A new method for diagnostics of plant diseases based on digital description of leaf images and forward propagation neural

Tutygin Vladimir Semenovich, Prokofiev Oleg Valerievich

A new approach to building a software system for diagnosing plant diseases from leaf images based on digital descriptions of leaf images in the form of sets of Haralik parameters in the form of png images for each of the color components of leaf images and a feed-forward neural network. The task of preprocessing is to prepare a data set (leaf images) by normalizing the sizes and generating digital descriptions as input to the neural network. This is achieved by using GLCM and calculating a set of 4 Haralik parameters that reflect the features of the image texture and at the same time have a much lower dimension. In this regard, convolution operations in the structure of the neural network are excluded, which significantly reduces the time spent on training the neural network and diagnostics. Real images of leaves for all plant diseases are used to train the neural network. The large amount of initial data required for training is created using randomization.

Keywords:Photo image normalization, GLCM - matrix, Haralik parameters, digital descriptions of leaf images, feed-forward neural network.