基于自适应主成分分析维度寻优的脑力负荷识别

Mental Workload Recognition Based on Adaptive Principal Component Analysis Dimensional Optimization

脑力负荷识别对提高作业操作人员工作效率,减少人因事故具有重要意义。然而,由于脑电(electroencephalogram,EEG)信号的采集是由多通道脑电帽采集的,并且分布在各个频带上,因此经过特征提取得到的特征维度过高,造成后续识别模型复杂度过高。对此,通常使用主成分分析(principal component analysis,PCA)对高维特征向量进行降维处理,但是降维维度的取值很难确定。提出了一种基于主成分分析的自适应维度寻优方法,将实验数据集中的高维特征使用主成分分析降维到寻优维数范围内的各个维度,并将各个维度的分类精度绘制成维度—分类精度曲线,通过识别该曲线的“肘部”来确定该实验数据集的最优降维维度,并将该维度应用到同实验的其他实验数据上进行脑力负荷识别。结果表明,该方法可以准确识别出在同实验数据集中通用的最优降维维度,有效提高识别效率。

Recognition of mental workload is considered to be of great significance to improve the work efficiency of operators and reduce human accidents.However,because the electroencephalogram(EEG)signal is collected by a multi-channel EEG cap and distributed in various frequency bands,the high complexity of the subsequent recognition model is attributed to the high dimension features obtained by feature extraction.In this regard,principal component analysis(PCA)is usually used to perform dimensionality reduction processing on high-dimensional feature vectors,but it is difficult to determine the value of the dimensionality reduction dimension.An adaptive dimensional optimization method based on principal component analysis was proposed.The high-dimensional features in the experimental data set were reduced to each dimension within the optimized dimension range using principal component analysis,and the classification accuracy of each dimension was plotted as a dimension-classification accuracy curve.The optimal dimensionality reduction dimension of the experimental dataset was determined by identifying the“elbow”of the curve,and this dimension was applied to the same experimental data set and then the mental workload recognition was performed on it.The results show that the method can accurately identify the optimal dimensionality reduction dimension commonly used in the same experimental data set,and effectively improve the recognition efficiency.