A Research Of Feature Selection On Emotion Recognition From Electrocardiography Based On Evolutionary Algorithm

As a result of emotion playing a significant role in human decision and perception making, along with the continuous exploration in the area of cognitive,so more and more people began to urgently engage in the research on emotion recognition.However,more research could be limited in facial recognition and voice recognition,those features sometimes could not reflect our true emotional states,and of course,we would not be prone to analyze our emotions from physiology.In order to solve such conflicts,we have adopted physiological signals for the study to analyze the emotional identification.Because we know that through the analysis of physiological signals,we could identify internal feelings and emotional changes,and the most reliable is that it could be more ideal to get real physiological data according with the realistic environment.Professor Picard and its partners,from MIT Media Lab Affective Computing Research Group,firstly have extracted abundant features from physiological signals to identify the emotional states,and then more and more scholars and countries dedicated their attention to this area.However,the existing method of emotion recognition has some difficulties in ineffective identification,non-unified feature subset and poor robustness.Thus,this paper gives an introduction of evolutionary algorithm for recognizing emotions to break through the former limitations,achieve a better result and obtain an effective feature subset.Evolutionary algorithm is a new search optimization technology,which is based on genetic and biological evolution,and describes a realistic problem by gene and chromosome.Following the Darwinian "natural selection,survival of the fittest" principle,the random initial solution would gradually approaching the optimal solution by the genetic operation of reproduction,crossover and mutation.In fact,evolutionary algorithm is a mixed subject combining biological sciences and engineering technology.In addition to expert systems,artificial neural networks,it has become the third research focus.Since ECG signal contains the characteristics of rich emotions,you can reflect more pronounced changes in the human emotional states.Therefore,this paper gives an introduction of evolutionary algorithm for emotion recognition.And to some extent,a number of limitations which ever existed in the traditional method of feature selection on ECG have been effectively solved,and subsequently we turn into the further studies of ECG signal on emotion recognition.The main research contents are as follows:1) ECG data aequisition:ECG data are collected through inspiring 391 subjects from Southwest University with the state of joy or sadness by MP150;2) Original ECG preprocessing:remove noise such as high-frequency power supply and interference factors such as baseline drift by wavelet function and filter;3) Extraction of emotional features:apply the abilities of time-frequency localization and signal detection of singular points on wavelet transform to remove band noise and baseline drift of the ECG signal by Continuous Wavelet Transform and reconstruct the high-frequency part of ECG signal,then extract 84 features through its own analysis of ECG characteristics,and finally select 108 groups of more effective ECG signals to form a database;4) Selection of feature Subset:lead evolutionary algorithm(evolution strategy and genetic algorithm)combined with KNN classifier into classifying two emotions(joy and sadness) with the correct classification rate as the evaluation criteria,and select the effective subset of respectively representing their own emotional states;5) Finally,through experimental simulation on emotional database from Augsburg University and accordingly comprehensive comparative analysis from the above experiments in recognizing two emotions(joy and sadness),it is verified that use evolutionary algorithm with KNN neighbors to emotion recognition based on physiological signals is effective,and it not only has better recognition rate,but also exits effective feature subset of recognizing emotion.