RECOGNITION OF TEXT PHRASES DISTORTED BY INTERFERENCE BY BACK PROPAGATION NEURAL NETWORK

Abstract

The paper takes into consideration risk systems that can use not only in nuclear reactions but other plants with frequent risks for people's life, such as mining, and other. Such facilities apply information systems in which take place exchange text messages through free space. The main problem of information radio reception is an increasing number of emitting means that equal the increase of noise level receiving set. As an additional means of processing distorted textual information, it is proposed to use a neural network, which must be pre-configured. For analysis, the back propagation neural network was selected. The adjustment is carried out by an algorithm assuming a double differentiation of the error function, which ensures a high network convergence rate. Learning is stopped according to the total criterion for the deviation of the output signal from the reference. The paper formulates the conditions of quadratic convergence of the back propagation network with one new tuning procedure, and also offers examples of the construction of a neural network for recognizing a text message in various reception conditions. The fed to the neural network is sequence of the letters of English alphabet. A feature of the structure of the neural network that provides correct recognition is the use of completely nonlinear neurons. Comparison of options for the structure of the neural network when recognizing text phrases is carried out according to indicators of the probability of recognition, error, and training time. The established properties of the neural network are useful in the design of efficient information system.