Emotion Database Establishment And Feature Extraction Algorithm

Researchers are focusing on the rationality of designing experiment for dataacquisition, and the validity of the collecting raw data is the foundation and key issue ofemotion recognition. It seems that establishing the high quality emotional databasebecomes a research hotspot. Electrocardiograph (ECG) emotional signals are veryimportant physiological signals. Numerous studies have demonstrated that there aresignificant correlations between ECG emotional signals and the human emotional state,and the ECG signal is widely used in clinical whose collection and analysis technology isrelatively mature. Therefore, the ECG emotion signals were collected, and the ECGemotion database is established in this paper by conducting emotion induced experiment.Studies have found that video material is more successfully in inducing humanemotion than pictures or music material. In this paper, video materials are clipped fromnumerous of films and televisions to induce emotion. ECG signals of happy, anger,sadness, and fear state from25students who are health and without heart disease historywere collected by emotion induced experiment. Intercepting out60seconds valid datafrom each ECG emotion signals as sample, using wavelet transform in pretreatment, suchas removing baseline drift and so on, and then establishing the ECG signal emotiondatabase.The result of emotion analysis and recognition is influenced directly by extraction ofemotion feature. Heart rate variability (HRV) reflects the disturbance of human bodyinternal and external environment on the cardiovascular system and the reaction ofcardiovascular system by autonomic neural system and humoral regulation to thedisturbance, and is sensitive to the changes of different physiological state and evenemotional state, contains a lot of information about emotions, therefore, can be used foremotion recognition. Wavelet transforms and independent component analysis (ICA) arecombined for heart rate variability of time-frequency analysis in this paper, and a total of21features of heart rate variability are extracted. Also combined with79time-domainfeatures and36wavelet features of ECG signals, and21normal time-frequency features of heart rate variation, ultimately, a total of157emotion features are extracted. Emotionalfeature sets extracted in this paper are used in ECG emotion recognition analysis. Forfurther research, feature selection and optimized combination conducted by Relief-Falgorithm, maximum relevance and minimum redundancy, principal component analysisand independent component analysis combined algorithm, and the optimized supportvector machine based on the genetic algorithm is adopted respectively for single emotionrecognition whose average recognition accuracy reached above90%. The resultsdemonstrated the ECG emotion database established in this paper has a good performanceand the emotion features extracted in this paper are with strong emotion recognition ability.Compared with regular HRV features, ECG time-domain features, and ECG waveletfeatures, the heart rate variability features extracted by the algorithm also has a remarkableability of emotion recognition.