IMPROVING THE WORK OF CONVOLUTIONAL NEURAL NETWORKS FOR RECOGNITION OF SOIL CONDITION IMAGES IN AGRICULTURE

Abstract

In modern agriculture, the greatest problem is the preservation of yield, which depends on many factors. One of the most important is the condition of the soil. There are several types of soil condition parameters, namely chemical, physical, biological, sanitary-bacteriological, and erosive. Each of the above conditions must be evaluated separately and in combination with others. The parameters are partly determined by laboratory tests, and partly in field conditions. Laboratory studies are performed by specialists in specific scientific fields and are based on a set of data collected in the field. Therefore, it is the field studies that are essential for determining the final result regarding the condition of a specific soil area.

Effective assessment of the agrophysical condition of soil located on large territories requires regular comprehensive, interdisciplinary research.

Such an analysis is a time-consuming and lengthy process, especially if it is necessary to study in detail not only the condition of the soil, but also to predict its parameters taking into account various conditions.

This article describes the problems of soil condition research and proposes ways to solve them. It considers current methods, tools, and devices of control. Emphasis is placed on the modern use of unmanned aerial vehicles and the methods of pre-processing soil condition images. Subsequent image processing using convolutional neural networks follows. The scientific contribution of the article is the proposed use of the influence operator in the processing workflow.