Research On Expression Recognition Technology Based On Deep Learning

Facial expressions play an important role in the communication between people.Traditional facial expression recognition technology relies more on manual feature extraction and is susceptible to external environment interference.With the emergence of machine learning,researchers use convolutional neural networks to overcome the manual dependence of facial features,automatically extract the feature information required for expression recognition,and better adapt to complex and changeable external environments.This paper studies the problems of poor recognition accuracy,large amount of deep learning parameters,and slow training in the natural environment where expression recognition exists.After learning and exploring facial expression recognition algorithms based on deep learning,two superior expression recognition algorithms are proposed.Based on the proposed algorithm,the application design is carried out,and a set of psychological detection system based on facial expression recognition is constructed.The main research content of this paper is as follows:(1)Aiming at the problems of low recognition accuracy of facial expression recognition in natural environment and large amount of deep learning model parameters.A channelweighted multi-way attention network is proposed.The network designs a multi-branch facial expression recognition feature extraction network,and designs and adds adaptive weights in the auxiliary network of feature extraction.Secondly,design and modify the spatial attention mechanism to accelerate the acquisition of important facial feature information areas.Finally,the experimental design and analysis of the algorithm proposed in this chapter is carried out.This algorithm achieved a superior recognition accuracy of 73.81%,achieving superior performance compared to the current recognition level.(2)Aiming at the problem of poor generalization and slow training of single network training in expression recognition network,a multi-task expression recognition algorithm is proposed.The algorithm uses key points and LBP features to assist in the accurate positioning of the face in the process of expression recognition and obtains facial texture features.Designing an adaptive loss function to jointly optimize the expression recognition agent network.Fusion of attention modules for accurate classification of expression recognition.(3)In order to verify the effectiveness and practicability of the expression recognition network,a psychological monitoring and analysis system based on expression recognition was developed based on NVIDIA hardware..Enables psychoanalysis of remote subjects by centering on a non-contact emotion detection algorithm.According to the algorithm in this paper,the system extracts an ultra-lightweight network model,and analyzes the collected images in real time,taking into account both accuracy and speed.The system collects scene images in real time through the camera,the hardware terminal is responsible for algorithm analysis,and writes the analysis results into the database.The background visualization platform reads the real-time reality from the database to help users analyze the current psychological state of the characters in the scene.