Research On Emotion Recognition Of Pulse Signal For Improved Support Vector Machine

With the development of society,the theory and technology of human-computer interaction has become more sophisticated and the range of applications has already penetrated into all areas of human life.As emotion recognition is an important branch of human-computer interaction,computer can recognize human emotions in real time,and may make an appropriate adjustment,which plays an important role in creating a harmonious human-computer interaction environment.The pulse signal is a physiological signal,which itself contains a wealth of physiological and pathological information.Therefore,emotion recognition based on the pulse signal has important significance.The classifier in emotion recognition affects the accuracy of recognition directly,and the key parameters of support vector machines such as the penalty factor C and kernel function parameters take different values,the performance of the classifier has been greatly affected.This paper uses swarm intelligence optimization algorithm to optimize the parameters of the support vector machine automatically.The firefly algorithm is applied in support vector machine to optimize the key parameters,and the average recognition rate of such state is 7.9% higher than the support vector machine without any treatment.The research of emotion recognition based on pulse signal is mainly from the following parts:(1)This paper designed a photoelectric pulse signal acquisition board and developed a reasonable acquisition scheme.Collected the pulse signal of healthy undergraduates for a total of 80 sample as the experimental database,and laid a foundation for further research work.(2)Using empirical mode decomposition algorithm to the pulse signal denoising and aiming at the endpoint Effect problem of EMD algorithm which affects the decomposition result,this paper uses the method of maximum correlation extension to improve it.Simulation results show that the improved EMD can restrain the end effect to some extent and it has a better denoising effect.Three types of characteristics which are closely related to emotion,are extracted from the pulse signal,namely the statistical value,pulse area feature quantity K and the empirical mode energy entropy feature.(3)Describes the influence of the support vector machine and its key parameters(Cand ?)on the performance of the classifier,analyzes the shortcomings of the traditional method of parameter optimization.This paper selects particle swarm optimization algorithm and the firefly algorithm from swarm intelligence optimization algorithm,and the experimental results show that firefly algorithm converges fast,high robustness and accuracy better than particle swarm optimization.Therefore,this paper uses firefly algorithm to optimize support vector machine key parameters C and ?.The classification results show that,the average recognition rate of support vector machine for samples which optimized by the firefly algorithm is higher than without any treatment of the support vector confidential up 7.9%.Thus,this method can improve the recognition rate of emotion to a certain extent,so that the computer can recognize human emotion more accurately.