Abstract

An optical neural network with 128 input and 128 output units is successfully implemented. This system consists of a 128 two-dimensional ferroelectric liquid crystal (FLC) spatial light modulator (SLM) used to present one-dimensional input patterns, a 240 × 220 LCTV to form the interconnections, a CCD camera to detect the output of the optical system, and a personal computer for updating the interconnections and controlling the CCD and SLMs. The algorithm used to train the network is Delta Rule. The size of the network allows us to test its operation on solving practical problems. We trained the system on samples of handwritten Arabic figures and tested with another set of samples. Laboratory results show that this network recognizes these characters with less than 10% error. In the conference we will present details about this network and further results. We will also present a second generation of optoelectronic neural network architecture utilizing semiconductor diode laser and VLSI/FLC modulators to speed up the training and testing cycles of the Delta Rule algorithm.

© 1992 Optical Society of America