Optical implementation of neocognitron and its applications to radar signature discrimination

摘要

Abstract: A feature extraction based optoelectronic neuralnetwork is introduced. The system implementationapproach applies the principle of the neocognitronparadigm first introduced by Fukushima. A multichannelcorrelator is used as a building block of a genericsingle layer of the neocognitron for shift-invariantfeature correlation. Multilayer processing is achievedby iteratively feeding back the output of the featurecorrelator to the input spatial light modulator.Successful pattern recognition with intra-class faulttolerance and inter-class discrimination is achievedusing this optoelectronic neocognitron. Detailed systemanalysis is described. Experimental demonstration ofradar signature processing is also provided.!27