Generalized hebbian learning for principal component analysis and automatic target recognition, systems and method

• US 6,894,639 B1
• Filed: 12/18/1991
• Issued: 05/17/2005
• Est. Priority Date: 12/18/1991
• Status: Expired due to Term

• Current Assignee: Intellectual Ventures II LLC (Intellectual Ventures LLC)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Katz, Alan Jerry
• Primary Examiner(s): Gregory, Bernarr E.
• Application Number:
  US07/810,005
• Time in Patent Office: 4,899 Days
• Field of Search: 342/27, 342/28, 342/61, 342 62- 81, 342 89-103, 342159-164, 342/175, 342/192, 342193-197, 342/22, 367/87, 367 93- 98, 367/135, 367/136, 704/200, 704/231, 704232-257, 706 15- 44, 382/103
• US Class Current: 342/90
• CPC Class Codes: G01S 7/417 involving the use of neural networks
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06T 2207/10044 Radar image
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 7/11 Region-based segmentation

Abstract

This application discloses a method for distinguishing targets from clutter, comprising the steps inputting data, calculating data statistics from said data and using said data statistics to select target specific feature information to distinguish specific targets from background clutter, generating said target specific feature information from said data statistics, extracting said target specific feature information from said data, using said target specific feature information to distinguish specific targets from background clutter, and outputting target and background clutter information. Classification systems, including hardware and software embodiments, are also disclosed.

Claims

1. A method for distinguishing targets from background clutter, comprising the steps of

(a) inputting data;

(b) calculating data statistics from said data and using said data statistics to select target specific feature information to distinguish specific targets from background clutter;

(c) generating said target feature information from said data statistics;

(d) extracting said target specific feature information from said data;

(e) using said target specific feature information to distinguish specific targets from background clutter; and

(f) outputting target and background clutter information.

2. The method of claim 1, wherein said data is one-dimensional, two-dimensional, or multi-dimensional.

3. The method of claim 1, wherein said data is digitized imagery, speech signals, or radar signals.

4. The method of claim 1, wherein parallel hardware architectures are employed to use said feature information to distinguish particular targets from background clutter.

5. The method of claim 4, wherein said parallel hardware architectures comprise neural networks.

6. The method of claim 1, wherein said feature information is contrast-based or texture-based.

7. The method of claim 1, wherein said feature information is further used to distinguish two target classes or a plurality of target classes from one another.

8. The method of claim 1, wherein Hebbian learning techniques are used to select said target specific feature information.

9. A classification system, comprising

(a) a calculator to calculate data statistics from a data input;

(b) a selector to use said data statistics outputed from said calculator to select target specific feature information to distinguish specific targets from background clutter, wherein said target specific feature information uniquely describes a target for recognition purposes;

(c) a generator to generate said target specific feature information from said data statistics outputed from said calculator;

(d) a feature extractor to extract target specific feature information from said data input; and

(e) a classifier to use said target specific feature information extracted from said data input by said feature extractor to distinguish targets from background clutter and output target and background clutter data.

10. The classification system of claim 9, wherein said data input is one-dimensional, two-dimensional, or multidimensional.

11. The classification system of claim 9, wherein said data input is digitized imagery, speech signals or radar signals.

12. The classification system of claim 9, wherein parallel hardware architectures are employed in the classifier to use said feature information to distinguish particular targets from background clutter.

13. The classification system of claim 12, wherein said parallel hardware architectures comprise neural networks.

14. The classification system of claim 9, wherein said feature information is contrast-based or texture-based.

15. The classification system of claim 9, wherein said selector employs Hebbian learning techniques to select said target specific feature information.

16. A classification hardware system, comprising

(a) a calculator to calculate data statistics from a data input;

(b) a selector to use said data statistics outputed from said calculator to select target specific feature information to distinguish specific targets from background clutter, wherein said target specific feature information uniquely describes a target for recognition purposes;

(c) a generator to generate said target specific feature information from said data statistics outputed from said calculator,

(d) a feature extractor to extract target specific feature information from said data input; and

(e) a classifier to use said target specific feature information extracted from said data input by said feature extractor to distinguish targets from background clutter and output target and background clutter data.

17. The classification hardware system of claim 16, wherein said selector employs Hebbian learning techniques to select said target specific feature information.

18. A classification software system, comprising

(a) calculation software to calculate data statistics from a data input;

(b) selection software to use said data statistics outputed from said calculator to select target specific feature information to distinguish specific targets from background clutter, wherein said target feature information uniquely describes a target for recognition purposes;

(c) generator software to generate said target specific feature information from said data statistics outputed from said calculator;

(d) feature extraction software to extract target specific feature information from said data input;

(e) classification software to use said target specific feature information extracted from said data input by said feature extractor to distinguish targets from background clutter and output target and background clutter data; and

(f) wherein the calculation software, selection software, generator software, feature extraction software, and classification software reside on at least one computer-readable medium.

19. The classification software system of claim 18, wherein said selector employs Hebbian learning techniques to select said target specific feature information.

20. A method for distinguishing targets from background clutter, comprising the steps of

(a) inputting data having targets and background clutter;

(b) applying neural network learning procedures to said data to construct a set of operators;

(c) applying said set of operators to data to construct a set of features;

(d) choosing optimal features from said set of features to distinguish targets from clutter; and

(e) using said optimal features to distinguishing targets from background clutter.