Industrial Automatic Control Systems and Controllers Annotation << Back Automation of Behavioral Audit Based on Recognition of Human Emotions Using a Neural Network D.A. Loktev, O.S. Lokteva In this paper, using the example of railway operators-train drivers, one of the options for automating a behavioral safety audit by analyzing human images by recognizing his emotional characteristics directly in the process of performing a work task is considered. Automation of behavioral audit will make it possible to more objectively assess the actions of employees and their emotional state in the labor process, as well as assess the reaction to its possible negative changes in real time. The main emotional reactions, such as happiness, anger, surprise, as well as the definition of a person’s fatigue, can be determined by individual segments of the image of a person - the state of the eyes, mouth, eyebrows, head position. For image segmentation, it is proposed to use a convolutional neural network with the U-net architecture, as well as to specify the obtained segments by updating their centers and boundaries. Keywords: behavioral audit, convolutional neural networks, emotion recognition, image analysis. DOI: 10.25791/asu.10.2022.1390 Pp. 24-30.