The paper presents a neural network study of the data of wheat seed quality. It is established that the analysis of bioelectrical signals of wheat seeds based on a neural network can be used in practice for the solution of two problems - diagnostics of seed material quality and the evaluation of cleaning line quality (separation into fractions). The paper presents the results of initial data preparation, formation of a neural network, analysis of training data for two practical problems of classification. It was established using a neural network that there is a nonlinear dependence of the membrane potential maximum value and the signal rise time on the seed yield. The model makes it possible to predict yield in terms of the seed material quality. A nonlinear dependence of the maximum membrane potential, the signal rise time of wheat seeds and the seeds variety to one or another faction (speed of separation into the fractions in this example) was also established in this paper. Studies have shown that the seeds variety is an important informative feature for solving the problem of classifying seeds by fractions. Therefore, it is necessary to conduct additional studies with other wheat seeds varieties to apply this method in practice.
Read full abstract