AutoML Technologies for Animal Monitoring

Abstract

Decision making in animal husbandry is critical to improving animal health and welfare, increasing production efficiency and minimizing environmental impact. Deep learning algorithms can be trained to recognize signs of animal condition in photographs and patterns of animal behavior in videos. The results of recognizing animal condition and behavior can provide valuable information about animal health and
welfare. However, the use of deep learning algorithms has its limitations in the form of the need for large amounts of high-quality data and machine learning experts. In this paper, a system for complex synthesis of deep learning models for animal monitoring based on Automated Machine Learning approach has been presented. This system can be used by non-experts in machine learning and users not capable of programming to create animal monitoring programs based on deep learning algorithms. The versatility of the created system was proved by testing it on the tasks of recognizing fundamentally different species of animals - reindeer and brant.