Driver Assistance System for Agricultural Machinery for Obstacles Detection Based on Deep Neural Networks

Abstract

The article deals with an important scientific task of developing and researching models and methods of deep learning for the purpose of detecting and recognizing objects in environmental images in agricultural intelligent transport systems. The direction and obstacles are determined based on the processing of video information generated by the cameras of the onboard system, taking into account the operations performed, such as plowing, harrowing, weeding and fertilizing. Electric power poles, trees, rocks, bird nests, animals, people and field roads are considered as obstacles. The relevance of the introduction of intelligent transport systems considered in the article is determined by the processes of digital transformation of the economy in this industry. The latter are defined by the concept of “Smart Agriculture”, one of the directions of which is “Smart Field”. Digital technologies are being actively developed in this area. The system considered in this paper refers to Advanced Driver Assistance Systems (ADAS). Existing technologies for detecting and recognizing objects in images can be divided into methods based on classical machine learning and methods based on deep learning. At the same time, the choice of an approach for specific application conditions is an independent scientific task that needs to be solved, especially in the case of creating new systems and considering new objects.