Research on Multimodal Emotion Recognition Based on Double Deep Networks and FR-ABCR

This paper proposes a new multi-mode emotion recognition method to improve the emotion recognition rate. The outstanding characteristic of this study are：1) Making full use of deep learning model can extract deep abstract features of emotional signals. a multimodal emotion feature extraction method based on double deep networks is proposed, which can not only effectively represent the characteristics of emotion, but also reduce the amount of calculation; 2) Use the deep residual network as the basic convolutional neural network model, the degradation and gradient disappearance of deep convolutional neural networks due to increasing depth are solved; 3) According to the ensemble learning ideas for common decision of multiple classifiers, a multimodel emotion information fusion and recognition algorithm based on AdaBoost and Class Reliability (FR-ABCR) is proposed. Both the decision level and the feature level are integrated, so that the whole emotion category recognition process is closer to the process of human brain judgment of emotion. Simulation experiments with a large number of data also prove the reliability and effectiveness of the proposed emotion recognition method.