The Feature Extraction And Emotion Analysis Based On Chaotic Theory

Physiological signal is an external manifestation of emotions, the emotional fluctuations can be expressed via the variation of physiological signals, this performance is not controlled by the individual’s subjective conscious, so the result is more objective and true. Beacause the complex variations of the physiological signal is difficult to describe by the statistical feature approach, the paper verifies the physiological signals are non-stationary and chaotic, on the basis of this, presenting the chaotic characteristics of physiological signals through analysis of physiological signals. The paper extracted chaotic characteristic parameters from physiological signals (ECG, EMG, RSP, SC) under four different emotions (joy, anger, sadness, pleasure), such as the largest Lyapunov exponent, correlation dimension, approximate entropy and complexity. On this basis, the paper uses the J48decision tree to train and recognize the chaotic characteristics of physiological signals, J48decision tree classifier is a supervised learning method, it has many advantages in solving the multi-class classification problem, such as the fast classification speed, the simple classification rules, the high accuracy etc. The chaotic characteristic matrix is made of the extracted Chaotic characteristic parameters, Combined J48decision tree classifier to recognize the four different emotional. The results show it is feasible emotion recognition of physiological signals based on chaotic theory. Multiple physiological signals to identify different emotions, the recognition of happy and angry rate are100%and96%respectively, the recognition of sadness and pleasure recognition rate are88%and92%respectively. The results of Recognition rate are compared with the Augsburg University Laboratory results, discovering the identification of sadness and pleasure is better, the sadness recognition rate of Augsburg University Laboratory is77.27%compared to the recognition rate of the paper is88%, and the pleasure recognition rate is68.18%compared to the paper is92%.