Research And Application Of Facial Expression Recognition Based On Deep Generative Adversarial Learning Technology

Human facial expressions is one of the most common ways to reflect human emotions,and understanding different categories of facial expressions is an important way to analyze human perception and emotional states.Facial Expression Analysis(FEA)has been extensively studied over the past few decades.Classical theories of facial expressions categorize facial expressions in various regions of the world and races into roughly 6 or 7 categories.However,in real life,the recognition of facial expressions will be interfered by various conditions,such as posture,illumination,occlusion,environment,and the complexity of human emotions,which brings great challenges to facial expression analysis.Using computers to recognize facial expressions and obtain human emotions has also become one of the hotspots in the field of computer vision.With the continuous update and iteration of hardware,deep learning technology has also developed by leaps and bounds.Deep learning technology has gradually been widely used in various fields and has made breakthrough progress,especially in the field of computer vision.As a core technology in the field of computer vision,facial expression recognition has already produced relatively rich research results,but there are still some shortcomings,such as low recognition accuracy of related algorithm models,redundant model frameworks,and high recall rates.Poor and other issues.In order to solve the above problems,this paper introduces the method of deep learning into the multi-label,single-label and video-based facial expression recognition technology,constructs a new framework,improves the algorithm and improves the accuracy of recognition,and enhances the facial expression recognition technology.Robustness of expression recognition technology.At the same time,this paper also discusses some applications of facial expression recognition technology in social management,and puts forward some suggestions and countermeasures.The main research contents of this paper are as follows:(1)A novel multi-label facial expression recognition and multi-feature joint ensemble learning framework(MF-JLE)is proposed.It balances global features with several different local key features,taking into account multiple expression labeling factors in many facial action units.Compared with other frameworks,this framework adopts the method of ensemble learning and introduces the binary cross-entropy loss function,and obtains a better ensemble learning effect.(2)A multi-model joint ensemble learning framework for facial expression recognition based on generative adversarial networks is proposed.The framework is augmented by Star Gan,followed by ensemble learning by a convolutional neural network and a Swin-transformer module.Under this framework,the data augmentation method based on the generative adversarial network can alleviate the defect of the Transformer-type model that does not perform well in small data sets,while the use of Swin-transformer can effectively improve the perception of the field of vision,while integrating with channel attention and The convolutional neural network of the spatial attention mechanism can enhance the overall recognition accuracy of the model.In general,the framework combines the advantages of traditional convolutional neural networks with the high performance of Transformers,thereby improving the learning effect of the network.(3)A facial expression recognition framework based on generative adversarial networks and video image sequences is proposed.The Transformer is used to extract the time-series information of different frames in the video and the generative adversarial network is used to generate the facial expression picture of the peak frame in this method,which alleviates the defect of insufficient data of the facial expression picture of the peak frame.At the same time,the Transformer network is used to extract the features of the peak frame pictures and the extracted image sequence pictures.The features of the extracted video sequence pictures are input into the Net VLAD module to convert the video sequence features into the picture features of multiple video frames.The time-series information and the facial expression features of the pictures are used for classification.In general,this work hopes to use only Transformer-type networks to solve the tasks of extracting time-series information and image features while obtaining better classification results.In summary,this paper proposes three algorithm frameworks,which respectively prove the effectiveness of the corresponding algorithm frameworks in the three fields of multi-label facial expression recognition,single-label facial expression recognition,and video-based facial expression recognition,and the optimization process of Algorithm is explored in depth.Finally,it is verified in some standard datasets,and the algorithm framework proposed in this paper has achieved excellent results.In addition,this paper also discusses and analyzes the application of facial expression recognition technology in education management,medical management,traffic management and national security management.