Research Of P300Deception Detection Based On Multi-Domain Integration And Genetic Algorithm

The deception detection research is a hot spot in the study of psychology application. In recent decades, with the development of event relate potential (ERP) technique, the ERP-based deception detection method turned into one of the main development directions of modern polygraph technique. In the laboratory ERP-based deception detection study, the P300potential, whose effectiveness in lie deception has been fully verified, is a kind of most commonly used ERP component. Research and application of P300-based deception detection will benefit the criminal investigations, counterterrorism, security fields, etc., and provide new approach for crime inhibition and safe protection to the countries and regions.Typically, P300-based deception detection can be divided into three parts: paradigm and stimulus presentation, EEG (electroencephalography) signal acquisition, signal processing and analysis. The EEG signal processing, which is the vital part of P300-based deception detection technique, is generally composed of signal preprocessing, feature extraction and classification. The capability of EEG signal processing method would directly determine the performance of P300-based deception detection method. Therefore, the research of EEG signal processing is of great importance.Commonly, the EEG signal has low signal to noise ratio (SNR) with strong and various noises. In addition, traditional signal processing method in P300-based deception detection research has multiple week points, such as single feature extraction, which cannot represent the essential properties of the EEG signal in the round, low separability of the extracted features. With the aim to solve these shortcomings, the signal processing methods was studied in this thesis.First of all, a new EEG signal processing method based on multi-domain integration and genetic algorithm was proposed by combining the theory of multi-domain integration and genetic algorithm. The proposed method extended the signal preprocessing and feature extraction from the temporal-spectral domain to the spatial domain. Then the multi-domain feature set was filtered using genetic algorithm to exclude the redundant or harmful features and obtain an optimal feature subset. According to the offline verification using left-right hand motor imagery movement EEG signal, the proposed method outperformed the traditional feature extraction method with an increase of1.9and3.6percentage point in average and highest classification accuracy rate respectively. What’s more, the amount of features reduced by80%. When compared with our previous work, although the average and highest classification accuracy rate of the proposed method decreased by0.6and1.2percentage point respectively, the number of features reduced by70%, which is conducive to real-time online application. Moreover, in comparison with band-power feature extraction methods, which extracted features only from spectral domain, the proposed method has a larger increase in the overall performance.The above results demonstrate that the proposed EEG signal processing method is of valuable and effective.Secondly, according to the EEG signal processing method based on multi-domain integration and genetic algorithm, a new P300-based deception detection method was proposed. A P300-based deception detection experiment using three-stimulus GKT technique with facial image was implemented. Twelve subjects participated in the experiment, and raw EEG signal was acquired from them. Then signal processing method based on multi-domain integration and genetic algorithm was used for signal preprocessing, feature extraction and classification. Finally, the mean classification accuracy rate, P stimulus recognition rate and I stimulus recognition rate of twelve subjects using leave-one-out cross validation were obtained with the value of95.53%,90.99%and96.66%respectively. The mean classification accuracy rate has an increase of4.44and3.46percentage point above the methods in two related literatures. Moreover, our method outperformed the two related literatures with an increase of11.99and5.99percentage point in P stimulus recognition rate, simultaneously, with an increase of2.56and2.83percentage point in I stimulus recognition rate.Thirdly, according to the EEG signal processing method based on multi-domain integration and genetic algorithm, a new P300-based deception detection method was proposed. Simulated network fraud was conducted and text information correlated with the crime details was extracted as probe stimulus in the three-stimulus GKT paradigm. EEG signal of twelve subjects was acquired during the deception detection experiment. The proposed method based on multi-domain integration and genetic algorithm was applied for simulated network fraud investigations, verifying the feasibility of this P300-based deception detection method in simulated network fraud condition. Inaddition, with the aim to improve the practicability of the proposed P300-based deception detection method, optimizing has been made in some detail. With leave-one-out cross validation,96.81%mean classification rate of twelve subjects was obtained. When determining the role type of the subjects, if the Pp%and PI%were set to be93%and92%, the individual diagnostic rate of the proposed method can reach100%, where Pp%was the probability of P stimulus identified as P stimulus, and PI%was the probability of P stimulus identified as I stimulus.Finally, we introduce our P300-based deception detection platform, which is designed and constructed using E-prime software, NATION NCERP-D36EEG and Matlab software.