Research On EEG Detection Algorithm Based On Event-related Potential

Polygraph technology's development is as remote as that of human society, and it has developed a lot from the outdated polygraph method to the scientific method based on scientific theory, the reliability and stability of which has been improved significantly. Polygraph technology has been used widely in commercial field and police field abroad which plays an important role in cases investigation and interrogation. With the application advantages of polygraph technology the Supreme People's Procuratorate and the Ministry of Public Security has confirmed that the polygraphy technology is a kind of effective investigative technology in 2000. The recognition of judicial organs of the state promotes the development and application of polygraph technology in our country. At present, the investigation departments of different provinces, municipalities and autonomous regions are equipped with multi-channel physiological lie detector. Considering the defects of present recognition rate and the lie detection method, we urgently need to seek a new lie detection method.Since the ERP has the advantages of high space-time resolution, using convenience and low cost, lie detection research based on ERP has got the attention of both domestical and international scholars. In recent years, although the lie detection research based on ERP has made many achievement, however, more research is based on personal information and crime knowledge tested by P300 polygraph technology. They extract wave amplitude, wave area and peak-to-peak value as characteristic parameter through three channels or only single channel. However, geometric properties can't reflect the characteristics of brain electrical signal effectively. What's more, the current recognition accuracy of the classifier commonly used can still not reach to the standard of the practical application. Therefore, this thesis uses ERP technology to do lie detection research on the basis of existing theory, collects the P300 brain signal of 15 subjects and 24 subjects of N400 brain signal which are used for feature extraction and classification recognition research. In this thesis, the main work is as follows:Firstly, in terms of feature extraction about brain electrical signal, aiming at the little study based on the feature extraction of P300 and low classification accuracy, a method of P300 feature extraction is proposed based on extreme learning machine auto-encoder. This thesis extract 15 subjects' features, then calculating the characteristic parameters of the SVM classifier classification recognition rate. The classification accuracy is up to 89.5%, which confirm the validity of the feature extraction and provide a feature extraction method of P300. Aiming at the shortcoming of the existing technology and according to the non-linear and non-stationary characteristics of brain electrical signal, this thesis adopt the sample entropy measure N400 to induce brain electrical signal complexity. This thesis find the threshold between lying and not lying by analysising data which overcome the existing polygraph judgment. Experiment results show the method is effective.Secondly, in terms of classification about brain electrical signal, This thesis integrates the artificial immune algorithm and the extreme learning machine, which proposes a method of evoked potential polygraphy based on AIA-ELM and N400.24 subjects are divided into a crime group and a control group respectively to extract the multi-channel N400 peak value, average amplitude and median frequency, and all of them just constitute the eigenvectors. AIA-ELM algorithm is applied to classify the probe stimulus and the irrelevant stimulus, and the recognition rate of the crime group is 97.60 percentage. Experimental result indicates that this method can distinguish lies effectively and provide a new reference for the polygraph based on N400.Above all, ELM-AE and sample entropy can be used for the extraction and analysis of brain electricity signal, which can also distinguish whether a person tells a lie or not effectively. The AIA-ELM algorithm proposed can improve the classified recognition precision of brain electrical signal effectively. These three methods are mainly used to do classified researches on brain electricity signal of from two aspects: feature extraction and classification algorithm, and good results have been achieved. Therefore, the study of feature extraction and classification algorithm can effectively promote the development of ERP lie detection technology.