Research on Automatic Identification and Real-Time Monitoring Technology of Mental Health Status Based on Multilevel Neural Network

Abstract

In this paper, the densely connected DenseNet is used as a feature extraction network to solve the problem of overfitting due to the increase of network parameters and small data samples, and the classifier uses Softmax for facial expression classification and localization. Aiming at the problem of lack of maximum attenuation in automatic identification, Soft-NMS algorithm is used instead of the traditional NMS algorithm to improve the detection accuracy. Design of remote mental health monitoring system based on B/S structure, including the analysis of the hardware structure of the Web server, sound signal module, power supply module, and other analysis, as well as personality traits, personalized testing, data transmission and so on. Software development psychology teachers, design evaluation results analysis, permission management and other modules. Through the above series of measures, the user's mental health can be effectively, accurately and safely remotely monitored and evaluated. The results show that the Soft-NM algorithm in this paper has a recognition rate of 90.14% in recognizing mental health conditions. When monitoring from 10 to 100 times, the accuracy of the system remains above 94%, up to 97% with little fluctuation. Moreover, the system in this paper takes less than 5.8s to run, which verifies the effectiveness and stability of multilevel neural networks in practical applications.