The two major difficulties associated with SAR (synthetic aperture radar) image change detection are addressed. These are the removal of speckle noise and the registration of information between images. Due to the unpredictable nature of the aircraft track, the problem of image registration is severe in airborne SAR imagery. Autofocus techniques are used to measure residual aircraft motions, thus allowing the production of large geometrically accurate images. The problem of speckle reduction is approached in two ways. The first technique applies an intensity segmentation algorithm to each image. The regions generated by the segmentation are then compared by the change detection algorithm. An alternative approach is to use neural network methods to learn the speckle removal and region generation task. To reduce this problem to a manageable size a factorization method for the multilayer-perceptron has been invented.<<ETX>>
[1]
Richard P. Lippmann,et al.
An introduction to computing with neural nets
,
1987
.
[2]
Anthony Freeman,et al.
The Prediction of Geometric Distortions in Airborne Synthetic Aperture Radar Imagery from Autofocus Measurements
,
1987,
IEEE Transactions on Geoscience and Remote Sensing.
[3]
E. Jakeman,et al.
Non-Gaussian models for the statistics of scattered waves
,
1988
.
[4]
J. W. Wood.
The removal of azimuth distortion in synthetic aperture radar images
,
1988
.
[5]
R. G. White,et al.
Change detection in SAR imagery
,
1990
.
[6]
R. G. White,et al.
Radar clutter classification based on noise models and neural networks
,
1990,
IEEE International Conference on Radar.
[7]
T. J. Stonham,et al.
Guide to pattern recognition using random-access memories
,
1979
.