Feature Learning Approach Based On Multi-modal Human Body Data For Emotion Recognition

The human body as the entire structure of a person contains physiological and physical reactions that are connected with emotions.Emotions play very vital role in our day-to-day living activities,not only in the way we interact with fellow human beings but also in our decision-making processes.Emotions also affects the way we view the world and everything that happens therein.Emotion recognition from multi-modal signals therefore allows a direct assessment of the “inner-most” state of a subject,which is regarded an important component of HumanComputer Interactions(HCI).In this dissertation,we conducted an experiment on the Database for Emotion Analysis using Physiological signals(DEAP dataset)by combining all the physiological signals based on the Circumplex model of emotion recognition also known as the valence arousal scale evaluation.These signals include electroencephalography,hEOG,vEOG,zEMG,respiration belt,galvanic skin response,Plethysmograph,and temperature.We defined the emotions in three non-similar categories: two and three classes using 1-9 distinct self-rating table and another model using three(3)non-similar emotional coded key-words to launch the three defined areas in the valence-arousal 2D model.In order for us to arrive at a more robust and a better accuracy that is fairly comparable with other works,we divided the DEAP database into training,testing and validation data sets..We finally finished the classification task by using the broad learning system,a powerful neural network and learning algorithm as a classifier.In applying the broad learning system to train,validate and test our model,we obtained 75.68 and 71.43 for valence and arousal respectively which proved that our work has more promising results as compared to others.