基于皮肤电信号与文本信息的双模态情感识别系统

人机交互离不开情感识别，目前无论是单模态的情感识别还是多生理参数融合的情感识别都存在识别率低，鲁棒性差的问题.为了克服上述问题，故提出一种基于两种不同类型信号的融合情感识别系统，即生理参数皮肤电信号和文本信息融合的双模态情感识别系统.首先通过采集与分析相应情感皮肤电信号特征参数和文本信息的情感关键词特征参数并对其进行优化，分别设计人工神经网络算法和高斯混合模型算法作为单个模态的情感分类器，最后利用改进的高斯混合模型对判决层进行加权融合.实验结果表明，该种融合系统比单模态和多生理参数融合的多模态情感识别精度都要高.所以，依据皮肤电信号和文本信息这两种不同类型的情感特征可以构建出识别率高，鲁棒性好的情感识别系统.

Bimodal Emotion Recognition System Based on Skin Electrical Signals and Text Information

Human-computer interaction is inseparable from emotion recognition. At present, there is a problem of low recognition rate and poor robustness both in single-modality emotion recognition and multi-physiological parameters fusion emotion recognition. Therefore, a fusion emotion recognition system based on two different types of signals is proposed, that is, a dual-modality emotion recognition system that integrates physiological parameters of skin electrical signals and text information. Firstly, by collecting and analyzing the characteristic parameters of the corresponding emotional skin electrical signals and the emotional keyword features of the textual information, the artificial neural network algorithm and the Gaussian mixture model algorithm are designed as a single mode emotion classifier, respectively. The Gaussian mixture model weights the decision layers. Experimental results show that this kind of fusion system has higher accuracy than multi-modality emotion recognition combined with single mode and multiple physiological parameters. Therefore, based on the two different types of emotional characteristics of skin electrical signals and text information, an emotion recognition system with high recognition rate and sound robustness can be constructed.