Research On Facial Expression Recognition Method Based On Contrastive Learning

Facial expression is one of the important factors in human communication to help understand the intentions of others.At present,facial expression analysis has become an important direction in the field of computer vision and artificial intelligence.Facial expression recognition technology has broad applications in driver fatigue detection,service robots,classroom student listening quality evaluation,and advertising video design.The task of the facial expression recognition system is to output the expression category corresponding to the image for a given face image.At present,the accuracy of facial expression recognition in a controlled environment is relatively high,but facial expression recognition in a real environment is affected by various factors such as pose changes,occlusion,and lighting differences,resulting in low accuracy of facial expression recognition.In order to better reduce the influence of various factors in the real scene on the recognition effect and improve the accuracy of expression recognition,the specific research work of this paper is as follows:(1)Research on facial expression recognition algorithm based on contrastive learning.Using the method of comparative learning to realize facial expression recognition,in view of the influence of factors such as pose transformation,occlusion,and illumination difference in facial expression recognition under realistic conditions,on the basis of the Mo Co v2 framework,the original image and the enhanced image after occlusion are added.For comparison of positive samples,random data enhancement operations are performed on positive samples to increase the type of positive sample images and enhance the recognition effect of the model in real environments;the Vi T-small network based on Transformer is used as the backbone feature extraction network to improve the model’s ability to extract features;The recognition model is pre-trained on the Image Net dataset,and the pre-trained model is applied to the expression recognition dataset for fine-tuning to improve the classification accuracy of the expression recognition task.Combined with comparative learning pre-training,the recognition accuracies of 89.02%,90.84%,64.94%,and 60.63% were obtained on the public expression recognition data sets RAF-DB,FERPlus,Affect Net-7,and Affect Net-8 data sets,respectively.Compared with popular algorithms,the effectiveness of the algorithm in this paper is proved.(2)Development of facial expression recognition system under natural conditions.Aiming at the needs of facial expression recognition under natural conditions,this paper develops a facial expression recognition system,which mainly includes five steps: face acquisition,face detection,image preprocessing,feature extraction,and expression classification.Collect face images through cameras and other hardware devices,use the Retina Face face detection framework and use the lightweight Mobile Net V3-small network as the backbone network to detect face areas,in order to make the model focus on important features in both channel and spatial dimensions,and To be able to pay more attention to location information,this paper uses the lightweight and efficient attention mechanism CANet to replace SENet in Mobile Net V3-small,and then performs image preprocessing operations such as data enhancement on the detected face images,and uses contrastive learning trained Vi T-The small network extracts expression information,uses the extracted features for expression classification,and outputs classification results.Finally,experimental analysis is carried out to evaluate the recognition effect of the system.