Program Contact: Dr Tristrom Cooke Phone: (08) 8302 5846.
Radar imagery enjoys the advantage of being independent from
a passive illumination source, such as sunlight, and thus offers
imaging capability at night and through clouds. By utilising synthetic
aperture processing methods, modern day radar imaging systems
are capable of sub-meter resolution. For defence applications
the operational requirement is to employ SAR imagery as an aid
in finding small targets, and if possible, classify the detected
targets; the requirement is of course an all weather one.
Defence units rely upon a variety of sensor information to locate
and track oppositional forces; the surveillance problem becomes
difficult over large land masses that are sparsely populated.
Modern warfare is dependent upon several types of image data to
aid in the surveillance task. These include optical data, infrared
data, and radar image data. The volume of image data would overwhelm
the available image analysis capabilities unless the imagery is
first prescreened to detect potentially significant targets. The
detection problem increases in difficulty when the targets are
small and the land area is large. Clearly, we desire to maximise
the probability of target detection, the probability of correct
classification, and to minimise the probability of a false alarm,
which are competing goals.
Also studied are Support Vector Machines (SVM's) in the classification
of SAR images. SVM's have been shown to perform better than traditional
classifiers with real world data in many situations and without
the need for the time consuming search of target features to use.
CSSIP's research concentrates on the investigation of the different
effects of error penalties and then extending the SVM's capabilities
to self-select the required error penalties in training.
Previous Projects
JP129 Project Award
Analysts' Detection Support System
Funding for SAR automatic target detection was secured under
Defence Project JP129, Airborne Surveillance for Land Operations.
CSSIP and DSTO/SSD jointly developed feature extraction/target
detection algorithms under a two year contract which commenced
in April 1998. The CSSIP project team included Dr. Jim Schroeder
as Program Manager, Dr. Jingxin Zhang, Senior Research Fellow,
Dr. Tristrom Cooke, Research Fellow, and Dr. Dahong Tang, Research
Fellow. Australian defence requirements included use of SAR images
for detection of small vehicles in remote regions. The high volume
of airborne image data mandated the use of automatic computer-based
pre-screening/processing algorithms to relieve ground based analysts
of viewing all available imagery.
Feature Extraction for Automatic Target Recognition
RLM Systems, Ltd.
Melbourne, Victoria
For high resolution SAR imagery it is possible to automatically
recognise and classify different types of military targets. For
example, a tank may be distinguished from a jeep, thus allowing
the battlefield commander to refine "On the Ground"
decisions for maximum force effectiveness.
CSSIP and RLM undertook a joint study effort to identify critical
target features in high resolution complex SAR imagery. Selected
features would then be input to a pattern classification algorithm
for Automatic Target Recognition.
PUBLICATIONS:
Cooke T and Peake M
The optimal classification using a linear discriminant for two
point classes having known mean and covariance, Journal of Multivariate
Analysis, Vol.82, No.2, pp 379-394, August 2002.
Cooke T
A note on the classification error of an SVM in one dimension,
Proceedings of IDC 2002, Adelaide, Australia, February 2002.
Cooke T
Two variations on Fisher's linear discriminant for pattern recognition,
IEEE Transactions on Pattern Analysis and Machine Intelligence,
Vol.24, No.2, pp 268-273, February 2002.
Tang D, Schroeder J, Zhang J and Cooke T
A New Approach of Regressionn via Support Vector Machines WoSPA
2000, Brisbane, December 2000.
Schroeder J and Gunawardena A
Speckle reduction in SAR imagery for enhanced automatic target
detection,
EURASIP Signal Processing Journal (Special Issue - Defence Signal
Processing),
Accepted for special issue, 2000.
Cooke T, Redding N, Schroeder J and Zhang J
Comparison of selected features for target detection in synthetic
aperture radar imagery, Digital Signal Processing, Vol.10, No.4,
pp 286-296, October 2000.
Cooke T, Redding N, Schroeder J and Zhang J
Target discrimination in complex synthetic aperture radar imagery,
Asilomar 2000, Monterey, California, October 2000.
Zhang J, Schroeder J, Redding N, Cooke T and Tang D
Singular value features of images, Proceedings of the SPIE Conference
on Visual Communications and Image Processing 2000, Perth, Australia,
June 2000, Vol.4067, Pt I, pp.894-903.
Cooke T
A Radon transform derivative method for faint trail detection
in SAR imagery, DICTA'99 conference proceedings, pp.31-34, December
1999.
Schroeder J
Multiscale modeling for manmade object discrimination in synthetic
aperture radar imagery, Proceedings of the Asilomar Conference
on Signals and Systems,
24-27 October, 1999.
Howard D and Schroeder J
Multiscale Models for Target Detection and Background Discrimination
in Synthetic Aperture Radar Imagery, Digital Signal Processing:
A Review Journal, July, 1999.
Bose, T., Xu, G-F, and Schroeder, J.,
Image Enhancement Using an EDS Adaptive Filter, ISCAS'99, Orlando,
Florida, May, 1999.
Schroeder J and Howard D
Multiscale modelling for target detection in complex synthetic
aperture radar imagery,
Asilomar'98, Monterey, CA, Nov'98.
Bose T, Campbell E and Schroeder J
A 2-D Switched Mean/Median Filter For Image Restoration
ISPACS'98, Melbourne, Vic, Nov'98.
Gunawardena A and Schroeder J
Polynomial Hough Transform Based Feature Extraction From SAR Imagery,
EUSAR'98, Friedrichshafen, Germany, 25-27 May, 1998, pp. 273-276.
|
