Research On Facial Expression Recognition Based On Deep Learning Method

Facial expression recognition is a cross-disciplinary field that spans multiple disciplines,covering many disciplines such as machine vision,psychology,neurology,and computer science.Facial expression recognition has broad application prospects in medical,intelligent monitoring,education and other aspects.With the rise of deep learning,expression recognition is divided into traditional machine learning methods and deep learning methods.The traditional expression recognition method includes three steps: Face detection,feature extraction and expression recognition.The research of expression recognition in the field of traditional learning has been a mature technology.In recent years,the research direction has shifted from the traditional field to the deep learning field.The deep learning network automatically learns the features related to facial expressions,and combines feature extraction and expression recognition,which enables researchers to achieve expression recognition even without knowledge of facial features.However,the amount of expression database data is insufficient to meet the needs of parameter training.For this,the method of transfer learning is introduced.In this paper,the training strategy of “pre-training-fine tuning-secondary fine tuning” was adopted.In the experiment,the ImageNet database was used to perform network parameter pre-training to have an excellent ability to extract feature.According to the transfer learning method,the FER-2013 database and the expanded CK+ database were used for parameter fine-tuning and training.We also used multi-task cascade convolution network for face detection to preprocess the experimental data for reducing the influence of non-expression data on network recognition.This article introduces the principles of Alexnet,VGG,ResNet,and GoogLeNet and uses these networks for expression recognition experiments.At the same time,This paper introduces the residual network architecture and SVM principle,and analyzes the feasibility of SVM replacing Softmax.An improved residual network(ResNet)expression recognition algorithm was proposed in this paper.The network architecture used a linear support vector machine(SVM)for classification.The experiment overcame the shortcomings of insufficient data through transfer learning,which can effectively prevent overfitting and overcome the problem that shallow networks rely on manual features and deep networks are difficult to train.The algorithm used a small convolution kernel and a deep network structure to solve the problem of accuracy reduction with the increase of depth by the residual module.The recognition rates of 91.333% and 95.775% were obtained on the CK+ database and the GENKI-4K database,respectively.The experimental result is about 1% higher than that of Softmax.In this paper,a real-time expression recognition system based on residual network is designed.The system uses Haar,LBP and MTCNN for real-time face detection and send faces to the residual network for real-time expression recognition.We tested the system performance through volunteers.