Automatic learning of belief functions

• US 6,125,339 A
• Filed: 12/23/1997
• Issued: 09/26/2000
• Est. Priority Date: 12/23/1997
• Status: Expired due to Fees

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Chen, Yang, Baker, Todd L., Reiser, Kurt
• Primary Examiner(s): Hoff, Marc S.
• Assistant Examiner(s): Bui, Bryan
• Application Number:
  US08/996,776
• Time in Patent Office: 1,008 Days
• Field of Search: 702/108, 702/181, 702/119, 702/123, 702/120, 702/121, 702/122, 706/47, 706/48
• US Class Current: 702/181
• CPC Class Codes: G06F 18/25 Fusion techniques
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A method providing automatic learning belief functions enabling the combination of different, and possibly contradictory information sources. The present invention provides the ability to determine erroneous information sources, inappropriate information combinations, and optimal information granularities, along with enhanced system performance.

Claims

1. A method of recognizing an object, comprising the steps of:

gathering a set of information from a sensor, wherein said information is representative of a characteristic of said object;

creating Dempster-Shafer basic probability assignments (bpa'"'"'s) based on said set of information, wherein said Dempster-Shafer bpa'"'"'s represent a probability that said object comprises a target;

creating combinations of said Dempster-Shafer bpa'"'"'s, wherein said combinations of said Dempster-Shafer bpa'"'"'s represent the probability that said object comprises a target;

measuring an error present in said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s;

calculating updates to said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said error; and

refining said probability of said object comprising a target by modifying said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said updates.

2. The method of claim 1 wherein said set of information comprises rules.

3. The method of claim 1 wherein said set of information comprises opinions.

4. The method of claim 1 wherein said set of information comprises a size of an object.

5. The method of claim 1 wherein said set of information comprises a shape of an object.

6. The method of claim 1 wherein said set of information comprises heat associated with an object.

7. The method of claim 1 wherein said step of measuring error comprises a comparison between said Dempster-Shafer bpa'"'"'s and a known desired result.

8. The method of claim 1 wherein said step of measuring error comprises a comparison between said combinations of said Dempster-Shafer bpa'"'"'s and a known desired result.

9. The method of claim 1 wherein said step of measuring error comprises a comparison between said Dempster-Shafer bpa'"'"'s and a set of characteristics of a desired result.

10. The method of claim 1 wherein said step of measuring error comprises a comparison between said combinations of said Dempster-Shafer bpa'"'"'s and a set of characteristics of a desired result.

11. The method of claim 1 wherein said updates to said Dempster-Shafer bpa'"'"'s are calculated using a gradient-descent rule.

12. An apparatus for recognizing an object comprising:

a signal processing unit;

a sensor that couples a set of information to said processing unit; and

a memory in communication with said signal processing unit, said memory containing data representative of a process to be executed by said signal processing unit;

said process comprising the steps of;

i) gathering a set of information from said sensor, wherein said information is representative of a characteristic of said object

ii) creating Dempster-Shafer basic probability assignments (bpa'"'"'s) based on said set of information, wherein said Dempster-Shafer bpa'"'"'s represent the probability that said object comprises a target;

iii) creating combinations of said Dempster-Shafer bpa'"'"'s, wherein said combinations of said Dempster-Shafer bpa'"'"'s represent a probability that said object comprises a target;

iv) measuring an error present in said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s;

v) calculating updates to said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said error; and

vi) refining said probability of said object being a target by modifying said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said updates.

13. The apparatus of claim 12 wherein said set of information comprises rules.

14. The apparatus of claim 12 wherein said set of information comprises opinions.

15. The apparatus of claim 12 wherein said error measurement comprises a comparison between said Dempster-Shafer bpa'"'"'s and a known desired result.

16. The apparatus of claim 12 wherein said error measurement comprises a comparison between said combinations of Dempster-Shafer bpa'"'"'s and a known desired result.

17. The apparatus of claim 12 wherein said error measurement comprises a comparison between said Dempster-Shafer bpa'"'"'s and a set of characteristics of a desired result.

18. The method of claim 12 wherein said error measurement comprises a comparison between said combinations of Dempster-Shafer bpa'"'"'s and a set of characteristics of a desired result.

19. The method of claim 12 wherein said updates to said Dempster-Shafer bpa'"'"'s are calculated using a gradient-descent rule.

20. A method of recognizing an object, comprising the steps of:

gathering a set of information from a sensor, wherein said information is representative of a characteristic of said object;

creating Dempster-Shafer basic probability assignments (bpa'"'"'s) based on said set of information, wherein said Dempster-Shafer bpa'"'"'s represent a probability that said object comprises a target;

creating combinations of said Dempster-Shafer bpa'"'"'s, wherein said combinations of said Dempster-Shafer bpa'"'"'s represent the probability that said object comprises a target;

measuring an error present in said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s, wherein said error is calculated according to (μ

_d -μ

_o)² wherein μ

_d is a desired Dempster-Shafer bpa and μ

_o is a Dempster-Shafer bpa based on said set of information;

calculating updates to said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said error; and

refining said probability of said object comprising a target by modifying said Dempster-Shafer bpa'"'"'s and said combinations of said Dempster-Shafer bpa'"'"'s based on said updates.