Design And Implement An Emotional Soothing System Based On Facial Expression Recognition

Emotion affects our daily behavior,and many situations require us to have a peaceful emotional state.There are many ways to affect the mood,music is an effective method.Relevant studies have shown that people can relieve their emotions by listening to music.However,people's preference for music is influenced by many factors such as personality,emotion and environment.In order to ensure the objective effectiveness of using music to relieve emotions,it is necessary to identify the emotional fragments of music.Expression is the main way to express emotions,and the current emotional state can be obtained by recognizing people's facial expressions.Facial expression research is a hot field and has a wide application prospect in daily life and business environment.Most of the expressions in life scenes are collected by camera equipment belong to sequence images.Most of the relevant researches focus on peak expression discrimination,and these models have poor recognition ability for sequence images.Therefore,this paper designs and implements an emotional soothing system based on facial expression recognition,which can judge emotions by recognizing users' facial expressions and play corresponding music at the same time to relieve emotions.Main work of this paper is as follows:First,a sorting network model that can recognize inconspicuous expression is proposed and validated by relevant experiments.The model takes the differences between same expressions into account,and carries out the micro-expression sorting recognition for different levels of the same expressions,which achieves high accuracy in the recognition of sequential images.In order to solve the lack of facial expression data,a two-stage transfer learning strategy is adopted to improve the model accuracy.Secondly,a method to identify the emotional characteristics of music is proposed.Based on the requirements of emotional soothing system,adopt the features including zero crossing rate,spectral center of mass,chroma characteristic,Mel-frequency cepstral coefficients,waveform energy and spectrum bandwidth.Based on the theory of emotional state,a double SVM model was designed.Experiments are designed to verify the validity of these features,the influence of audio sampling time on the recognition accuracy is analyzed,and the feasibility of music emotion feature recognition is verified.Thirdly,the finished model was deployed on x86 computers and android systems.Based on the differences of computing power and memory size between computer and mobile devices,the system is adjusted accordingly.The computer deployment scheme adopts the design scheme of the front end of the browser and the back end of the server.The experiment proves that the deployment can achieve the accuracy and real-time performance required by the theory.The mobile terminal deployment adopts the built-in scheme of compressing the sorting network model,which realizes the emotional relief function of the single mobile device offline,and designs experiments to verify the feasibility of the scheme.