NEURAL NETWORK MODEL OF FACIAL EXPRESSION RECOGNITION TO DETERMINE HUMAN FATIGUE

Abstract

The article is devoted to the use of a convolutional neural network model to determine a person’s fatigue by facial expression. For this purpose, the architecture of the convolutional neural network ResNet50 is used, which is trained on images and finds signs indicating a person’s fatigue and vigor. The prepared dataset in-cludes 6000 images, of which 3000 images correspond to a tired person (in these images the person yawns or closes his eyes) and 3000 images correspond to a cheerful person. When creating the ResNet50 neural network model, the Python programming language, the Jupyter Notebook development platform and the Anaconda3 development environment were used. The neural network model was trained for 100 epochs, with each training example consisting of 32 elements. Thanks to the use of the Adam optimization algorithm, it was possible to train a neural network to correctly classify images into two classes: fatigue and vigor. During training, the neural network model achieved a classification accuracy level of 95%. As a result of calculating the values of classification quality metrics Recall and Precision on a test data sample using an error matrix, it was possible to obtain high results. For the “tired” class, the value of the Recall metric was 0.9426, and the Precision metric was 0.9587. For the “vigorous” class, the value of the Recall metric was 0.9576, and the Precision metric was 0.9417. These indicators indicate the adequacy of the constructed model and the possibility of its practical use. As a result, we can conclude that the study demonstrated the successful use of a neural network model for facial expression recognition to determine human fatigue with a high degree of accuracy.