Автоматизированная система распознавания эмоций по лицу человека с использованием разделяемой по глубине сверточной нейронной сети (Automated Human Facial Emotion Recognition System Using Depthwise Separable Convolutional Neural Network)

Аннотация

Automatic human facial emotion recognition (AHFER) system plays the important role in many fields, especially in the field of human-computer collaboration, human-robot communication, and emotion aware smart agent systems. Human beings can recognize and interpret facial expressions, but achieving the same task with an automatic system is quite challenging. In this study, we implement a Depthwise Separable convolutional neural network (DS-CNN) architecture, which includes depth-wise separable convolution (depth-wise and point-wise convolution), max pooling, global average pooling, dropout, and dense layers. In addition, image preprocess has been implemented using dataset splitting, intensity normalization, image cropping, and grayscale conversion. The proposed AHFER system is capable of recognizing four human emotions consist of happy, sad, angry, and neutral from image data. The comprehensive experimental results show that the proposed method achieved a training accuracy 99% and a validation accuracy 93%, respectively. Also, we determined the confusion matrix with the precision, recall, and f1-score for each class from the test dataset. The proposed MM-CNN architecture has improved the performance compared to existing state-of-the-art methods. In the future, developing a web or mobile application for a real-time facial emotion recognition system could be improved.