Group Emotion Recognition Based on Multilayer Hybrid Network

Aiming at improve the accuracy of the group emotion recognition, a multilayer hybrid network based on multiple deep Learning models is proposed to get different features from different aspects. In addition, considering the contributions of different features to the accuracy of group emotion recognition, a parallel feature fusion algorithm based on weight scaling inference rule is proposed to get the deep feature vector. Finally, based on the deep features obtained by the fusion algorithm, a random forest algorithm for classifying the group emotion is proposed to get high classification accuracy. The experimental results show that the proposed method achieves 77.76% accuracy rate on the test set, 24.14% higher than the benchmark method and 13.08% and 14.33% higher than the other two deep learning methods respectively.