CLASSIFICATION OF SUBSURFACE OBJECTS USING SINGULAR VALUES DERIVED FROM SIGNAL FRAMES

• US 20130082858A1
• Filed: 08/26/2011
• Published: 04/04/2013
• Est. Priority Date: 08/26/2010
• Status: Active Grant

• Current Assignee: Lawrence Livermore National Security LLC (Government of the United States of America)
• Original Assignee: Lawrence Livermore National Security LLC (Government of the United States of America)
• Inventors: Chambers, David H., Paglieroni, David W.
• Granted Patent: US 8,717,223 B2
• Time in Patent Office: Days
• US Class Current: 342/22
• CPC Class Codes: G01S 13/003 Bistatic radar systems; Multistatic radar systems
• CPC Class Codes: G01S 13/885 for ground probing prospecting or detecting using electromagnetic waves G01V3/12
• CPC Class Codes: G01S 13/90 using synthetic aperture techniques , e.g. synthetic aperture radar [SAR] techniques
• CPC Class Codes: G01V 3/12 operating with electromagnetic waves operating with millimetre waves G01V8/005
• CPC Class Codes: G01V 3/15 specially adapted for use during transport, e.g. by a person, vehicle or boat
• CPC Class Codes: G06N 99/00 Subject matter not provided for in other groups of this subclass

Abstract

The classification system represents a detected object with a feature vector derived from the return signals acquired by an array of N transceivers operating in multistatic mode. The classification system generates the feature vector by transforming the real-valued return signals into complex-valued spectra, using, for example, a Fast Fourier Transform. The classification system then generates a feature vector of singular values for each user-designated spectral sub-band by applying a singular value decomposition (SVD) to the N×N square complex-valued matrix formed from sub-band samples associated with all possible transmitter-receiver pairs. The resulting feature vector of singular values may be transformed into a feature vector of singular value likelihoods and then subjected to a multi-category linear or neural network classifier for object classification.

Claims

1. A method in a computing device for classifying objects, the method comprising:

providing a classifier that identifies a class for an object, the classifier inputting a feature vector representing the object and outputting an indication of the class for the object;

receiving a signal frame for a detected object, the signal frame representing return signals acquired by receivers based on signals emitted by transmitters operating in multistatic mode, the signal frame including, for each transmitter and receiver pair, a plurality of samples acquired by the receiver at different sampling times for the signal emitted by the transmitter;

transforming the samples of the return signal for each transmitter and receiver pair to a return spectrum in the frequency domain;

generating singular values from the samples of the return spectrum of each transmitter and receiver pair;

generating a feature vector representing the detected object from the generated singular values; and

applying the classifier to the generated feature vector to identify the class of the detected object.

2. The method of claim 1 wherein the signal frame is generated by a detection system that detects a presence of subsurface objects.

3. The method of claim 1 wherein the signal frame is represented as an N×

N matrix where N is a number of transceivers with each element of the matrix having the samples for transmitter and receiver pairs.

4. The method of claim 1 wherein the transform of return signals to the return spectra applies a Fast Fourier Transform to the return signals of each transmitter and receiver pair.

5. The method of claim 1 wherein the return spectra include a sample for each of a plurality of frequency bands and the generating of the singular values is performed by applying a singular value decomposition to the return spectral samples of a frequency sub-band.

6. The method of claim 1 wherein the classifier is generated by:

providing training data that includes, for each of a plurality of training objects, a signal frame and a class for that training object;

for each training object, generating a feature vector for that training object from the signal frame for that training object; and

training the classifier based on the generated feature vectors and classes of the training objects.

7. The method of claim 1 wherein the classifier is selected from a group consisting of a linear classifier, a stochastic classifier, and a hybrid linear-stochastic classifier.

8. A computer-readable storage medium containing computer-executable instructions for controlling a computer to generate a classifier for classifying objects detected below a surface, by a method comprising:

providing training data that includes, for each of a plurality of training objects, a signal frame and a class for that training object, each signal frame representing return signals acquired by receivers based on signals emitted by transmitters in multistatic mode, each signal frame including, for each transmitter and receiver pair, a plurality of samples acquired by the receiver at different sampling times for the signal emitted by the transmitter;

for each training object, generating a feature vector for that training object from the signal frame for that training object by;

transforming the return signal samples of each transmitter and receiver pair to return spectral samples in the frequency domain; and

generating singular values for the return spectral samples of each transmitter and receiver pair; and

training the classifier based on the generated feature vectors and classes of the training objects.

9. The computer-readable storage medium of claim 8 wherein the signal frames are generated by a detection system that detects the presence of training objects having known classes.

10. The computer-readable storage medium of claim 9 wherein the detection system has transmitter and receiver pairs operating in a multistatic mode.

11. The computer-readable storage medium of claim 8 wherein the return spectral samples include a sample for each of a plurality of frequency bands and the generating of the singular values is performed by applying a singular value decomposition to the transformed samples of a frequency sub-band.

12. The computer-readable storage medium of claim 8 wherein the generating of the feature vector focuses on samples of a signal frame that contribute to the detection of a training object.

13. The computer-readable storage medium of claim 8 including:

after generating the feature vectors, generating a likelihood function for each class from the feature vectors, each likelihood function for a class to indicate a likelihood that a feature vector represents an object in that class; and

for each training object, generating a likelihood feature vector using the generated likelihood functions,

wherein the training of the classifier includes training a linear classifier based on the likelihood feature vectors and the classes of the training objects.

14. A computing device for classifying objects, the computing device comprising:

a component that receives a signal frame for a detected object, the signal frame representing return signals acquired by receivers based on signals emitted by transmitters in multistatic mode, the signal frame including, for each transmitter and receiver pair, a plurality of samples acquired by the receiver at different sampling times for the signal emitted by the transmitter;

a component that generates a feature vector for the detected object by transforming the return signal samples of each transmitter and receiver pair to return spectral samples in the frequency domain and generates singular values for the return spectral samples of each transmitter and receiver pair; and

a component that applies a classifier to the generated feature vector to identify the class of the detected object.

15. The computing device of claim 14 wherein the return spectral samples include a sample for each of a plurality of frequency bands and the generating of the singular values is performed by applying a singular value decomposition to the transformed samples of a frequency sub-band.

16. The computing device of claim 14 wherein the component that generates the feature vector focuses on samples of the signal frame that contribute to the detection of the detected object.

17. The computing device of claim 14 wherein the classifier is a hybrid linear-stochastic classifier.