Emotional problems such as anxiety, stress and tension may have a long-term impact on athletes’ competitive performance and physical and mental health. The identification method of athletes’ psychological state based on video analysis has the advantages of objectivity, real-time and high efficiency. It provides a more accurate evaluation tool for coaches and psychologists. This article aims to design an efficient facial emotion identification model for athletes to improve the accuracy and real-time performance of emotion identification. The results show that the accuracy and recall rate of the convolutional neural network (CNN) algorithm are higher than those of the traditional algorithm in most cases, and the CNN model has high real-time and fast response ability. This research result is valuable for athletes’ psychological state monitoring and performance analysis. By studying athletes’ psychological state, coaches can better understand their inner world in the competition and provide them with more accurate psychological support and intervention. This will not only help to improve athletes’ competitive performance, but also help them better cope with the challenges and pressures in the competition and protect their physical and mental health.