Research Of Emotion Recognition On Pulse Signal For Feature Classification

With the rapid development of society,science and technology is changing with each passing day,which brings convenience to people,but also increases the pressure of modern life.People’s emotional problems have become a killer to human health.In this paper,the emotion is analyzed by the way of human-computer interaction.After the pulse signal is collected,he computer is used to analyze and process the signal features,so as to obtain the human emotion.The emotional feelings of the entity.The emotional state is the most real,can avoid the appearance of people’s emotional disclosure of camouflage.The research show that the portable medical devices can be used to collect human pulse signals in real-time.Through the analysis of the program in time to obtain people’s emotional state,to remind the parties to pay attention to emotional changes.If necessary,notify the doctor and guardian to help protect.Pulse signal is a kind of traditional bioelectrical signal.Since ancient times,it has been used to analyze the physiological state of Qi and blood,viscera and so on.Compared with other physiological signals,the pulse signal is less noisy and easy to collect.A large number of studies have shown that it is feasible to use the pulse signal to identify the emotion.Therefore,we can find out some characteristics of the emotion from the pulse signal to establish the mapping relationship between the pulse signal and the emotion.Therefore,it is of great significance for us to study the emotion recognition of pulse signals for the human psychology and the development of artificial intelligence in the future.This experiment was used to extract the pulse signals in the four emotion states: anger,excitement,sadness and calm from MIT Emotion recognition database,which was used to carry on the research.The study is divided into the following sections:1)The pretreatment of pulse signal: the first step is to denoise the pulse wave in order to obtain the characteristics of the pulse wave accurately.The baseline drift noise is removed by wavelet decomposition and reconstruction while keeping the pulse signal effective information.2)Feature extraction of pulse signal: the selection of different features affects the accuracy of the final classification,so the selection of features is very important.It is difficult to guarantee the final classification accuracy using the limitation of single feature.Therefore,this study uses a combination of features,including: the statistical characteristics of the pulse wave in the time domain.The feature points are detectedby the differential threshold method,and the characteristic points of the pulse wave peak value and the double notch of the depression are obtained.The wavelet coefficients of the high frequency and low frequency of each layer can be obtained by the five wavelet decomposition of the ’DB7’ wavelet.3)Emotion recognition feature classification : constructing a more powerful emotion recognition algorithm based on hierarchical support vector machine model to solve the problem that more support vector machines are needed in classification,and classification training sample set is more,take up memory,calculation time is long.The algorithm combines two-dimensional emotion model to build a hierarchical structure,classification of the emotional information carrying pulse signal feature extracted according to the dimensions of emotion itself,to achieving the purpose of identification of different emotions.For the N class classification problem,one to one SVM classification requires n(n-1)/2 classifiers,one to many SVM classification requires n classifiers,and training samples are too much.The hierarchical SVM classification only need to construct n-1 SVM classifier,and the training sample decline.The experimental results show that the support vector machine model for the four classification problem needs only 3 SVM,which is less than 4 support vector machines for one to many SVM classification requiring and less than 6 support vector machines for one to one SVM classification requiring by,while reducing the number of samples in the training process.This model can ensure the classification accuracy,at the same time,reduce the number of traditional classification algorithm of support vector machine,reduce the storage time of training samples and classification algorithm,improve the classification speed greatly.It is able to meet the need of accomplishing emotion recognition.