Substance detection system

• US 20050244978A1
• Filed: 04/30/2004
• Published: 11/03/2005
• Est. Priority Date: 04/30/2004
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Uluyol, Onder
• Granted Patent: US 7,531,137 B2
• Time in Patent Office: Days
• US Class Current: 436/149
• CPC Class Codes: G01N 33/0034 comprising neural networks or related mathematical techniques

Abstract

A substance detection system having an electronic nose and classifier. The system may extract feature from the electronic signals representing the smells from the electronic nose sensing an unknown substance. The features may be formatted as smellprints that are synthesized as data. The features may be classified, and in a binary tournament fashion, as an illustrative example, may be mapped to be compared or correlated with features of known substances. The known substance or substances having the highest scores for good mapping, comparison or correlation with the sensed substance, may be reviewed in view of decision criteria to determine an identification of the sensed substance.

Claims

1. A substance detection system comprising:

an electronic nose;

a feature extraction module connected to the electronic nose;

a data synthesis module connected to the feature extraction module;

a classification module connected to the data synthesis module; and

a decision criteria module connected to the classification module.

2. The system of claim 1, wherein the classification module comprises a support vector machine.

3. The system of claim 2, wherein the feature extraction module provides a smellprint to the data synthesis module.

4. The system of claim 3, wherein the smellprint comprises a final activation and initial response data.

5. The system of claim 4, wherein the output classification module has an output for classification results.

6. The system of claim 5, wherein the decision criteria module is for filtering the classification results.

7. The system of claim 6, wherein a first criterion for filtering the classification results is a threshold.

8. The system of claim 6, wherein a criterion for filtering the classification results is a winner-take-all standard.

9. The system of claim 6, wherein the feature extraction module comprises:

a baseline flag;

a baseline resistance processor having a first input from the electronic nose and a second input from the baseline flag;

a fractional normalizer having a first input from the electronic nose and a second input from the baseline resistance;

a filter having an input connected to the fractional normalizer;

a gradient indicator having an input connected to the filter;

a steady-state period indicator at the baseline purge having an input connected to the baseline flag;

a sample draw flag;

a transient period indicator at sample draw having an input connected to the sample draw flag;

an indicator of average or absolute variation at baseline having a first input connected to the steady-state period indicator at baseline purge and a second input connected to the gradient indicator;

an indicator of maximum absolute variation at initial response having a first input connected to the transient period indicator at sample draw and having a second input connected to the gradient indicator;

a relative variation processor having a first input connected to the indicator of average of absolute variation at baseline and a second input connected to the indicator of maximum absolute variation at initial response;

a vector normalizer having an input connected to the relative variation processor; and

an initial response feature vector indicator having an input connected to the vector normalizer and an output connected to the data synthesis module.

10. The system of claim 6, further comprising a leave-one-out connection from the classification module to the data synthesis module.

11. A method for detecting a substance comprising:

sensing a substance with an electronic nose providing a signal corresponding to the substance;

extracting at least one feature from the signal;

synthesizing the at least one feature;

classifying the at least one feature;

associating the at least one feature with at least one feature of a set of known substances;

selecting a substance of the set that most likely resembles the sensed substance according to the at least one feature; and

identifying the sensed substance.

12. The method of claim 11, wherein:

the at least one feature is a first smellprint; and

the classifying the at least one feature involves a support vector machine.

13. The method of claim 12, further comprising:

selecting one of each pair of substance smellprints of a plurality of smellprints of substances that compares most closely to the first smellprint; and

determining which smellprint of the plurality of smellprints of substances is most selected.

14. The method of claim 13, further comprising:

applying a decision criterion relative to the smellprint that is most selected; and

determining whether the most selected smellprint is the same as the first smellprint so as to identify the substance sensed by the electronic nose.

15. Means for detecting a substance comprising:

means for sensing a substance via a smell;

means for converting the smell into at least one signal;

means for extracting features from the at least one signal;

means for mapping features to features of each of a plurality of known substances;

means for indicating a score of mapping for each of the known substances;

means for selecting the known substance having the highest score; and

means for determining whether the sensed substance is identified as the same as the known substance having the highest score.

16. The means of claim 15, wherein the means for determining comprises a criterion selected from a group of the threshold levels and winner-take-all criteria.

17. The means of claim 16, wherein the mapping of features is a binary tournament classification.

18. The means of claim 17, wherein the binary tournament classification comprises a support vector machine.

19. A substance detection system comprising:

an electronic smelling mechanism having a signal output representing a detected smell;

a feature extraction device connected to the electronic smelling mechanism; and

a feature classification module, connected to the feature extraction device, having an output indicating at least one known substance that has a smell similar to the detected smell.

20. The system of claim 19, further comprising a decision maker connected to the output of the feature classification module.

21. The system of claim 19, wherein the decision maker contains criteria relating to whether the detected smell is from a known substance.

22. The system of claim 21, wherein the feature classification module comprises a support vector machine.