| Cognitive concealed information detection or deception detection has a special role in areas such as national security,criminal justice and staff screening.Traditional polygraph has been widely used in some areas in several countries.However,its scientificity and effectiveness is not sufficient and has been questioned by some scientists.In order to apply deception detection as a legal basis to criminal justice and national security,it is important to strength the research of its testing theory and analytical methods.Besides,it is necessary to explore other alternative and enhancement technologies to improve the accuracy of the testing.In recent years,with the development of brain and cognitive science and brain imaging technology,brain computer interface(BCI)based lie detection technology has received extensive attention and rapid development.In general,cognitive concealed information detection is often referred as concealed information test(CIT).Lying,in essence,is a complex cognitive process which involves a range of advanced cognitive functions.While the brain computer interface and brain imaging based technology can reflect this cognitive activities intuitively.Only after we have a comprehensive understanding of the basic brain cognitive mechanisms can we design a more effective lie detection system.Besides,with the development and widely use of machine learning in artificial intelligence,combining brain computer interface and machine learning to construct a pattern recognition system for concealed information test is necessary.Based on the brain computer interface and lie detection paradigm,this thesis designed several lie detection experiments under visual,audio and tactile stimulation.Through the synchronization analysis of multiple channel EEG signals from the whole brain,brain activities under lying and truth telling is analyzed in details from the perspective in time,frequency and spatial domain and functional brain networks,and combined with the pattern classifiers to complete condition classification and recognition.The main contributions and innovations of this dissertation are listed as below:1.One method based on multi-channel P300 amplitude geometric difference and another method based on "feature ratio for key node" from brain network were proposed.In order to study the time,frequency and spatial features of brain signals from the whole brain under lying and truth telling condition,a concealed information test based on visual stimulus was designed.Analysis was done in time and frequency domain for P300 from the whole brain.It was found in the first method that geometric difference between target and irrelevant stimulus was significantly greater than which between target and probe stimulus in lying group.However,the difference was not significant in truth telling group.The results in the second method showed that the feature ratio between target and irrelevant stimulus was significantly different from the ratio between target and probe stimulus in lying group,but not for truth telling group.Based on these differences,it can distinguish lying from truth telling.Test results showed that both methods have higher recognition accuracy and verified the feasibility of these methods.2.A method based on EEG binary network and support vector machine was proposed.In order to carry out quantitative analysis of functional brain network during lying,through which can achieve the computerization of the test process,concealed information tests based on visual stimulation and auditory stimulation were designed,corresponding binary networks under different conditions were constructed based on nonlinear interdependence index.Feature recognition between lying and truth telling can be achieved by combining the network features and support vector machine.Test results showed it can achieve higher classification accuracy,which verified the feasible and effective of this method.Compare analysis of brain network under different conditions showed that functional networks under both visual and auditory stimuli have small world-ness,and small world feature of networks are enhanced during lying process.In addition,it showed that there is no significant difference between visual and auditory paradigm,but the recognition rate of visual stimuli is higher than which of auditory stimuli.It is the first time to apply functional connectivity analysis to concealed information test and achieved good results,which provides a reference for further understanding of the cognitive mechanism during the lying process.3.A method based on EEG weighted network and quantum neural network classifier was proposed.The main disadvantage of binary network is that lots of valuable information is lost after the threshold filtering,while the weighed network can reflect the connection structure of the network more realistically.Based on the results of above mentioned binary network,a concealed information test based on audio-visual synchronous stimulation was designed,and the comparison with visual stimulation test was completed.The weighted network was constructed by phase lag index and a new concept,which can achieve quantitative analysis of overall complexity for the whole brain,global feature entropy was proposed.Finally,a neural network classifier based on quantum gate nodes was designed,and the feature parameters and global feature entropy are combined to realize the detection of the concealed information.Test results showed that the proposed method can effectively identify deception from truth telling,which verify the effectiveness and feasibleness of this method.In addition,the comparison between audio-visual synchronization and visual stimulation showed that the combination of visual and auditory modality is better than visual stimuli only,which provide a support for the fusion of multimodal stimulation.4.Causal network induced by the information of familiar and stranger person was constructed,and a threshold selection method based on the "ratio between flow and degree”was proposed.Cognitive process of brain involves coordination and cooperation between different brain regions,but how is the information in different regions transmitted during lying process?And is there any certain dependence between brain regions?It is impossible to answer these questions through the above mentioned synchronization methods.The foundation of concealed information test is that brain’s response to the information of familiar person is different from stranger person.Therefore,based on the test data from audiovisual synchronization stimulus,corresponding directional functional network was constructed by transmit entropy in delta,theta,beta and alpha band for ERPs induced by familiar and strange information.Then,by calculating the ratio between flow and degree of the network in each frequency band to screen the network.Comparative analysis found that there are significant differences of the location for key nodes and the information flow direction between different regions between familiar and strange networks,which means cognitive involvement of brain regions for familiar person is different from stranger person.The directed function network analysis in this thesis provides an important reference for understanding the brain’s response to familiar information and the corresponding information flow between different brain regions.In addition,it is also useful for exploring the application of other bands except P300 to the deception detection.5.Comparative analysis of network features between truth telling and lying condition in BCl based concealed information test were carried out by using six methods which respectively present time and frequency domain,linear and nonlinear domain,general synchronization and phase synchronization,found the best method for network construction for CIT.Numerous synchronization methods are available for function network construction.However,the theoretical basis and calculation process of different methods are not the same.It means that the representation of the dependence of the same data sets is different,which will have a certain impact of the test results.Based on the graph theory,this thesis made a comparative analysis of the functional networks constructed by several synchronization methods under lying and truth telling conditions.The results showed that compared with truth telling condition,connection strength of networks and the small-worldness were enhanced in lying condition,and the network parameters showed significant differences between these two conditions.In addition,network parameters did not have significant differences between methods from linear and nonlinear,time domain and frequency domain.During these methods,function network constructed by mutual information has the best effect to distinguish lying from truth telling.These results provided a reference for synchronization selection for function network construction in BCI based concealed information test.6.A CIT method based on visual stimulation and tactile feedback was proposed.Tactile perception is another important system of human body in addition to visual and auditory system,and it is an important sensory channel for brain computer interface.Next concealed information test system will be a comprehensive of multiple stimulus,virtual reality and artificial intelligence.At the same time,add of tactile feedback stimulus in CIT has certain value for the design of countermeasure resistant system.Thus it is necessary to understand the cognitive mechanisms of brain under tactile stimulus first.This thesis designed a multi-stable tactile perception experiment and a prethreshold tactile perception experiment with EEG recorded simultaneously.It was found that there is no difference of brain activities between different perceptual conditions in multi-stable experiment,while the prethreshold experiment can express the processes of subject’s perception of tactile stimulation better.In addition,it also shows that the more certain the sensory state of the brain,the less active it is.Based on the discussion results,the prethreshold tactile stimulus was added into the visual based CIT system,in which tactile stimulus was appeared before visual stimulus to deal with the possible countermeasure in the test.The results showed that this system has good effect in resistance to countermeasures.The addition of tactile stimulus to visual based CIT system has no effect of brain’s recognition ability to concealed information,thereby can improve the stability and reliability of the CIT system.This thesis studied the time,frequency and spatial features of ERP signals and the functional brain network characteristics in the process of BCI based concealed information test.It is the first time to give a detail analysis of the neural oscillatory activity under deception from the whole brain,which is important for understanding the brain cognitive mechanism of deception.Besides,we made a deep analysis of the brain cognitive activities under visual,auditory and tactile stimulus,which provides a theoretical basis for the design of multi-modal BCI based concealed information test.The methods proposed in this thesis are of value as a reference in both theoretical and practical aspects. |
PDF Full Text Request