Feature discovery on segmented objects in SAR imagery using self-organizing neural networks.

摘要

In this paper we investigate the applicability of the feature extraction mechanisms found in the neurophysiology of mammals to the problem of object recognition in synthetic aperture radar imagery. Our approach is to present multiple views of objects to be recognized to a two-stage self-organizing neural network architecture. The first stage, a two-layer Neocognitron, performs feature extraction in each layer The resulting feature vectors are presented to the second stage, an ART-2A classifier self-organizing neural network which clusters the features into multiple object categories. The feature extraction operators resulting from the self-organization process are compared to the feature extraction mechanisms found in the neurophysiology of vision. In a previous paper, the Neocognitron was trained on raw SAR imagery. The architecture was able to recognize a simulated vehicle at arbitrary azimuthal orientations at a single depression angle while rejecting clutter as well as other vehicles. Feature extraction on raw imagery yielded features that were robust but very difficult to interpret. In this paper we report the results of some new experiments in which the self-organization process is applied separately to shadow and bright returns from objects to be recognized. Feature extraction on shadow returns yield oriented contrast edge operators suggestive of bipartite simple cells observed in the striate cortex of mammals. Feature extraction on the specularity patterns in bright returns yield a collection of operators resembling a twodimensional Haar basis set. We compare the performance of the earlier two-stage neural network trained on raw imagery with a modified network using the new feature set.