General frame wavelet classifier

• US 5,481,269 A
• Filed: 05/27/1994
• Issued: 01/02/1996
• Est. Priority Date: 05/27/1994
• Status: Expired due to Fees

• Current Assignee: Northrop Grumman Corporation
• Original Assignee: Westinghouse Electric Corporation (Paramount Global (f/k/a ViacomCBS Inc.))
• Inventors: Woolfson, Martin G., Imhoff, Scott A., Brinsley, James R., Morici, Martin M., Bendor, Giora A., Cheesman, Guy W. III
• Primary Examiner(s): Eldred, J. Woodrow
• Application Number:
  US08/250,226
• Time in Patent Office: 585 Days
• Field of Search: 342/89, 342/90, 342/159, 342/162, 342/175, 342/195, 364/485, 364/487
• US Class Current: 342/90
• CPC Class Codes: G01S 7/411 Identification of targets based on measurements of radar reflectivity G01S7/415 takes precedence
• CPC Class Codes: G06F 17/14 Fourier, Walsh or analogous domain transformations , e.g. Laplace, Hilbert, Karhunen-Loeve, transforms for correlation function computation G06F17/156; spectrum analysers G01R23/16
• CPC Class Codes: G06F 2218/08 Feature extraction
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 10/431 Frequency domain transformation; Autocorrelation

Abstract

A classifier employing general frame wavelets for improving target recognition while decreasing computational intensity is disclosed. A wavelet selection means compares a portion of a time dependent input signal from a receiver with a set of sample wavelets, each wavelet being bandpass limited, locally supported, and mean zero, and selects a prototype wavelet resembling the input portion. A resolving means expands the prototype wavelet onto a basis of wavelet functions having variations in scale and delay, and resolves the input signal onto the basis of functions. A regionpass filter identifies a region from a grid of resolved coefficients that contains features of a target from the input signal and discards coefficients outside the region. A separability filter passes select coefficients from a scale band within the identified region where separability among known targets is high. A comparator compares the selected coefficients with an exemplar of known targets to indicate a recognition of the target.

Claims

1. A classifier for recognizing features of a target in a received signal, comprising:

means for selecting a prototype wavelet corresponding in amplitude and phase per unit of time to a portion of said received signal;

means for resolving said received signal onto a basis of wavelet functions in a wavelet domain by calculating inner product transform coefficients between said received signal and each of said wavelet functions, said wavelet functions corresponding to variations in scale and delay for said prototype wavelet, said variations in scale and delay exceeding a frame condition for said prototype wavelet and said basis, said frame condition enabling an inverse wavelet transform of said calculated inner product transform coefficients to converge on said received signal;

a regionpass filter for identifying a target region of said basis and for passing said inner product transform coefficients within said target region, said target region including said target features in the received signal;

a separability filter coupled to said regionpass filter for passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable; and

a comparator coupled to said separability filter for comparing said select inner product transform coefficients to an exemplar, said comparison indicating a recognition of said features of the target.

2. The classifier of claim 1, wherein said basis of wavelet functions is orthonormal.

3. The classifier of claim 1, wherein said prototype wavelet is bandpass limited, locally supported, and mean zero.

4. The classifier of claim 1, wherein said basis of wavelet functions is overcomplete.

5. The classifier of claim 4, wherein said overcomplete basis of wavelet functions is general frame.

6. The classifier of claim 5, wherein said frame condition is critical frame, said critical frame indicating the variation for scale and delay for said prototype wavelet where a number of said inner product calculations is a minimum for said inverse wavelet transform to converge.

7. The classifier of claim 6, wherein said critical frame for said prototype wavelet h_m,n (t) occurs at h_m,n (t)=(1/√

a_o^m)h(t/a_o^m -nb_o) when a_o is a minimum for said inverse wavelet transform to converge, b_o being a real constant, m and n being integers.

8. The classifier of claim 6, wherein said critical frame for said prototype wavelet is determined using a non-continuous function within a complement of a space of continuous functions within Hilbert space for said input signal.

9. The classifier of claim 8, wherein said non-continuous function within Hilbert space is a square wave.

10. The classifier of claim 1, wherein said comparator shifts a delay variable of said select inner product transform coefficients to increase a correlation between said select inner product transform coefficients and said exemplar.

11. The classifier of claim 1, wherein said comparison is a Pearson'"'"'s r correlation.

12. A preclassifier for discriminating features of a target in a received signal, comprising:

means for selecting a prototype wavelet corresponding in amplitude and phase per unit of time to a portion of said received signal;

means for resolving said received signal onto a basis of wavelet functions in a wavelet domain by calculating inner product transform coefficients between said received signal and each of said wavelet functions, said wavelet functions corresponding to variations in scale and delay for said prototype wavelet, said variations in scale and delay exceeding a frame condition for said prototype wavelet and said basis, said frame condition enabling an inverse wavelet transform of said calculated inner product transform coefficients to converge on said received signal;

a regionpass filter for identifying a target region of said basis and for passing said inner product transform coefficients within said target region, said target region including said target features in the received signal; and

a separability filter coupled to said regionpass filter for passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable.

13. The preclassifier of claim 12, wherein said basis of wavelet functions is orthonormal.

14. The preclassifier of claim 12, wherein said prototype wavelet is bandpass limited, locally supported, and mean zero.

15. The preclassifier of claim 12, wherein said basis of wavelet functions is overcomplete.

16. The preclassifier of claim 15, wherein said overcomplete basis of wavelet functions is general frame.

17. The preclassifier of claim 16, wherein said frame condition is critical frame, said critical frame indicating the variation for scale and delay for said prototype wavelet where a number of said inner product calculations is a minimum for said inverse wavelet transform to converge.

18. The preclassifier of claim 17, wherein said critical frame for said prototype wavelet h_m,n (t) occurs at h_m, n(t)=(1/√

a_o^m)h(t/a_o^m -nb_o) when a_o is a minimum for said inverse wavelet transform to converge, b_o being a real constant, m and n being integers.

19. The preclassifier of claim 17, wherein said critical frame for said prototype wavelet is determined using a non-continuous function within a complement of a space of continuous functions within Hilbert space for said input signal.

20. The preclassifier of claim 19, wherein said non-continuous function within Hilbert space is a square wave.

21. A filter for passing select inner product transform coefficients from a resolution of a received signal onto a basis of wavelet functions in a wavelet domain, said wavelet functions corresponding to variations in scale and delay for a prototype wavelet, said select inner product transform coefficients including features of a received target, comprising:

a regionpass filter for identifying a target region of said basis and for passing said inner product transform coefficients within said target region, said target region including said features of the received target; and

a separability filter coupled to said regionpass filter for passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable.

22. The filter of claim 21, further comprising an amplitude filter coupled to said separability filter for passing a portion of said select inner product transform coefficients, said portion having an amplitude exceeding a predetermined value.

23. The filter of claim 21, wherein said regionpass filter identifies a target region by correlating said inner product transform coefficients occurring inside a sample region in a scale-delay plane of said wavelet domain with said inner product transform coefficients occurring outside said sample region.

24. The filter of claim 23, wherein said regionpass filter adjusts a size of said sample region to decrease said correlation.

25. The filter of claim 23, wherein said regionpass filter shifts a location of said sample region in said scale-delay plane to decrease said correlation.

26. The filter of claim 23, wherein said correlation is a Pearson'"'"'s r correlation.

27. The filter of claim 23, wherein said sample region has an ellipsoidal shape.

28. A method for classifying features of a target in a received signal, comprising the steps of:

(a) selecting a prototype wavelet corresponding in amplitude and phase per unit of time to a portion of said received signal;

(b) resolving said received signal onto a basis of wavelet functions in a wavelet domain by calculating inner product transform coefficients between said received signal and each of said wavelet functions, said wavelet functions corresponding to variations in scale and delay for said prototype wavelet, said variations in scale and delay exceeding a frame condition for said prototype wavelet and said basis, said frame condition enabling an inverse wavelet transform of said calculated inner product transform coefficients to converge on said received signal;

(c) identifying a target region of said basis, said target region including said target features in the received signal;

(d) passing said inner product transform coefficients within said target region;

(e) passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable; and

(f) comparing said select inner product transform coefficients to an exemplar, said comparison indicating a recognition of said features of the target.

29. A method for preclassifying features of a target in a received signal, comprising the steps of:

(a) selecting a prototype wavelet corresponding in amplitude per unit of time to a portion of said received signal;

(b) resolving said received signal onto a basis of wavelet functions in a wavelet domain by calculating inner product transform coefficients between said received signal and each of said wavelet functions, said wavelet functions corresponding to variations in scale and delay for said prototype wavelet, said variations in scale and delay exceeding a frame condition for said prototype wavelet and said basis, said frame condition enabling an inverse wavelet transform of said calculated inner product transform coefficients to converge on said received signal;

(c) identifying a target region of said basis, said target region including said target features in the received signal;

(d) passing said inner product transform coefficients within said target region; and

(e) passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable.

30. A method for passing select inner product transform coefficients from a resolution of a received signal onto a basis of wavelet functions in a wavelet domain, said wavelet functions corresponding to variations in scale and delay for a prototype wavelet, said select inner product transform coefficients including features of a received target, comprising the steps of:

(a) identifying a target region of said basis, said target region including said received target features; and

(b) passing select inner product transform coefficients in said target region, said select inner product transform coefficients being a scale band within said target region and having a separability between known targets exceeding the separability in other scale bands within said target region, a scale band defined by a set of inner product transform coefficients having a common scale variable.

31. The method of passing select inner product transform coefficients of claim 30, further comprising the step of:

(c) filtering a portion of said select inner product transform coefficients having an amplitude less than a predetermined value.

32. The method of passing select inner product transform coefficients of claim 30, wherein said identifying step comprises the steps of:

(i) selecting a sample region in a scale-delay plane of said wavelet domain;

(ii) correlating said inner product transform coefficients occurring inside said sample region with said inner product transform coefficients occurring outside said sample region;

(iii) adjusting a size of said sample region to decrease said correlation; and

(iv) shifting a location of said sample region in said scale-delay plane to decrease said correlation.