Machine learning for recognition of events in hostile environments

Ekaterina Panteley, Viacheslav Abrosimov
The paper considers the problem of machine learning for recognizing events that can occur with a control object operating in its environment. Information about the state of the environment is formed by means of technical vision systems as a result of its targeted online monitoring. Such events, triggered by a set of recorded facts and influencing environmental factors, become a training sample. The sample is used for machine learning with deep neural networks. The neural model obtained during the learning process is used in the object control system to recognize and predict potential events and their consequences.