基于优化的LSTSVM的多模态生理信号情感识别

Emotion recognition of multimodal physiological signals based on optimized LSTSVM

采用最小二乘双支持向量机(LSTSVM)进行情感识别,针对LSTSVM模型的惩罚系数及核函数参数难以确定的问题,使用改进的萤火虫算法(MFA)来优选LSTSVM的各项参数,使分类器取得最优的性能。基于脑电、皮肤电、肌电和呼吸4种模态的生理信号,使用该算法进行情感识别,并与使用标准LSTSVM和粒子群LSTSVM算法的识别结果比较。仿真分析表明,提出的MFA-LSTSVM算法识别准确率更高,需要的训练时间更短。

The least squares twin support vector machine（ LSTSVM） is used for emotion recognition. The penalty coefficients, and kernel function parameter of LSTSVM model are difficult to determine, so the modified firefly algorithm（ MFA） is used to select the best parameters of the LSTSVM to achieve optimal performance. Based on four modal of physiological signals, which are EEG, skin electrical, electromyography and respiratory signal, the proposed algorithm is used for emotion recognition, and comparisons are made with standard LSTSVM and particle swarm optimization LSTSVM algorithm. Simulation results show that the proposed MFA-LSTSVM algorithm has higher accuracy and shorter training time.