Unsupervised learning neural networks with applications to data fusion

Abstract

An unsupervised learning artificial neural network, DIGNET, is used to design a multi-sensor data fusion system. DIGNET is a self-organizing neural network model with its system parameters analytically determined from self-organization during the learning process. The fast and stable clustering of DIGNET on statistical pattern recognition is used to supplement the decision making on multi-sensor detection problems. Features of the received signals are extracted by using signal processing techniques at each sensor stage before presented to data fusion. The data fusion architecture consists of DIGNET models and decision making algorithms. The function of DIGNET is to perform feature clustering prior to data fusion. The clusters of features created by DIGNET are fused by a decision making algorithm for an integrated decision. Experimental results in a multi-sensor moving target indication system show that data fusion with DIGNET successfully detects and tracks multiple moving targets embedded in clutter.