论文部分内容阅读
近年来,脑力负荷估计已经经历了广泛的研究,因为监测认知负荷的能力能够防止认知超负荷并且改善工作场所安全。脑电图(EEG)信号已经被发现是一种客观和非侵入性的脑力负荷的测量方式。然而,作为实时脑力负荷监测和脑机接口研究的重要一步,基于单试验EEG数据的认知负荷的评估一直是一个重大的挑战。最近,许多高级的特征提取方法和机器学习算法已经被采用于基于EEG的脑力负荷评估中。在本研究中,使用在具有2个难度水平的n-back任务的执行期间记录的EEG数据进行了单试验脑力负荷分类,测试了3种类型的特征提取的有效性(谱功率、离散小波变换和公共空间滤波),并评估了4种分类算法的性能(支持向量机、K-近邻、随机森林和梯度推进分类器)。研究结果表明,公共空间滤波是性能最好的基于单试验的脑力负荷分类的特征提取方法,而且最佳性能可以通过将来自谱功率或离散小波变换的特征与来自公共空间滤波的特征相结合,并采用随机森林分类器来实现。这项研究可能对基于单试验脑电图数据的脑力负荷评估中的特征提取方法以及机器学习算法的选择提供一些有用的指导。
In recent years, brain load estimation has undergone extensive research because the ability to monitor cognitive load can prevent cognitive overload and improve workplace safety. Electroencephalography (EEG) signals have been found to be an objective and non-invasive measure of mental stress. However, assessment of cognitive load based on single-trial EEG data has been a significant challenge as a significant step in real-time brain load monitoring and brain-computer interface studies. Recently, many advanced feature extraction methods and machine learning algorithms have been employed in EEG-based brain load assessment. In this study, single-trial mental workload classification was performed using EEG data recorded during the execution of an n-back task with two difficulty levels, and the validity of the three types of feature extraction (spectral power, discrete wavelet transform And public space filtering), and evaluated the performance of four classification algorithms (SVM, K-nearest neighbor, stochastic forest and gradient propulsion classifier). The results show that public spatial filtering is the best feature extraction method based on single-trial mental load classification, and the best performance can be obtained by combining features from spectral power or discrete wavelet transform with features from common-space filtering, And using random forest classifier to achieve. This study may provide some useful guidance on the methods of feature extraction and the selection of machine learning algorithms in brain load assessment based on single trial EEG data.