Emotion Pattern Recognition From Physiological Signals

Emotion recognition from physiological signals is a kind of emotion pattern recognition, and the features are extracted from the physiological signals sampled. This paper regarded the emotion recognition as the pattern recognition, and GA is combined with the k-nearest neighbors to recognize affective states from physiological signals. Regarding searching the optimal feature subset in the process of emotion recognition as the combinational optimization problems, and attempting to find out the optimal feature subset which represents exactly the relevant affective states. The originality that the modern intelligence optimization algorithms are applied to physiological signals for emotion recognition is presented.The main works of this dissertation are summarized as follows:(1) The affective database supplied by Augsburg University of Germany are analyzed andprocessed, and the original affective feature matrices are constructed to use in the future.(2) Taking into account the combinational optimization problem of emotion recognition fromthe physiological signals, and GA is combined with the k-nearest neighbors to recognize the following four affective states: joy, anger, pleasure, sadness. In order to verify the feasibility and effectivity of the method.(3) A new algorithm that adaptive GA and is combined with the k-nearest neighbours waspresented and was applied to recognize again the following four affective states: joy, anger, pleasure, sadness, and observe whether the recognition rates were improved and the results are consistent with the conclusions from the relevant literatures nowadays. According to the three aforementioned aspects, three conclusions after simulation are made as follows:(1) The affective feature matrices which are constructed from the affective database(Supplied by German's Augsburg University) are feasible and effective to recognize the following four affective states: joy, anger, pleasure, sadness. Because of the redundancy, this doesn't mean the more amount of features, the better recognition rates are. There exists a great difference in the process of emotion recognition between different physiological signals and different affective states.(2) The method that GA is used to search the optimal feature subset in the process of emotion recognition is feasible and effective. Furthermore, feature's quantities that the recognition rates are best in the process of emotion recognition between different physiological signals and four affective states were different, and the relevant features are different, so the feature combinations have a great differences. Great differences have taken place while recognizing the different affective states using one and the same physiological signal, which shows the physiological signals have a great difference in different affective states, such as the 'anger' recognition accuracy gained 97% using the physiological signal RSP. On the contrary, the affective state 'pleasure' gained only 34%; it showed that the 'anger' was easier recognized than the 'pleasure' using the physiological signal RSP. Except these, a great deal of experiment results showed that searched feature combinations which gained the best recognition accuracy included as following features: RSP1diff-max, RSP2diff-max. all of these indicated the RSP represented the anger easier than the others.(3) The recognition accuracy has improved using the improved algorithms, emotion recognition from four physiological signals made the better conclusion than from single physiological signal. What's more, the more the feature's quantities and diversification, the more the searched feature combinations represented exactly the relevant affective states. In terms of affective dimensionality, it was easier to distinguish emotions along the arousal axes than along the valence axes. Statistical methods were used to find out which features are significant for a specific emotion demonstrated: The feature SC1diff-mean is significant for the specific emotion anger. The anger is characterized by high RSP-levels. All of these conclusions are consistent with emotion psychology and the relevant literatures nowadays.At last, the research results are summarized, the future research directions which can bedeveloped further at the same time are pointed out and application foregrounds for physiological signals for emotion recognition are expected.