Research And Implementation Of Facial Expression Recognition System Based On Deep Learning

With the advent of the 5G era,facial expression recognition will have broad application prospects in many fields.However,at present,facial expression recognition still has difficult problems such as lack of dataset,single feature,and poor recognition effect in natural scenes.Aiming at the above difficulties,this article is committed to designing a deep learning-based facial expression recognition system with good recognition effect,strong anti-natural interference ability.The research content of this system mainly includes: face detection and face key point algorithm selection,dataset expansion,local geometric feature selection,fusion neural network design,system platform construction and testing.(1)According to the basic system framework,through the analysis and comparison of multiple algorithms,supplemented by experimental data,comprehensively considering the correct recognition rate,real-time performance and stability,this article selects the face detection method based on HOG features and Face key point detection method based on ERT algorithm.(2)Aiming at the difficult problem of lack of facial expression dataset,this article is proposed to use DCGAN fusion network to generate pictures,expand CK+ and Jaffe expression datasets,so that the number of expression category pictures in the data setis evenly distributed,and this expression datasets are used by the fusion neural network to improve the robustness of the network model.(3)this article is proposed to use the local geometric features of facial expressions to judge the expressions of other people's faces to improve recognition accuracy for facial expression classification.Through comparative experiments,the final features is 22 key points with high characterization ability selected by Light GBM method as the local geometric features,and confirms that the local geometric features of the face have strong expression discrimination ability.(4)The fusion neural network that combines image characteristics and local geometric features to jointly distinguish facial expression categories is designed to solve the problem that facial expression pictures in natural scenes are easily affected by light intensity,noise and other factors.According to the series of comparative experiments,the experimental results confirmed that the expression recognition accuracy of the fusion network is higher than that of a single feature network.Comparing the accuracy of some published documents,the average recognition accuracy of the fusion neural network on the expression datasets(CK+,GENKI-4K,Jaffe)reached 98.25%,93.9%,and96.80%,which confirmed that the proposed fusion network showed higher performance and better recognition effect.(5)This article integrates the design of various modules,embeds the fusion neural network,builds a set of facial expression recognition system platform based on deep learning,and conducts test experiments in actual scenes to verify that the system has a good facial expression recognition effect and has a certain anti-interference ability and high practical value in our real life.