PSO optimization of synergetic neural classifier for multichannel emotion recognition

In the world of technology, human-machine interaction is becoming more common and will perhaps be a part of our life in the future. Human-machine interaction is more natural if machines are able to perceive and respond to human non-verbal communication such as emotions instead of relying only on audio-visual emotion channels. A particle swarm optimization (PSO) of synergetic neural classifier for multimodal emotion recognition is proposed in this paper. In the experiments, a music induction method which elicits natural emotional reactions from the subject is used and four-channel biosensors are used to obtain electromyogram (EMG), electrocardiogram (ECG), skin conductivity (SC) and respiration changes (RSP) of the subject. The most significant features are extracted via testing several feature selection/reduction methods. Four classes of emotions, that is, joy, anger, sadness, and pleasure are considered and the synergetic neural classifier is used for multimodal emotion recognition. Weights are assigned to the different channels of the classifier and PSO is applied to optimize the weights for enhancing performance. Fast classification speed has been achieved and the experimental results look promising.