CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
REALIZATION OF THE PERCEPTRON MODEL USING AN EIGHT-BIT MICROCONTROLLER
 |
Read | Chernukhin Dmitriy A., Shurshev Valeriy F. REALIZATION OF THE PERCEPTRON MODEL USING AN EIGHT-BIT MICROCONTROLLER // Caspian journal : management and high technologies. — 2020. — №3. — pp. 103-113. |
Chernukhin Dmitriy A. - Astrakhan State Technical University, madlabman@mail.ru
Shurshev Valeriy F. - Astrakhan State Technical University, v.shurshev@mail.ru
Most modern stand-alone devices are developed on the basis of microcontrollers with 32-bit architecture. At the same time, the use of the "younger" family (eight-bit microcontrollers) has a number of advantages: it allows to reduce the cost of components; to increase the energy efficiency of the device, and, hence, its battery life; achieve higher reliability of the device. For definiteness, a software and hardware implementation of the perceptron model based on the use of an 8-bit microcontroller is considered. A neural network has been developed for the task of recognizing numbers displayed on a seven-segment indicator. A perceptron topology with one hidden layer was used. It was trained using the back-propagation method. Highlighted the characteristics of the neural network, critical from the point of view of implementation on an 8-bit microcontroller. Estimates have been developed to determine various parameters of the neural network for specific characteristics of the 8-bit microcontroller. An assessment of the reaction speed of a network built on the basis of the 8-bit microcontroller is given. The conclusion is made about the admissibility and expediency of using an 8-bit microcontroller for solving this and similar problems.
Key words: eight-bit microcontroller, perceptron, hardware implementation of a neural network, artificial neuron, layer, input signal, output signal, activation function, sigmoid function, training set, operation with floating point numbers, performance evaluation