Emotion Recognition From Physiological Signals Based On BPSO

Emotions play a significant role in human perception and decision making. For a long time, research on emotion intelligence has been done in the fields of psychology and cognitive science. Along with the development of artificial intelligence these years, the combination of emotion intelligence and computer technology brings the novel research area named affective computing. This combination will greatly advance the development of the computer technology. Emotion recognition is one of the key technologies of affective computing for it is the foundations of harmonious human-machine interaction. The target of emotion recognition is supply theory and experiment foundations for correct select emotion physiological signals, and with reliable original data to support the comprehension and expression of emotion. Its application is widely. At present, the methods of emotion recognition mostly based on facial expressions, gestures and analysis of vocal signals, as well as questionnaires which are frequently used in psychology, the results of these methods are often influenced by subjects. However, we can continuously gather information about the users' emotional changes while they are connected to biosensors, since they are directly controlled by the human autonomous nervous system and incretion system, not by subjects. Therefore, the data will be more subjective obtained through recording physiological signals. Whether there is uniquely map between physiological patterns and specific emotion types, is dispute for a long time. But Ekman and colleagues had done a series of experiments and concluded that there exist some relations between physiological signal and specific emotions. Picard and colleagues at MIT Media Laboratory had testified that it is feasible to recognize emotion from physiological signals.In emotion recognition, too many irrelevant or redundant features also affect the recognition speed and accuracy, so feature selection is necessary. In fact, the feature selection problem is a feature searching problem. It has been proved to be a NP-hard problem. Although there are many proposed searching algorithms, no one is far superior to the other. Discrete binary particle swarm optimization(BPSO) is one of the swarm intelligent global optimal algorithms. The algorithm is famous for its simple code, small population and parameters, easy to understand and realize. Therefore, this paper tries to propose with discrete binary particle swarm optimization to select useful features from a large of emotion physiological signal features, and expects it can improve the correct classification rate of emotion state.This paper has mainly finished three research jobs based on existent research results:(1) To remove the redundant features in emotion recognition of physiological signals, proposed that introduce the idea of computational intelligence to the feature selection of emotional physiological signals, try to improve the correct recognition rate of emotion. And then taking BPSO feature selection methods to select useful features from emotional physiological signals; studied with single signal to recognize single emotion and multiple emotions, moreover, studied multiple physiological signals to recognize single and multiple emotions based on single physiological signal.(2) Aimed at improving the particle swarm optimization algorithms converges and escape from the local optima during search, proposed improved BPSO algorithm to select useful physiological features of emotion, improved the adaptability of BPSO, and then studied single physiological signal recognize multiple emotions and the relations between mutation ratio of particles and the correct recognition rate.(3) Taking different feature selection methods and classifiers to recognize emotion for exploring the relationship between emotion and physiological signals.This paper did many simulative experiments, verified the feasibility and correctness of the above job, and obtained some corresponding results:(1) Overall 193 features were extracted from four physiological signals, these features were extracted from four different emotions corresponding physiological signals. The feature set was selected by BPSO, nearest neighbor method is applied to classify the emotion classes, the whole correct recognition rate of four emotions is up to 86%. Among four emotion types, the recognition results of Joy based on ECG and SC are better, correct recognition rate are up to 88% and 72% separately; and the recognition results of Anger based on EMG and RSP are better, correct recognition rate are up to 80% and 100% separately. With single physiological signal to recognize four emotions, the RSP signal recognition result is best during four physiological signals, is up to 69.86%.(2) From many simulative experiments' results we can conclude that four emotions recognition rate is better when the mutation of particles' dimension is two, raised from 66% up to 81.35%. If take k-nearest neighbor as classifier, the whole average correct recognition rate of four emotions is up to 82.2%.(3) From the recognition results of SFS,SFFS feature selection combined with KNN,LDF methods, we know that the recognition rate of Joy and Anger are better, and Pleasure is worse.When the feature selection from physiological signal was regarded as a combinatorial optimization problem, this paper adopted BPSO as feature selection method. Experimental results demonstrate that BPSO is an effective way to emotion physiological signals feature selection. This work will put foundation for the application of emotion recognition.