member of a succeeding generation is identified as having sufficient accuracy of output information to qualify as the optimized neural network

(d) means for irradiating the unknown mixture at the first set of wavelengths;

(e) means for recording the at least one parameter from the unknown mixture at the second set of wavelengths; and

(f) means for identifying the at least one known constituent in the unknown mixture by observing the output of the optimized neural network after it is supplied with the parameters measured at the second set of wavelengths from the unknown mixture at the inputs of the optimized neural network.

57. The system of claim 56 wherein the excitation source is a flash lamp equipped with a plurality of filters to provide the first set of wavelengths.

58. The system of claim 56 wherein the excitation source is a laser tuned to the first set of wavelengths.

59. The system of claim 56 wherein the parameter measured is the magnitude of uv fluorescence.

60. The system of claim 56 wherein the at least one known constituent is at least one chemical.

61. The system of claim 56 wherein at least some of the input layer neurons are connected exclusively to inputs providing information about the operational characteristics of the apparatus performing the irradiating and recording.

62. The system of claim 56 wherein the identification includes a determination of concentration.

65. A system for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

means for establishing a database containing responses of samples of the known constituents to irradiation at a first set of wavelengths of electromagnetic radiation measured at a second set of wavelengths, the database comprising a plurality of two-dimensional patches of pixels containing information measured at the second set of wavelengths indicative of the known constituents;

means for establishing an optimized neural network comprising an evolutionary algorithm which evolves the structure of an at least three level neural network by training the neural network against a set containing the responses of a plurality of known constituents to the first set of wavelengths which set changes the relative concentrations of its members as each generation of the neural network evolves, said optimized neural network providing the amount of the known constituents when provided with the response of the known constituents at the second set of wavelengths;

means to irradiate the unknown mixture with the first set of wavelengths;

means to record the responses of the unknown mixture at the second set of wavelengths;

means to process the responses with the optimized neural network to ascertain a first value for the amounts of the known constituents in the unknown mixture;

means to select from the database those two-dimensional patches indicative of the known constituents ascertained by the optimized neural network;

means to minimize the residuals by multivariate analysis between the patches selected from the database by the optimized neural network and the corresponding patches in the responses from the unknown mixture, said means providing the amounts of the constituents.

66. The system of claim 65 further comprising means to compare the amounts of the constituents from the optimized neural network and the amounts of the constituents from the means for minimizing the residuals by multivariate analysis.

67. A method for identifying the amount of at least one known constituent contained within an unknown mixture comprising:

establishing a database containing responses of samples of the known constituents to irradiation at a first set of wavelengths of electromagnetic radiation measured at a second set of wavelengths, the database comprising a plurality of two-dimensional patches of pixels containing information measured at the second set of wavelengths indicative of the known constituents;

establishing an optimized neural network comprising an evolutionary algorithm which evolves the structure of an at least three level neural network by training the neural network against a set containing the responses of a plurality of known constituents to the first set of wavelengths which set changes the relative concentrations of its members as each generation of the neural network evolves, said optimized neural network providing the amount of the known constituents when provided with the response of the known constituents at the second set of wavelengths;

irradiating the unknown mixture with the first set of wavelengths;

recording the responses of the unknown mixture at the second set of wavelengths;

processing the responses with the optimized neural network to ascertain a first value for the amounts of the known constituents in the unknown mixture;

selecting from the database those two-dimensional patches indicative of the known constituents ascertained by the optimized neural network;

minimizing the residuals by multivariate analysis between the patches selected from the database by the optimized neural network and the corresponding patches in the responses from the unknown mixture, thereby providing the amounts of the constituents.

68. The method of claim 67 further comprising comparing the amounts of the constituents from the optimized neural network and the amounts of the constituents from the means for minimizing the residuals by multivariate analysis.