PROJECT OF A SOFTWARE COMPLEX FOR THE IMPLEMENTATION OF AN APPLICATION FOR THE IDENTIFICATION OF MEDICINAL PLANTS

Abstract

A software product has been developed designed to improve the search for necessary plants and reduce unforeseen cases when using the wrong medicinal plants. Analyzes of existing analogues have been made, and their shortcomings have been considered. A programming language, a software product, and a neural network were selected with the help of expert evaluation. The basic architecture and activities of the own product have been designed. With the help of SADT design methods, the basic principle of the system was designed, which includes the identification of medicinal plants by image. Structured chart of precedents, which reflects the system of identification of medicinal plants by image. A sequence diagram has been constructed, which contains a system for identifying medicinal plants by image. To build a neural network, 60 types of plants were used - 38,815 images, which provide a clear identification that will protect people from dangerous mistakes when using medicinal plants. In this work, such tasks as identification and recognition of images, in the specific case of images of plants from photographs, are considered. Creation of an architecture based on InceptionV3, to create a model of a pre-trained neural network. A plant image classifier based on a pre-trained neural network. The images in the learning network were divided into categories depending on the part of the plant depicted on them: Entire (plant as a whole), Branch, Flower, Fruit (fruit or berry), LeafScan, Leaf, Stem. For each of these categories, the most suitable method of pre-processing has been selected. Trained the network as a whole using augmentation and the Imgaug library. Those transformations that occur in real life were chosen for augmentation. Top-metrics were used to measure the ability of the model to give the true class of the plant in the list of the most probable classes.