Research On Facial Expression Recognition Based On CNN-ELM

As an advanced form of human-computer interaction technology,facial expression recognition has become a key criterion for measuring the level of intelligent human-computer interaction.It is one of the research hotspots in the fields of machine vision,artificial intelligence and emotional computing.It has been widely used in the fields of fatigue driving detection,distance education interactive system,and criminal detection.At present,most of the research on facial expression recognition still stays in the experimental stage.The traditional method of extracting artificially defined features and then classifying it has the disadvantages of complex feature extraction process,poor feature representation ability and weak expression classification ability.The convolutional neural network has the ability to extract features implicitly and combine low-level features to form high-level abstract features.The extracted features have high resolution,but can further improve the classification accuracy and generalization ability.This paper combines convolutional neural network and extreme learning machine to study facial expression recognition,which has important theoretical and practical value.The main innovations of this paper are:1)In the process of training,the convolutional neural network does not consider the lack of information between categories,and introduces the supervised local preserving projection loss into the convolutional neural network,improves the feature extraction ability of the convolutional neural network,and keeps the sample features in high dimension.At the same time as the local manifold structure of the space,the distance between the features of the heterogeneous samples is increased,and the distance between the features of the same sample is reduced;2)For the insufficiency of the convolutional neural network's generalization ability is relatively poor and easy to fall into the local extreme point,the combination of convolutional neural network and extreme learning machine is used to further improve the classification accuracy of facial expression recognition system.;3)For the disadvantage that the extreme learning machine can not classify the samples in the nonlinear manifold space well,the manifold regularization is introduced into the extreme learning machine to further improve the sample classification accuracy of the extreme learning machine in the nonlinear manifold structure.The experimental and comparative experimental researches on the proposed algorithm are carried out.The experimental results show that the proposed algorithm has stronger generalization ability and higher classification accuracy than traditional methods in feature extraction and expression classification.Satisfactory effect.