Search

The discrete binary particle swarm optimization algorithm proposed in this work can address the difficulty of imprecise emotion-related typical features and localization of emotion recognition. Eighteen classes of features in the time domain, frequency domain and time-frequency domain and differential entropy features are used as features for holistic emotion recognition accuracy using linear and nonlinear methods. Then, the next particle position, velocity and fitness are updated according to the recognition rate accuracy, so as to obtain the optimal feature combination by the emotion recognition accuracy, and lastly the subjects were identified using SVM to get the emotion recognition outcome. The initial experiments reveal that the percentage of correctness was 88.95% and 76.12% for BPSO and PSO, respectively. and the highest correct rate of happy among the four categories of emotions is 91.78%, which is improved by up to 12.83% using the BPSO algorithm.