Facial Expression Recognition Algorithm Research And Design Of An Intelligent Diagnosis Labeling System For Depression

As a mental illness,depression can have a serious impact on patients,their families and society.In the diagnosis of depression,doctors use questionnaires to ask patients questions,and analyze the contents of patients’ answers,behaviors,etc.for diagnosis.This paper conducts research on facial expression recognition that is important in the diagnosis of depression.In order to imitate the doctor’s diagnosis process,an intelligent diagnosis scheme is established,and an intelligent diagnosis labeling system for depression is further established.The main work of this article is as follows:(1)Establish an expression recognition model VBLCA based on covariance pooling and attention mechanism.To solve the problem that the CNN-RNN expression recognition algorithm cannot extract second-order features and longer time series information well,the covariance pooling is used to extract the second-order change features of the face,and the self-attention mechanism is used to realize the effective information of the time series Better capture,thereby effectively improving recognition accuracy.VBLCA achieved an accuracy rate of 51.05% in the AFEW verification set of the seven classifications,which was improved by 2.05% compared to the latest single-model algorithm.In addition,tears,as a manifestation of deep expression,are very important for the diagnosis of depression.In this paper,the tear data set was self-established through network search and multi-person annotation methods.VBLCA’s tear data set in the second classification reached 92.04% accuracy rate.(2)Establish an expression recognition model RNTA based on time transfer module and attention mechanism and conduct model fusion research Aiming at the problem of huge parameter and complicated calculation of 3DCNN expression recognition algorithm,the time transfer module is used to extract spatio-temporal features without incre asing the calculation amount of 2DCNN.At the same time,during the classification process,the attention mechanism is used to merge the results of multiple frames to improve recognition Accuracy.RNTA achieves an accuracy rate of 54.47% in the AFEW verification set,which is 5.47% higher than the latest single-model algorithm and 93.83% accuracy in the tear data set.In order to further improve the performance of the model,three model fusion schemes were designed,such as average score fusion,fully connected layer fusion,and attention-based feature layer fusion.On the AFEW verification set,the accuracy rates were 56.05%,50.79%,and 51.57%,and on the tear data set,the accuracy rates were 95.26%,93.05%,and 93.55%,respectively.The experimental results show that the average score fusion can further improve the accuracy of facial expression recognition.(3)Implement the intelligent diagnosis and labeling system for depression.This article integrates "Montgomery Depression Rating Scale","Hamilton Depression Rating Scale" and expert opinion to establish the diagnosis entity and attribute value of depression.The system can use the data collection module to collect information,use the scale answer module to obtain the patient’s current situation,use expression recognition and other algorithms and the labeling module to label the entity attribute values,and display the final diagnosis results.After testing,the system has reached the set target and has been put into use.In this paper,the improved algorithm proposed for facial expression recognition has a certain improvement in accuracy.The system proposed for the intelligent diagnosis of depression considers many factors,all of which have high application value.