<title>Character recognition using novel optoelectronic neural network</title>

Abstract

BROOKS AIR FORCE BASE, TEXAS NOTICESThis technical paper is published as received and has not been edited by the technical editing staff of the Armstrong Laboratory.The interpretations, conclusions, recommendations, and opinions are those of the author and are not necessarily endorsed by the U.S. Air Force or the Department of Defense.When Government drawings, specifications, or other data are used for any purpose other than in connection with a definitely Government-related procurement, the United States Government incurs no responsibility or any obligation whatsoever. The fact that the Government may have formulated or in any way supplied the said drawings, specifications, or other data, is not to be regarded by Implication, or otherwise in any manner construed, as licensing the holder, or any other person or corporation; or as conveying any rights or permission to manufacture, use, or sell any patented invention that may in any way be related thereto.The Office of Public Affairs has reviewed this paper, and it is releasable to the National Technical Information Service, where it will be available to the general public, including foreign nationals. This paper has been reviewed and is approved for publication. Approved for public release; distribution is unlimited. ABSTRACT (Aaxnmum 200 words)A novel optoelectronic neural network has been designed and constructed to recognize a set of characters from the alphabet. The network consists of a 15Xl binary input vector, two optoelectronic vector matrix multiplication layers, and a 15X1 binary output layer. The network utilizes a pair of custom fabricated Spatial Light Modulators (SLMs) with 120 levels of gray scale per pixel. The SIMs realize the matrix weights. Previous networks of this type were hampered by limited levels of gray scale and the need to use two separate weight masks (matrices) per layer. The weight masks are operated in unipolar mode. This allows both positive and negative weights to be realized from the same mask. A hard limiting function is used for the network's nonlinearity. A modification of Widrow's lesser known MR2 training algorithm is used to train the network. Furthermore, the network introduces a novel lens-free crossbar matrix-vector multiplier. multiplier.