Improvement of Optical and Electronic Neuron Net Identification of Large Set of Input Signal Corrupted by Noise

An acceptable level of the input image noise during the associative image recognition is analyzed within the framework of the digital simulation of a singlelayer optical and electronic neuron net based on the large-sized optical vector-and-matrix multiplier. It is shown that with the help of uniting the similar reference memory images into groups and constructing the individual weight neuron matrices for each group and one initial matrix for a set of generalized prototypes of each group it is possible to increase considerably the acceptable noise level for the input images being identified without significant increase of time required for their complete and errorless recovery. Peculiarities of instruction of such a neuron net are considered.