基于P300和机器学习的测谎方法研究

Lie Detection Method Based on P300 and Machine Learning

为了克服传统测谎方法没有考虑到相同刺激下受试者思维状态变化的缺点,提出了基于P300和机器学习的测谎方法.该方法使用典型的3刺激测谎范式,首先记录30名随机划分的撒谎者和无辜者的12导脑电(EEG)信号,使用独立成分分析方法(ICA)分解由探针刺激产生的脑电信号,利用在Pz电极上分布强度大的独立分量重建Pz波形,将每名受试者的的若干个Pz波形进行平均,提取两步降噪后的每个Pz波形的时域和小波特征.最后,使用分类器区分P300和非P300波形,进一步计算出个体测谎诊断率.实验结果表明,支持向量机(SVM)适合于说谎意识状态的分类,提出的方法可以有效地改进单次刺激记录上的信噪比,提高P300成分的识别率,进而提高测谎诊断率.

For solving the unvariability of subject cognitive states under the same kind of stimulus in the conventional machine learning methods,the method based on P300and machine learning was proposed.The standard three-stimuli protocol was chosen.Thirty guilty and innocent subjects were randomly divided into two groups and their EEG signals were first recoded.Independent component analysis（ICA）was carried out to decompose the datasets in the probe stimuli.The ICs with the largest projection strength at Pz were selected to reconstruct the Pz waveforms.Then small number of Pz waveforms within each subject is further averaged.Afterwards,the time-domain and wavelet features were extracted from each denoised Pz waveforms.In terms of the classifier to identify the P300and non-P300waveforms,the individual diagnostic rate was evaluated.The experimental results show that the SVM classifier is suitable to identify the sense states of lying,and the proposed method enables to improve the SNR in single trails,enhancing the accuracy of identifying the P300and of individual diagnostic rate.