Using a neural network to identify diseased potato plants

Abstract

Relevance. In order to obtain high-quality seed material, seed farms should pay great attention to crop cultivation technologies. At the same time, an important role is played by the implementation of such breeding measure as phyto-cleaning of breeding and seed plots, in order to identify and eliminate infected plants. However, it is worth noting the fact that the implementation of such measure requires the presence of highly qualified specialists capable of detecting plant diseases at early stages. However, currently there is a shortage of such employees in agriculture, and therefore the development of innovative digital technologies aimed at detecting infected plants is an urgent task. Currently, machine vision and neural network technologies designed to solve such problems are actively developing.Methods. As part of the research, existing machine vision technologies were analyzed, as well as developed machine learning technologies. Then, based on the analysis, a software package based on a convolutional neural network was developed. During the training and testing of the neural network, framing technologies, affine transformation methods, information and logical analysis of the initial information were used.Results. To determine the quality of the software package for the identification of diseased potato plants, a series of tests was conducted. During the research, the accuracy with which the distribution of plants to a particular group was carried out was evaluated. The analysis of the results showed that the chosen neural network design successfully coped with the experimental task. At the same time, for the further development of this direction, it is necessary to create an extensive information base on potato diseases. That will allow in the future to develop a software and hardware complex for the analysis of potato plantings and the identification of infected plants in real time.