In order to diagnose depression and anxiety, clinicians will conduct interviews with subjects. If large-scale screening is carried out, this method is too costly and difficult to implement. Because facial expressions play an important role in the diagnosis of clinicians, this provides an opportunity to solve this problem. Therefore, we recorded 303 subjects who answered the self-rated anxiety scale (SAS) and the self-rated depression scale (SDS) Video. Based on Convolutional Neural Networks (CNN) and Long Short-Term Memory Networks (LSTM), by using either of these two types of videos alone as a binary classification experiment, the accuracy of the diagnosis of depression is 72.53%, and the diagnosis of anxiety is 72.08%. In addition, by fusing the two types of videos to diagnose anxiety, depression, and normal in three categories, the accuracy of the model is 80.22%. Through the comparison of the results, the …