Countermeasures method and system based on cognitive load

Abstract

The present application belongs to the field of psychological assessment and lie detection technology, and particularly relates to a counter-adversarial lie detection method and system based on cognitive load. The method comprises: performing a counter-adversarial test based on cognitive load on a subject, and obtaining a subject response in a test process; wherein a stimulus paradigm of the counter-adversarial test contains a tension peak test sequence, a cognitive load exercise sequence and a counter-adversarial test sequence, and each sequence contains a plurality of different types of stimuli as test questions; screening all subject responses in the tension peak test sequence, the cognitive load exercise sequence and the counter-adversarial test sequence that meet a preset response condition as target responses, and extracting a feature vector of the target responses; and performing lie detection based on the feature vector of the target responses. The present application aims to overcome the defect of detection misjudgment caused by an adversarial strategy in a test process.

Description

Technical Field

[0001] This invention belongs to the field of psychological testing and lie detection technology, specifically relating to a method and system for resisting lie detection based on cognitive load. Background Technology

[0002] Lie detection technology has important applications in criminal investigation and national security. The test experiment paradigm is the core issue design of the lie detection experiment process, namely how the test subject is questioned, how the question categories are arranged, and what psychological mechanisms are used to induce differential responses. It determines the data collection method, the organization logic of the questions, and the basis for interpreting the results in the lie detection experiment, which is very important for ensuring the validity and reliability of the assessment.

[0003] A lie detection paradigm refers to the design and presentation of specific questions to induce observable differences in the physiological or neurological responses of subjects during their answers, thereby inferring the subjects' true intentions or cognitive states. Currently, common lie detection paradigms mainly include the Control Question Test (CQT) and the Concealed Information Test (CIT). The Control Question Test alternates between "case-related questions" and "case-irrelevant but potentially stressful control questions," comparing the subjects' psychological, physiological, or behavioral responses to the two types of questions. If a subject's response to the case-related questions is significantly stronger than their response to the control questions, it is inferred that they are deceiving the subject. The Concealed Information Test presents a set of options, of which only one or some are relevant to the case. If a subject exhibits significant psychological, physiological, or behavioral responses to the relevant options, it is inferred that they possess case information and are deceiving the subject.

[0004] However, current lie detection testing paradigms generally suffer from insufficient adaptability, especially in highly adversarial and deceptive scenarios, where traditional experimental assessment methods exhibit low sensitivity and effectiveness. For example, subjects can suppress the psychological and physiological signal differences between "relevant information" and "irrelevant information" through simple adversarial strategies such as deliberate deep breathing and muscle tension, leading to a high false negative rate. Innocent individuals, due to fear of being wrongly accused, fear of the lie detector, or a sensitive and easily stressed personality, may experience equal or even stronger psychological and physiological arousal, resulting in a high false positive rate.

[0005] Chinese patent document CN111616702A discloses a lie detection analysis system based on cognitive load enhancement, and Chinese patent document CN118592958A discloses an anti-lie detection method based on P300 and CNV with a feedback mode. Both methods increase the cognitive load of the test subject through a feedback interface. However, the feedback interface does not enhance cognitive load after the subject engages in deception, and cognitive resources can still be allocated to counter-lie detection. Therefore, the counter-countermeasure effect is limited. It is necessary to construct an anti-countermeasure experimental paradigm and analysis method that utilizes cognitive resources when the subject engages in deception. Furthermore, the above patent methods rely solely on EEG signals. However, EEG is susceptible to interference from eye movements, electromyography, and motion noise, and single-modal data cannot distinguish between reactions caused by lying and reactions caused by emotional tension. Therefore, it is necessary to construct a multimodal simultaneous acquisition and analysis system integrating behavioral, electrophysiological, and EEG data.

[0006] Chinese patent document CN113180668A discloses a real-time functional magnetic resonance imaging (fMRI) lie detection system based on changes in cognitive load. This system adjusts the cognitive load by varying the number of images the subject remembers, increasing the difficulty of lying under high-load memory conditions. However, under simple tasks, the subject experiences low memory pressure, leaving ample cognitive resources available for counter-lie detection. Furthermore, for subjects with weak memory, a small number of images may already represent a high load, while for subjects with strong memory, a large number of images may still represent a low load, failing to achieve stable and effective lie detection. In addition, this method relies on fMRI data acquisition, resulting in bulky and expensive equipment, and limitations in the detection environment.

[0007] Therefore, it is necessary to invent a new method and system for resisting lie detection. By introducing specific cognitive load tasks for lying subjects, the system enhances resistance through question design and task load setting, weakening the cognitive resources that lying subjects might allocate to suppressing psychological and physiological responses when lying, thereby improving the reliability and application value of lie detection results. Simultaneously, a differentiated judgment method based on intra-individual comparisons should be established to avoid misjudgments caused by group standards. Summary of the Invention

[0008] The present invention aims to overcome at least one of the defects of the prior art and provide a cognitive load-based anti-lie detection method and system to reduce detection misjudgments caused by adversarial strategies during the testing process.

[0009] The detailed technical solution of this invention is as follows: A cognitive load-based method for resisting lie detection, the method comprising: The participants were subjected to a cognitive load-based opposition test, and their responses during the test were obtained. The stimulus paradigm of the opposition test included a tension peak test sequence, a cognitive load exercise sequence, and an opposition test sequence, and each sequence contained several different types of stimuli as test questions. The participants' responses included behavioral responses and physiological signals. All subject responses that meet the preset response conditions in the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence are selected as target responses, and the feature vectors of the target responses are extracted; wherein, the preset response conditions include correct behavioral response category and / or behavioral response duration less than the response duration threshold; Lie detection is performed based on the feature vector of the target response.

[0010] According to a preferred embodiment of the present invention, the type of stimulus includes target stimulus, control stimulus, and irrelevant stimulus, wherein: The target stimulus is defined as hidden information known only to the liar and the person involved. The control stimulus is defined as irrelevant information known to both the liar and the non-liar; The irrelevant stimulus is defined as irrelevant information that neither the liar nor the non-liar is aware of.

[0011] According to a preferred embodiment of the present invention, the cognitive load-based objection test performed on the subject specifically includes: First, the test is conducted based on the tension peak test sequence, which contains several target stimuli, control stimuli, and irrelevant stimuli in a ratio of 1:1:4. During the test, the control stimuli, irrelevant stimuli, irrelevant stimuli, target stimuli, irrelevant stimuli, and irrelevant stimuli are presented in sequence according to a preset sequence. The test was conducted sequentially based on the cognitive load training sequence, which contained several control stimuli and irrelevant stimuli with a probability ratio of 1:1. During the test, each stimulus was presented randomly twice, once instructing the subject to answer truthfully according to the actual situation, and once instructing the subject to answer ironically according to the opposite situation. The test is conducted sequentially based on the opposition test sequence, which includes several target stimuli, control stimuli, and irrelevant stimuli with a probability ratio of 1:1:4. During the test, each stimulus is presented randomly twice, once instructing the subject to answer truthfully according to the actual situation, and once instructing the subject to answer ironically according to the opposite situation.

[0012] According to a preferred embodiment of the present invention, selecting all subject responses that meet the preset response conditions from the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence as target responses specifically includes: During the test, the presented stimuli and the subjects' responses to the stimuli are labeled in real time, generating stimulus labels and behavioral response labels that include the type and duration of the behavioral response. Determine whether the behavioral response category and behavioral response duration under all behavioral response tags meet the preset response conditions. If the behavioral response category is correct and the behavioral response duration is less than the response duration threshold, then retain the subject's response under that stimulus; otherwise, discard it. All retained participant responses were used as target responses.

[0013] According to a preferred embodiment of the present invention, for the behavioral response category, the following is set: The correct response category for the target stimulus and irrelevant stimulus is "No"; the correct response category for the control stimulus is "Yes"; and under the instruction to give a truthful response, the correct response category for the target stimulus and irrelevant stimulus is "No", and the correct response category for the control stimulus is "Yes"; under the instruction to give a sarcasm response, the correct response category for the target stimulus and irrelevant stimulus is "Yes", and the correct response category for the control stimulus is "No".

[0014] According to a preferred embodiment of the present invention, extracting the feature vector of the target response specifically includes: The behavioral responses and physiological signals in the target response are preprocessed; wherein, the behavioral responses include behavioral response category and behavioral response time, and the physiological signals include skin conductance signals and electroencephalogram (EEG) signals; Feature extraction is performed on the preprocessed behavioral responses and physiological signals, including: The accuracy and reaction time of the behavior response category are extracted as feature vectors of the behavior response; Based on the stimulation markers generated during the test and the preset time window, the preprocessed skin conductance signal is segmented and mean-centered. The skin conductance level (SCL) and skin conductance response (SCR) are decomposed from the skin conductance signal using the NeuroKit2 tool to calculate the average skin conductance level, SCR amplitude, rise time and recovery time, and these are used as the feature vector of the skin conductance signal. Based on the stimulus markers generated during the test, the continuous EEG signal is segmented. Taking the moment of stimulus occurrence as a reference point, EEG data segments within a preset time window are extracted. The average signal value within the preset time window before the stimulus occurrence is used as the baseline. The EEG signal of each time window is baseline corrected. The effective trials of the same type of stimulus are time-aligned and averaged point by point to obtain a stable ERP waveform. Target features are extracted from the ERP waveform as the feature vector of the EEG signal. The target features include peak amplitude, peak latency, and waveform slope.

[0015] According to a preferred embodiment of the present invention, preprocessing of the behavioral responses and physiological signals in the target response specifically includes: Based on the stimulus markers generated during the test, the behavioral responses and physiological signals are time-aligned and synchronized; the physiological signals are band-pass filtered to retain signals within a preset frequency band; Furthermore, wavelet transform is used for noise reduction of electrodermal signals; for electroencephalogram (EEG) signals, average reference or binaural reference is used to rereference the signals, and independent component analysis is used to remove interference caused by eye movement, electromyography, electrocardiography, or motion.

[0016] According to a preferred embodiment of the present invention, lie detection based on the feature vector of the target response specifically includes: For the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence, the feature vectors of all extracted target responses are normalized by zscore within the corresponding sequence, and all feature vectors under each sequence are grouped based on stimulus labels to form feature vector groups corresponding to target stimulus, control stimulus, and irrelevant stimulus under each sequence. Calculate the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the tension peak test sequence, and perform a t-test to generate the first t-statistic and the first p-value. Calculate the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the anti-resistance test sequence and perform a t-test to generate a second t-statistic and a second p-value; wherein the feature vector groups corresponding to the target stimulus and the irrelevant stimulus are both included in the feature vectors under the truth response instruction and the feature vectors under the anti-resistance response instruction. The first p-value and the second p-value are both compared with a preset p-test threshold. If both the first p-value and the second p-value are less than the preset p-test threshold, the subject is determined to be lying.

[0017] According to a preferred embodiment of the present invention, lie detection based on the feature vector of the target response further includes: Both the first p-value and the second p-value are compared with a preset p-test threshold. If at least one of the first p-values ​​and the second p-value is greater than or equal to the preset p-test threshold, then: Calculate the mean of the feature vector group corresponding to irrelevant stimuli and under the irrational response instruction in the cognitive load exercise sequence and the mean of the feature vector group under the truth response instruction, and perform a t-test to generate the corresponding third t-statistic; Calculate the mean of the feature vector group corresponding to the target stimulus and under the truth response instruction in the anti-resistance test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and under the truth response instruction, and perform a t-test to generate the fourth t-statistic. Calculate the mean of the feature vector group corresponding to the target stimulus and under the irrational response instruction in the anti-resistance test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and under the irrational response instruction, and perform a t-test to generate the fifth t-statistic. Sort all first, third, fourth, and fifth t-statistics in ascending order of their values. The higher the proportion of the t-statistics from the tension peak test sequence or the anti-confrontation test sequence, the greater the probability that the subject is lying; among which, This represents the total number of all t-statistics.

[0018] In another aspect of the invention, a cognitive load-based anti-lie detection system is also provided, which applies the cognitive load-based anti-lie detection method described above, the system comprising: The subject testing module is used to conduct cognitive load-based opposition tests on subjects and obtain subject responses during the testing process; wherein, the stimulus paradigm of the opposition test includes a tension peak test sequence, a cognitive load practice sequence, and an opposition test sequence, and each sequence contains several different types of stimuli as test questions; the subject responses include behavioral responses and physiological signals. The feature acquisition module is used to filter all subject responses that meet the preset response conditions in the tension peak test sequence, cognitive load exercise sequence and opposition test sequence as target responses, and extract the feature vector of the target response; wherein, the preset response conditions include correct behavioral response category and / or behavioral response duration less than the response duration threshold; The lie detection module is used to perform lie detection based on the feature vector of the target's response.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention introduces truth-telling stimuli and anti-truth stimuli, wherein anti-truth stimuli amplify the cognitive load of lying subjects, weaken the cognitive resources they allocate to suppressing psychological and physiological responses and other adversarial behaviors, and suppress the subjects' adversarial strategies. (2) This invention introduces a cognitive load training sequence and proposes an individual-specific method for judging lies. Compared with traditional group lie detection methods, it avoids misjudgment caused by the influence of different psychological qualities of individuals. (3) This invention uses multimodal signal synchronous acquisition and analysis, and combines behavioral, electrophysiological and EEG signals for multi-dimensional cross-verification, rather than the traditional single-modal lie detection. Attached Figure Description

[0020] Figure 1 This is a flowchart of the cognitive load-based anti-lie detection method described in this invention; Figure 2It is a schematic diagram of the execution process of a stimulation paradigm in Embodiment 1 of the present invention; Figure 3 It is a schematic diagram of multimodal signal acquisition in Embodiment 1 of the present invention; Figure 4 It is a schematic diagram of the process for lie determination of a subject in Embodiment 1 of the present invention. Detailed implementation manners

[0021] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are all exemplary and intended to provide further descriptions of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] In the case of no conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0025] Embodiment 1, Refer Figure 1 , this embodiment provides an anti-countermeasure lie detection method based on cognitive load, and the method includes: S1. Conduct an anti-countermeasure test on the subject based on cognitive load, and obtain the subject's responses during the test.

[0026] In this embodiment, the stimulation paradigm of the anti-countermeasure test includes a tension peak test sequence, a cognitive load practice sequence, and an anti-countermeasure test sequence, and each sequence includes several different types of stimuli as test questions.

[0027] Specifically, the types of the stimuli may include target stimuli, control stimuli, and irrelevant stimuli.

[0028] The target stimulus is defined as hidden information known only to liars and involved persons. In a specific exemplary case, when the lost item is a diamond and only the thief and the case handling personnel know this information, the target stimulus may be: "Was the stolen item a diamond?"

[0029] The controlling stimulus is defined as irrelevant information known to both the liar and the non-liar. In a specific example, when a picture of a pearl necklace is displayed on the screen, the controlling stimulus could be "Is this a pearl necklace?".

[0030] The irrelevant stimulus is defined as information that is unknown to both the liar and the non-liar. In a specific example, when a picture of a stranger is displayed on the screen, the irrelevant stimulus could be "Do you know this person?".

[0031] Understandably, the method for generating the target stimulus may be to ask whether or not to ask about key information that the subject wants to hide; the method for generating the control stimulus may be to ask whether or not to ask about items that the subject sees on the screen; and the method for generating irrelevant stimuli may be to ask whether or not to ask about items or people that the subject does not know (such as strangers).

[0032] In this embodiment, the participants undergo an objection test based on cognitive load, which may specifically include: First, the test is conducted based on the aforementioned tension peak test sequence. This sequence contains 24 stimuli used as test questions. The stimulus types include target stimuli, control stimuli, and irrelevant stimuli in a ratio of 1:1:4. The test begins with a control stimulus, and at least one irrelevant stimulus is present before and after the target stimulus. For example, the test could present the control stimulus, irrelevant stimulus, irrelevant stimulus, target stimulus, irrelevant stimulus, and irrelevant stimulus in a preset sequence during the test.

[0033] The tests are conducted sequentially based on the cognitive load training sequence. Specifically, the cognitive load training sequence test begins after the tension peak test sequence is completed. The cognitive load training sequence contains 60 stimuli used as test questions. The stimulus types consist only of control stimuli and irrelevant stimuli, with a probability ratio of 1:1. Each stimulus is presented randomly twice during the test; once the subject is instructed to answer truthfully according to the actual situation, and once the subject is instructed to answer ironically according to the opposite situation.

[0034] The tests are conducted sequentially based on the aforementioned opposition test sequence. That is, after completing the cognitive load exercise sequence test, the opposition test sequence test begins. The opposition test sequence contains 24 stimuli used as test questions. The stimulus types include target stimuli, control stimuli, and irrelevant stimuli, with a probability ratio of 1:1:4. During the test, each stimulus is presented randomly twice: once instructing the subject to answer truthfully according to the actual situation, and once instructing the subject to answer ironically according to the opposite situation.

[0035] Understandably, the above truth-telling response instruction is used to instruct the subject to answer truthfully during the test process to increase the cognitive load of the lying subjects. In a specific demonstration example, when a picture of a pearl necklace is displayed on the screen, the question "Is this a pearl necklace?" is asked and the subject is instructed to tell the truth, and the subject should answer "Yes".

[0036] The above reverse-telling response instruction is used to instruct the subject to give an answer opposite to the truth during the test process to increase the cognitive load of all subjects. In a specific demonstration example, when a picture of a pearl necklace is displayed on the screen, the question "Is this a pearl necklace?" is asked and the subject is instructed to speak in reverse, and the subject should answer "No".

[0037] See Figure 2 , for each stimulus in the sequence, the process of testing the subject can be as follows: First, a fixation point "+" appears in the center of the screen and lasts for 500 ms; The fixation point disappears, the stimulus appears and lasts for 2000 ms, and the subject receives the stimulus information but does not respond; The stimulus disappears, the options "Yes" and "No" appear and last for 1000 ms, and the subject makes a quick judgment and makes a choice within 1000 ms; After the subject makes a choice, the screen randomly gives feedback of "Honesty +2" or "Honesty -2" and lasts for 1000 ms; then a blank screen of 12000 ms is presented to allow the subject's physiological response to return to the normal level.

[0038] Furthermore, to ensure that the subject can receive the stimulus, this step can also include: When presenting the stimulus, the text information of the stimulus is displayed on the screen, the pre-configured lie-detection question voice is played synchronously, and the picture of the item involved in the stimulus is displayed. In a specific demonstration example, the voice file and picture file corresponding to each stimulus can be automatically generated through automatic synthesis software such as text-to-speech (TTS) and optical character recognition (OCR).

[0039] Based on the above test process, the subject's responses are collected in real time. The subject's responses include behavioral responses and physiological signals. The behavioral responses include the category of behavioral responses and the time of behavioral responses. The physiological signals include galvanic skin response signals and electroencephalogram signals.

[0040] Specifically see Figure 3 , the collection of behavioral responses and the presentation of stimuli can use the Eprime psychological experiment software; the body movement information of the subject is collected through a camera, and the category of the subject's behavioral responses and the time of behavioral responses are obtained. Understandably, the category of behavioral responses can be the answer given by the subject to the currently presented stimulus; the time of behavioral responses can be the time used by the subject to make an answer starting from the moment the options "Yes" and "No" appear on the screen.

[0041] Electrodermal signal acquisition can be performed using the Biopac MP160 system with AcqKnowledge software. When acquiring electrodermal signals, wipe the subject's finger area with an alcohol swab, then place the electrodes on the finger area, ensuring close contact between the electrodes and the skin. Electrodermal signals can be transmitted to the MP160 main unit via a BNC cable.

[0042] EEG signal acquisition can be performed using Neuroscan's 64-channel recording system, with Curry8 as the data acquisition software. The subject wears a 64-channel EEG cap connected to an EEG amplifier, and the electrode distribution follows the international 10-20 system, with the bilateral mastoid dual electrodes used as a reference. During the acquisition process, the impedance of each electrode is kept below 10KΩ.

[0043] Based on the above, the participants were tested on cognitive load-based resistance, and the corresponding participant response information was obtained.

[0044] S2. Select all subject responses that meet the preset response conditions from the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence as target responses, and extract the feature vector of the target responses.

[0045] Specifically, selecting all subject responses that meet the preset response conditions from the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence as target responses includes the following process: First, during the test, the presented stimuli and the subject's response to the stimuli are labeled in real time, generating stimulus labels and behavioral response labels containing the behavioral response category and duration.

[0046] That is, the stimulus occurrence event (i.e. the type of stimulus presented), the subject's reaction time under the stimulus, the subject's reaction behavior, etc. are simultaneously labeled in a multimodal manner. Electrical pulse signals are sent to all acquisition devices simultaneously to mark the stimulus sequence number, behavioral event type and other information in real time. This information is stored synchronously with the subject's EEG, TE data and other data, and finally a stimulus label and a behavioral response label containing the behavioral response category and behavioral response duration are formed.

[0047] In this embodiment, the labeling of the stimulus occurrence event (i.e., the type of stimulus presented) includes: labeling the target stimulus, control stimulus, and irrelevant stimulus without instruction, such as labeling them as 1, 2, and 3 respectively; labeling the target stimulus, control stimulus, and irrelevant stimulus under the instruction of truthful response, such as labeling them as 4, 5, and 6 respectively; and labeling the target stimulus, control stimulus, and irrelevant stimulus under the instruction of sarcasm response, such as labeling them as 7, 8, and 9 respectively. In this embodiment, the subject's reaction time under the stimulus is the duration from the moment the stimulus occurs to the moment the subject responds, as recorded by the Eprime program. In this embodiment, the labeling of the subject's response behavior is to mark the subject's "yes" or "no" response, such as labeling "yes" as 11 and "no" as 12.

[0048] Then, it is determined whether the behavioral response category and behavioral response duration under all behavioral response tags meet the preset response conditions. If the behavioral response category is correct and the behavioral response duration is less than the response duration threshold, the subject's response under that stimulus is retained; otherwise, it is discarded.

[0049] That is, based on the behavioral response markers collected from the subjects during the test, preset information for data filtering is extracted. Understandably, the preset information for data filtering includes the behavioral response category and the behavioral response duration.

[0050] The system compares the preset information of the subject's behavioral response with the corresponding preset response conditions. In this embodiment, the preset response conditions include a correct behavioral response category and / or a behavioral response duration less than a response duration threshold. When the preset information in the subject's behavioral response does not meet the corresponding preset conditions, the corresponding subject response information is deleted. For example, if the subject's response category is different from the preset correct response category, the subject response information is deleted; if the subject's response duration is greater than the response duration threshold, the subject response information is deleted.

[0051] Furthermore, for the behavioral response categories under different types of stimuli, the following are pre-defined: The correct response category for the target stimulus and irrelevant stimulus is "No"; the correct response category for the control stimulus is "Yes"; and under the instruction to give a truthful response, the correct response category for the target stimulus and irrelevant stimulus is "No", and the correct response category for the control stimulus is "Yes"; under the instruction to give a sarcasm response, the correct response category for the target stimulus and irrelevant stimulus is "Yes", and the correct response category for the control stimulus is "No".

[0052] The reaction time threshold can be understood as follows: starting from the moment the "yes" and "no" options appear, the subject needs to make a choice within a preset reaction time. However, if the time taken to make a choice exceeds the preset reaction time threshold, the subject's reaction information will be deleted.

[0053] It should be further explained that when the behavioral or physiological response information collected within a preset time period contains severe noise interference that cannot be removed by filtering or other preprocessing methods, the response information for that time period will be deleted. In a specific example, when the subject's body is moving significantly, the response information for that time period will be deleted; when the signal amplitude is too high (e.g., EEG signals exceeding ±100μV) or fluctuates irregularly, the response information for that time period will be deleted; when, after routine preprocessing, the power frequency (45-55Hz) energy accounts for more than 1 / 5 of the total energy, the response information for that time period will be deleted.

[0054] Finally, all retained subject responses will be used as target responses. The behavioral responses and physiological signals contained in the target responses will be used for subsequent lie detection assessments.

[0055] In this embodiment, the feature vector of the target response is extracted, specifically including the following process: First, the behavioral responses and physiological signals in the target response are preprocessed.

[0056] The behavioral response includes the behavioral response category and the behavioral response time, and the physiological signals include skin conductance signals and electroencephalogram (EEG) signals.

[0057] Based on the stimulus markers generated during the testing process, the behavioral responses and physiological signals are time-aligned and synchronized using different modalities; the collected physiological signals are bandpass filtered to remove low-frequency drift and high-frequency noise, retaining signals within the preset frequency band; and for electrodermal signals, wavelet transform is used for noise reduction; for electroencephalogram (EEG) signals, the signals are rereferenced using an average reference or binaural reference method, and independent component analysis (ICA) is used to remove interference caused by eye movements, electromyography (EMG), electrocardiography (ECG), or movement.

[0058] Then, feature extraction is performed on the preprocessed behavioral responses and physiological signals, including the following processes: The accuracy and reaction time of the behavior response category are extracted as feature vectors of the behavior response.

[0059] For the electrodermal signal, signal processing methods were used to extract features from the electrophysiological signal. Specifically, based on the stimulus markers (target stimulus, control stimulus, irrelevant stimulus) generated during the test and a preset time window, the preprocessed electrodermal signal was segmented and mean-centered (subtracting the mean of the entire segment). The NeuroKit2 tool was used to decompose the skin conductance level (SCL) and skin conductance response (SCR) from the electrodermal signal to calculate the mean skin conductance level (Mean SCL), SCR amplitude, rise time, and recovery time, etc., and these were used as the feature vector of the electrodermal signal.

[0060] For EEG signals, the event-related potential (ERP) method was used to extract ERP features. Specifically, based on stimulus markers (target stimulus, control stimulus, and irrelevant stimulus) generated during the test, the continuous EEG signal was segmented. Using the stimulus occurrence time as a reference point, EEG data segments within a preset time window (e.g., from 200ms before the stimulus to 800ms after the stimulus) were extracted. The average signal value within the preset time window before the stimulus occurrence was used as the baseline, and baseline correction was performed on the EEG signal of each time window to eliminate the effects of slow drift and reference offset. Furthermore, effective trials of the same type of stimulus were time-aligned and averaged point by point to obtain a stable ERP waveform. Target features, such as peak amplitude, peak latency, and waveform slope, were extracted from the ERP waveform and used as the feature vector of the EEG signal.

[0061] Based on the above, the screening and feature extraction of subject response information under the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence were completed.

[0062] S3. Perform lie detection based on the feature vector of the target response.

[0063] Specific reference Figure 4 For the tension peak test sequence, cognitive load exercise sequence, and anti-resistance test sequence, the feature vectors of all extracted target responses are normalized by z-score within the corresponding sequence, and all feature vectors under each sequence are grouped based on stimulus labels to form feature vector groups corresponding to target stimulus, control stimulus, and irrelevant stimulus under each sequence.

[0064] Next, the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the tension peak test sequence are calculated and a t-test is performed to generate the target stimulus-irrelevant stimulus t-statistic and the corresponding p-value of the tension peak test sequence, which are denoted as the first t-statistic and the first p-value.

[0065] Calculate the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the opposition test sequence, and perform a t-test. This generates the target-irrelevant stimulus t-statistic and the corresponding p-value for the opposition test sequence, denoted as the second t-statistic and the second p-value. It should be noted that the feature vector groups corresponding to the target stimulus and the irrelevant stimulus both include the feature vectors under the truth response instruction and the feature vectors under the irrational response instruction.

[0066] Both the first p-value and the second p-value are compared with a preset p-test threshold. Assuming the preset p-test threshold is 0.05, if both the first p-value and the second p-value are less than the preset p-test threshold, it means that there are significant differences between the target stimulus and irrelevant stimulus characteristics in the tension peak test sequence and the opposition test sequence, and the subject is judged to be lying.

[0067] If at least one of the first and second p-values ​​is greater than or equal to a preset p-test threshold, indicating that there is no significant difference between the target stimulus and irrelevant stimulus characteristics in the tension peak test sequence and the opposition test sequence, then the probability of the subject lying is estimated through comparative analysis, specifically including: Calculate the mean of the feature vector group corresponding to irrelevant stimuli under the sarcasm response instruction and the mean of the feature vector group under the truth response instruction in the cognitive load training sequence, and perform a t-test to generate the sarcasm response instruction-truth response instruction t-statistic of the cognitive load training sequence, denoted as the third t-statistic.

[0068] Calculate the mean of the feature vector group corresponding to the target stimulus and the truth response instruction in the opposition test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and the truth response instruction, and perform a t-test. Generate the target stimulus-irrelevant stimulus t-statistic in the opposition test sequence under the truth response instruction, denoted as the fourth t-statistic.

[0069] Calculate the mean of the feature vector group corresponding to the target stimulus and under the irrational response instruction in the anti-resistance test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and under the irrational response instruction, and perform a t-test to generate the target stimulus-irrelevant stimulus t-statistic in the anti-resistance test sequence under the irrational response instruction, denoted as the fifth t-statistic.

[0070] Sort all the first, third, fourth, and fifth t-statistics calculated above in ascending order of their values. If the first... The higher the proportion of samples in the t-statistic that come from the tension peak test sequence or the anti-confrontation test sequence (e.g., above 50%), the greater the probability that the subject is lying; conversely, if the proportion is lower, the probability is lower. If the proportion of samples from the tension peak test sequence or the anti-confrontation test sequence in the t-statistic is less than 50%, the subject is considered less likely to be lying. This represents the total number of all t-statistics.

[0071] The method described above calculates the proportion of t-statistics in the tension peak test sequence or the opposition test sequence that are greater than the t-statistics in the cognitive load exercise sequence, and uses this proportion to measure the subject's tendency to lie. That is, the larger the value, the higher the probability that the subject is lying. Compared with traditional group lie detection methods, it can avoid misjudgment caused by the influence of different individual psychological qualities and overcome the defects of detection misjudgment caused by the adversarial strategy during the testing process.

[0072] Example 2 This embodiment provides a cognitive load-based anti-lie detection system, which applies the cognitive load-based anti-lie detection method described in Embodiment 1. The system includes: The subject testing module is used to conduct cognitive load-based opposition tests on subjects and obtain subject responses during the testing process; wherein, the stimulus paradigm of the opposition test includes a tension peak test sequence, a cognitive load practice sequence, and an opposition test sequence, and each sequence contains several different types of stimuli as test questions; the subject responses include behavioral responses and physiological signals. The feature acquisition module is used to filter all subject responses that meet the preset response conditions in the tension peak test sequence, cognitive load exercise sequence and opposition test sequence as target responses, and extract the feature vector of the target response; wherein, the preset response conditions include correct behavioral response category and / or behavioral response duration less than the response duration threshold; The lie detection module is used to perform lie detection based on the feature vector of the target's response.

[0073] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for resisting lie detection based on cognitive load, characterized in that, The method includes: The participants were subjected to a cognitive load-based opposition test, and their responses during the test were obtained. The stimulus paradigm of the opposition test included a tension peak test sequence, a cognitive load exercise sequence, and an opposition test sequence, and each sequence contained several different types of stimuli as test questions. The participants' responses included behavioral responses and physiological signals. All subject responses that meet the preset response conditions in the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence are selected as target responses, and the feature vectors of the target responses are extracted; wherein, the preset response conditions include correct behavioral response category and / or behavioral response duration less than the response duration threshold; Lie detection is performed based on the feature vector of the target response.

2. The cognitive load-based anti-lie detection method according to claim 1, characterized in that, The types of stimuli include target stimuli, control stimuli, and irrelevant stimuli, wherein: The target stimulus is defined as hidden information known only to the liar and the person involved. The control stimulus is defined as irrelevant information known to both the liar and the non-liar; The irrelevant stimulus is defined as irrelevant information that neither the liar nor the non-liar is aware of.

3. The cognitive load-based anti-lie detection method according to claim 2, characterized in that, The aforementioned cognitive load-based opposition test for the subjects specifically includes: First, the test is conducted based on the tension peak test sequence, which contains several target stimuli, control stimuli, and irrelevant stimuli in a ratio of 1:1:

4. During the test, the control stimuli, irrelevant stimuli, irrelevant stimuli, target stimuli, irrelevant stimuli, and irrelevant stimuli are presented in sequence according to a preset sequence. The test was conducted sequentially based on the cognitive load training sequence, which contained several control stimuli and irrelevant stimuli with a probability ratio of 1:

1. During the test, each stimulus was presented randomly twice, once instructing the subject to answer truthfully according to the actual situation, and once instructing the subject to answer ironically according to the opposite situation. The test is conducted sequentially based on the opposition test sequence, which includes several target stimuli, control stimuli, and irrelevant stimuli with a probability ratio of 1:1:

4. During the test, each stimulus is presented randomly twice, once instructing the subject to answer truthfully according to the actual situation, and once instructing the subject to answer ironically according to the opposite situation.

4. The cognitive load-based anti-lie detection method according to claim 3, characterized in that, Selecting all subject responses that meet the preset response conditions from the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence as target responses, specifically including: During the test, the presented stimuli and the subjects' responses to the stimuli are labeled in real time, generating stimulus labels and behavioral response labels that include the type and duration of the behavioral response. Determine whether the behavioral response category and behavioral response duration under all behavioral response tags meet the preset response conditions. If the behavioral response category is correct and the behavioral response duration is less than the response duration threshold, then retain the subject's response under that stimulus; otherwise, discard it. All retained participant responses were used as target responses.

5. The cognitive load-based anti-lie detection method according to claim 4, characterized in that, For the aforementioned behavioral response category, the following settings are defined: The correct response category for the target stimulus and irrelevant stimulus is "No"; the correct response category for the control stimulus is "Yes"; and under the instruction to give a truthful response, the correct response category for the target stimulus and irrelevant stimulus is "No", and the correct response category for the control stimulus is "Yes"; under the instruction to give a sarcasm response, the correct response category for the target stimulus and irrelevant stimulus is "Yes", and the correct response category for the control stimulus is "No".

6. The cognitive load-based anti-lie detection method according to claim 4, characterized in that, Extracting the feature vector of the target response specifically includes: The behavioral responses and physiological signals in the target response are preprocessed; wherein, the behavioral responses include behavioral response category and behavioral response time, and the physiological signals include skin conductance signals and electroencephalogram (EEG) signals; Feature extraction is performed on the preprocessed behavioral responses and physiological signals, including: The accuracy and reaction time of the behavior response category are extracted as feature vectors of the behavior response; Based on the stimulation markers generated during the test and the preset time window, the preprocessed skin conductance signal is segmented and mean-centered. The skin conductance level (SCL) and skin conductance response (SCR) are decomposed from the skin conductance signal using the NeuroKit2 tool to calculate the average skin conductance level, SCR amplitude, rise time and recovery time, and these are used as the feature vector of the skin conductance signal. Based on the stimulus markers generated during the test, the continuous EEG signal is segmented. Taking the moment of stimulus occurrence as a reference point, EEG data segments within a preset time window are extracted. The average signal value within the preset time window before the stimulus occurrence is used as the baseline. The EEG signal of each time window is baseline corrected. The effective trials of the same type of stimulus are time-aligned and averaged point by point to obtain a stable ERP waveform. Target features are extracted from the ERP waveform as the feature vector of the EEG signal. The target features include peak amplitude, peak latency, and waveform slope.

7. The cognitive load-based anti-lie detection method according to claim 6, characterized in that, Preprocessing of the behavioral responses and physiological signals in the target response specifically includes: Based on the stimulus markers generated during the test, the behavioral responses and physiological signals are time-aligned and synchronized; the physiological signals are band-pass filtered to retain signals within a preset frequency band; Furthermore, wavelet transform is used for noise reduction of electrodermal signals; for electroencephalogram (EEG) signals, average reference or binaural reference is used to rereference the signals, and independent component analysis is used to remove interference caused by eye movement, electromyography, electrocardiography, or motion.

8. The cognitive load-based anti-lie detection method according to claim 4, characterized in that, The lie detection determination is based on the feature vector of the target response, specifically including: For the tension peak test sequence, cognitive load exercise sequence, and opposition test sequence, the feature vectors of all extracted target responses are normalized by zscore within the corresponding sequence, and all feature vectors under each sequence are grouped based on stimulus labels to form feature vector groups corresponding to target stimulus, control stimulus, and irrelevant stimulus under each sequence. Calculate the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the tension peak test sequence, and perform a t-test to generate the first t-statistic and the first p-value. Calculate the mean of the feature vector group corresponding to the target stimulus and the mean of the feature vector group corresponding to the irrelevant stimulus in the anti-resistance test sequence and perform a t-test to generate a second t-statistic and a second p-value; wherein the feature vector groups corresponding to the target stimulus and the irrelevant stimulus are both included in the feature vectors under the truth response instruction and the feature vectors under the anti-resistance response instruction. The first p-value and the second p-value are both compared with a preset p-test threshold. If both the first p-value and the second p-value are less than the preset p-test threshold, the subject is determined to be lying.

9. The cognitive load-based anti-lie detection method according to claim 8, characterized in that, The lie detection determination based on the feature vector of the target response specifically includes: Both the first p-value and the second p-value are compared with a preset p-test threshold. If at least one of the first p-values ​​and the second p-value is greater than or equal to the preset p-test threshold, then: Calculate the mean of the feature vector group corresponding to irrelevant stimuli and under the irrational response instruction in the cognitive load exercise sequence and the mean of the feature vector group under the truth response instruction, and perform a t-test to generate the corresponding third t-statistic; Calculate the mean of the feature vector group corresponding to the target stimulus and under the truth response instruction in the anti-resistance test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and under the truth response instruction, and perform a t-test to generate the fourth t-statistic. Calculate the mean of the feature vector group corresponding to the target stimulus and under the irrational response instruction in the anti-resistance test sequence and the mean of the feature vector group corresponding to the irrelevant stimulus and under the irrational response instruction, and perform a t-test to generate the fifth t-statistic. Sort all first, third, fourth, and fifth t-statistics in ascending order of their values. The higher the proportion of the t-statistics from the tension peak test sequence or the anti-confrontation test sequence, the greater the probability that the subject is lying; among which, This represents the total number of all t-statistics.

10. A cognitive load-based anti-lie detection system, employing the cognitive load-based anti-lie detection method as described in any one of claims 1 to 9, characterized in that, The system includes: The subject testing module is used to conduct cognitive load-based opposition tests on subjects and obtain subject responses during the testing process; wherein, the stimulus paradigm of the opposition test includes a tension peak test sequence, a cognitive load practice sequence, and an opposition test sequence, and each sequence contains several different types of stimuli as test questions; the subject responses include behavioral responses and physiological signals. The feature acquisition module is used to filter all subject responses that meet the preset response conditions in the tension peak test sequence, cognitive load exercise sequence and opposition test sequence as target responses, and extract the feature vector of the target response; wherein, the preset response conditions include correct behavioral response category and / or behavioral response duration less than the response duration threshold; The lie detection module is used to perform lie detection based on the feature vector of the target's response.

Images