Research And Application Of Expression Recognition Based On Deep Learning

Facial expressions are one of the most representative signals for human beings to express their emotional states and intentions.Facial expression recognition has attracted increasing attention in academia and industry,and has been widely used in robotics,intelligent security,medical monitoring,educational assessment,driving fatigue monitoring and other fields.Relying on the group’s video action recognition and intelligent analysis project,this paper applies deep learning technology to the study of facial expression recognition methods,designs a lightweight expression recognition model Dense-Mobile Net,and develops a prototype of an intelligent online learning system based on this,which can intervene and alert online learning students,and can assist teachers in analysing students’ emotions during class The system can provide a basis for teachers to adjust their strategies and optimise their teaching methods.The main research elements of this paper are as follows.A lightweight expression recognition model,Dense-Mobile Net,is proposed.first,a Dense DW-block for feature reuse is designed and embedded into Mobile Netv1 to obtain better accuracy and less computation;then width selection experiments and comparison experiments are conducted on the RAF-DB dataset to verify the proposed Dense-Mobile Net’s effectiveness and to select the best network parameters.The experimental results show that: 1.Among the three proposed models Dense-Mobile Net-1,Dense-Mobile Net-2 and Dense-Mobile Net-3,DenseMobile Net-2 has the best accuracy rate of 82.4%.2.Compared with Mobile Netv1,the recognition accuracy of our model improved by 2.5%,the number of parameters decreased by45.7%,and the computational effort decreased by 66.73%.Two problems exist in online learning: 1.teachers have difficulty in obtaining feedback on students’ emotional state during class;2.there is a lack of supervision and intervention for students’ learning.Based on the facial expression recognition model proposed in this paper,a prototype of an intelligent online learning system is designed and implemented.The system implements a camera to capture images of students’ faces during class,and uses face detection,face alignment and expression recognition models to process and analyse the input images,and save the recognition results as emotional states in a database.During student learning,the system will alert and intervene with students based on their emotional states,and will use data visualisation techniques to generate an emotional analysis report to visually display to teachers,providing them with feedback on their teaching.The proposed lightweight expression recognition model,Dense-Mobile Net,has better accuracy and less computation than the traditional Mobile Netv1 for online facial expression recognition applications.The proposed Dense DW-block serves as a standalone feature reuse module that can be used to design or optimise similar convolutional neural networks.The intelligent online learning system demonstrated in this paper can supervise students’ online learning,while providing teachers with a basis for adjusting teaching strategies and optimising teaching methods.The design and embodiment of this system also helps in the developing of expression recognition systems for similar scenarios.