Vision-based multi-model facial emotion recognition psychotherapy assistant system

Emotion is a state of integrating people’s thoughts and behaviors. It plays an important role in people’s communication. It includes people’s psychological response to external or self-stimulation, including the physiological response accompanying this psychological response. Emotion detection is one of the most important methods in psychotherapy. The patient’s psychology can be inferred from the change of emotion. However, the traditional mood detection is limited by the use conditions of some relevant devices. In addition, patients may have resistance when wearing equipment, which may aggravate the disease. As a solution, the author proposes a visual emotion recognition system. The system uses gray image as data set for training. Four models are used for training (Alex-Net, VGG-Net, Google-Net and ResNet). ResNet has the best performance and training accuracy is close to 90%. Alex-Net and Google-Net have the worst performance, with recognition accuracy of only 40% to 60%. Finally, in the emotion recognition system, ResNet was determined as the best model. At the end of this paper, optimization schemes were proposed for further study.