Sparse adaptive filter

• US 9,213,915 B2
• Filed: 03/15/2013
• Issued: 12/15/2015
• Est. Priority Date: 03/15/2013
• Status: Active Grant

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Sommese, Anthony, Flanders, Bradley A., Robinson, Ian S.
• Primary Examiner(s): Park, Chan
• Assistant Examiner(s): RICE, ELISA M
• Application Number:
  US13/837,866
• Publication Number:
  US 20150161475A1
• Time in Patent Office: 1,005 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 18/2113 by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
• CPC Class Codes: G06V 10/431 Frequency domain transformation; Autocorrelation
• CPC Class Codes: G06V 10/46 Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features colour feature extraction G06V10/56
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space
• CPC Class Codes: G06V 20/13 Satellite images

Abstract

The disclosure provides a filtering engine for selecting a subset of hyperspectral imaging wavebands having information useful for detecting a target in a scene. Selecting these wavebands, called “sparse bands,” is an iterative process. One or more search techniques of varying computational complexity are used in the process. The techniques rely on various selection criteria, including a signal to clutter ratio that measures the “goodness” of band selection. A convenient example of the filtering engine uses several of the techniques together in a layered approach. In this novel approach, simpler computational techniques are applied, initially, to reduce a number of bands. More computationally intensive techniques then search the reduced band space. Accordingly, the filtering engine efficiently selects a set of sparse bands tailored for each target and each scene, and maintains some of the detection capability provided with a full set of wavebands.

Claims

1. A method for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, the method comprising:

in a filtering engine provided with a plurality of spectra of a target and a covariance matrix of a scene, for each spectrum in the plurality of spectra of the target;

computing a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

selecting a subset of wavebands from the full set of wavebands;

computing a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

comparing the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determining whether the subset of wavebands is a good enough set of wavebands based on the comparison; and

providing the good enough set of wavebands to detect the target in the scene;

wherein selecting the subset of wavebands includes determining, in a number of iterations, reduced sets of wavebands by;

computing a matched filter weight for each waveband in an input set of wavebands, the input set of wavebands of a first iteration being the full set of wavebands and the input set of wavebands of subsequent iterations being the reduced sets of wavebands from previous iterations;

deleting a waveband with the smallest matched filter weight from the input set of wavebands resulting in a reduced set of wavebands; and

ending the iterations when the reduced set of wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the reduced set of wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined.

2. The method of claim 1, wherein selecting the subset of wavebands includes:

computing a matched filter weight for each waveband in the full set of wavebands; and

deleting wavebands with the smallest matched filter weights from the full set of wavebands resulting in a reduced set of wavebands, the reduced set of wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined.

4. The method of claim 1, wherein determining the good enough set of wavebands includes determining a set of wavebands providing a signal-to-clutter ratio that is better than or equal to a specified percentage of the signal-to-clutter ratio of the full set of wavebands.

5. The method of claim 1, wherein determining the good enough set of wavebands includes determining a fixed number of wavebands belonging to the good enough set of wavebands.

6. The method of claim 1, wherein determining the good enough set of wavebands includes determining the good enough set of wavebands having a fixed signal-to-clutter ratio or having fixed fraction of the total signal-to-clutter ratio for the full set of wavebands.

7. The method of claim 1, further comprising:

in the filtering engine provided with a plurality of pixels in the scene,

processing each pixel in the plurality of pixels with the good enough set of wavebands;

identifying which pixels in the scene are candidate detections based on the processed pixels;

processing each of the candidate detections with the full set of wavebands; and

detecting the target in the scene based on the processed candidate detections.

8. The method of claim 1, further comprises:

in the filtering engine provided with a plurality of pixels in the scene,

processing each pixel in the plurality of pixels with the good enough set of wavebands;

identifying which pixels in the scene are candidate detections based on the processed pixels;

processing each of the candidate detections with the full set of wavebands using a localized scene covariance; and

detecting the target in the scene based on the localized scene covariance processed candidate detections.

3. A method for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, the method comprising:

in a filtering engine provided with a plurality of spectra of a target and a covariance matrix of a scene, for each spectrum in the plurality of spectra of the target;

computing a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

selecting a subset of wavebands from the full set of wavebands;

computing a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

comparing the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determining whether the subset of wavebands is a good enough set of

wavebands based on the comparison; and

providing the good enough set of wavebands to detect the target in the scene;

wherein selecting the subset of wavebands includes determining, in a number of iterations, included sets of wavebands by;

selecting a waveband from an group of wavebands which when combined with wavebands of an input set of wavebands have, collectively, the highest signal-to-clutter ratio, the group of wavebands of a first iteration being a full set of wavebands and groups of wavebands of subsequent iterations being the full set of wavebands less wavebands selected in previous iterations;

adding the selected waveband to the input set of wavebands resulting in intermediate set of wavebands, the input set of wavebands of a first iteration having a number of wavebands, including zero, and input sets of wavebands of subsequent iterations being included sets of wavebands determined in previous iterations;

deleting one or more wavebands from the intermediate set of wavebands contributing the least to a signal-to-clutter ratio of the intermediate set of wavebands resulting in a set of included wavebands; and

ending the iterations when the set of included wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the set of included wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined.

12. The method of claim 3, wherein selecting the subset of wavebands includes:

computing a matched filter weight for each waveband in the full set of wavebands; and

deleting wavebands with the smallest matched filter weights from the full set of wavebands resulting in a reduced set of wavebands, the reduced set of wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined.

13. The method of claim 3, wherein determining the good enough set of wavebands includes determining a set of wavebands providing a signal-to-clutter ratio that is better than or equal to a specified percentage of the signal-to-clutter ratio of the full set of wavebands.

14. The method of claim 3, wherein determining the good enough set of wavebands includes determining a fixed number of wavebands belonging to the good enough set of wavebands.

15. The method of claim 3, wherein determining the good enough set of wavebands includes determining the good enough set of wavebands having a fixed signal-to-clutter ratio or having fixed fraction of the total signal-to-clutter ratio for the full set of wavebands.

16. The method of claim 3, further comprising:

in the filtering engine provided with a plurality of pixels in the scene,

processing each pixel in the plurality of pixels with the good enough set of wavebands;

identifying which pixels in the scene are candidate detections based on the processed pixels;

processing each of the candidate detections with the full set of wavebands; and

detecting the target in the scene based on the processed candidate detections.

17. The method of claim 3, further comprising:

in the filtering engine provided with a plurality of pixels in the scene,

processing each pixel in the plurality of pixels with the good enough set of wavebands;

identifying which pixels in the scene are candidate detections based on the processed pixels;

processing each of the candidate detections with the full set of wavebands using a localized scene covariance; and

detecting the target in the scene based on the localized scene covariance processed candidate detections.

9. A method for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, the method comprising:

in a filtering engine provided with a plurality of spectra of a target and a covariance matrix of a scene, for each spectrum in the plurality of spectra of the target;

computing a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

selecting a subset of wavebands from the full set of wavebands;

computing a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

comparing the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determining whether the subset of wavebands is a good enough set of

wavebands based on the comparison; and

providing the good enough set of wavebands to detect the target in the scene;

wherein selecting the subset of wavebands includes;

computing a matched filter weight for each waveband in the full set of wavebands; and

deleting wavebands with the smallest matched filter weights from the full set of wavebands resulting in a reduced set of wavebands;

in a first round of iterations, determining further reduced sets of wavebands including;

in each iteration of the first round of iterations, computing a matched filter weight for each waveband in a first round input set of wavebands, the first round input set of wavebands of a first iteration being the reduced set of wavebands determined from the full set of wavebands and the first round input sets of wavebands of subsequent iterations being the further reduced sets of wavebands from previous iterations of the first round of iterations;

deleting a waveband with the smallest matched filter weight from the first round input set of wavebands resulting in a further reduced set of wavebands; and

ending the first round of iterations when the further reduced set of wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands;

in a second round of iterations, determining included sets of wavebands including;

in each iteration of the second round of iterations, selecting a waveband from an group of wavebands which when combined with wavebands of a second round input set of wavebands have, collectively, the highest signal-to-clutter ratio, the group of wavebands of a first iteration being the further reduced set of wavebands determined in the first round of iterations and groups of wavebands of subsequent iterations being the further reduced set of wavebands less wavebands selected in previous iterations of the second round of iterations;

adding the selected waveband to the second round input set of wavebands resulting in intermediate set of wavebands, the second round input set of wavebands of a first iteration having a number of wavebands, including zero, and second round input sets of wavebands of subsequent iterations being included sets of wavebands determined in previous iterations of the second round of iterations;

deleting one or more wavebands from the intermediate set of wavebands contributing the least to a signal-to-clutter ratio of the intermediate set of wavebands resulting in a set of included wavebands; and

ending the second round of iterations when the set of included wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the set of included wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined.

10. A system for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, the system comprising:

a memory having computer executable instructions thereupon;

at least one interface receiving a plurality of spectra of a target and a covariance matrix of a scene;

a filtering engine coupled to the memory and the at least one interface, the computer executable instructions when executed by the filtering engine cause the filtering engine to, for each spectrum in the plurality of spectra of the target;

compute a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

select a subset of the full set of wavebands, including determining, in a number of iterations, reduced sets of wavebands by

computing a matched filter weight for each waveband in an input set of wavebands, the input set of wavebands of a first iteration being the full set of wavebands and the input set of wavebands of subsequent iterations being the reduced sets of wavebands from previous iterations,

deleting a waveband with the smallest matched filter weight from the input set of wavebands resulting in a reduced set of wavebands, and

ending the iterations when the reduced set of wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the reduced set of wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined;

compute a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

compare the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determine whether the subset of wavebands is a good enough set of wavebands based on the comparison; and

provide the good enough set of wavebands to detect the target in the scene.

11. A tangible non-transitory computer-readable storage medium having computer readable instructions stored therein for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, which when executed by one or more processors provided with a plurality of spectra of a target and a covariance matrix of a scene cause the one or more processors to, for each spectrum in the plurality of spectra of the target:

compute a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

select a subset of the full set of wavebands, including determining, in a number of iterations, reduced sets of wavebands by

computing a matched filter weight for each waveband in an input set of wavebands, the input set of wavebands of a first iteration being the full set of wavebands and the input set of wavebands of subsequent iterations being the reduced sets of wavebands from previous iterations,

deleting a waveband with the smallest matched filter weight from the input set of wavebands resulting in a reduced set of wavebands, and

ending the iterations when the reduced set of wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the reduced set of wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined;

compute a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

compare the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determine whether the subset of wavebands is a good enough set of wavebands based on the comparison; and

provide the good enough set of wavebands to detect the target in the scene.

18. A system for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, the system comprising:

a memory having computer executable instructions thereupon;

at least one interface receiving a plurality of spectra of a target and a covariance matrix of a scene;

a filtering engine coupled to the memory and the at least one interface, the computer executable instructions when executed by the filtering engine cause the filtering engine to, for each spectrum in the plurality of spectra of the target;

compute a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

select a subset of the full set of wavebands, including determining, in a number of iterations, included sets of wavebands by

selecting a waveband from an group of wavebands which when combined with wavebands of an input set of wavebands have, collectively, the highest signal-to-clutter ratio, the group of wavebands of a first iteration being a full set of wavebands and groups of wavebands of subsequent iterations being the full set of wavebands less wavebands selected in previous iterations,

adding the selected waveband to the input set of wavebands resulting in intermediate set of wavebands, the input set of wavebands of a first iteration having a number of wavebands, including zero, and input sets of wavebands of subsequent iterations being included sets of wavebands determined in previous iterations,

deleting one or more wavebands from the intermediate set of wavebands contributing the least to a signal-to-clutter ratio of the intermediate set of wavebands resulting in a set of included wavebands, and

ending the iterations when the set of included wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the set of included wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined;

compute a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

compare the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determine whether the subset of wavebands is a good enough set of wavebands based on the comparison; and

provide the good enough set of wavebands to detect the target in the scene.

19. A tangible non-transitory computer-readable storage medium having computer readable instructions stored therein for selecting a subset of hyperspectral imaging wavebands to detect a target in a scene, which when executed by one or more processors provided with a plurality of spectra of a target and a covariance matrix of a scene cause the one or more processors to, for each spectrum in the plurality of spectra of the target:

compute a signal-to-clutter ratio of a full set of wavebands representing the probability of detecting the target viewed in the scene;

select a subset of the full set of wavebands, including determining, in a number of iterations, included sets of wavebands by

selecting a waveband from an group of wavebands which when combined with wavebands of an input set of wavebands have, collectively, the highest signal-to-clutter ratio, the group of wavebands of a first iteration being a full set of wavebands and groups of wavebands of subsequent iterations being the full set of wavebands less wavebands selected in previous iterations,

adding the selected waveband to the input set of wavebands resulting in intermediate set of wavebands, the input set of wavebands of a first iteration having a number of wavebands, including zero, and input sets of wavebands of subsequent iterations being included sets of wavebands determined in previous iterations,

deleting one or more wavebands from the intermediate set of wavebands contributing the least to a signal-to-clutter ratio of the intermediate set of wavebands resulting in a set of included wavebands, and

ending the iterations when the set of included wavebands has a specified number of wavebands, specified signal-to-clutter ratio or has a specified fraction of the total signal-to-clutter ratio for the full set of wavebands, the set of included wavebands being the subset of wavebands that is selected and from which the good enough set of wavebands is determined;

compute a signal-to-clutter ratio of the selected subset of wavebands representing the probability of detecting the target viewed in the scene using the selected subset of wavebands;

compare the signal-to-clutter ratio of the subset of wavebands to the signal-to-clutter ratio of the full set of wavebands;

determine whether the subset of wavebands is a good enough set of wavebands based on the comparison; and

provide the good enough set of wavebands to detect the target in the scene.