Lie detection data analysis system and method thereof

Abstract

The invention relates to the technical field of data analysis, and particularly discloses a lie detection data analysis system and a lie detection data analysis method, and the lie detection data analysis system comprises a four-dimensional sensing data acquisition module, a sensing parameter confidence analysis module, a sensing weight calculation analysis module, a parameter fusion calculation analysis module, a sensing confidence optimization analysis module and a database. By monitoring power supply voltage and current fluctuation, analyzing the feature matching degree of data in a sliding window and historical data and evaluating and optimizing the confidence coefficient in real time, the confidence coefficient can be adjusted in time and stable operation of the system can be ensured in a complex and changeable environment scene when the sensor performance is influenced by the dry environment; if the power supply is unstable or the data feature matching degree is low, the confidence coefficient is correspondingly reduced, so that the system can more accurately evaluate the reliability of the sensor data in different environments, the adaptability and the stability of the system are improved, and the accuracy of lie detection analysis is further guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and specifically refers to a lie detection data analysis system and method thereof. Background Art

[0002] In the field of lie detection technology, traditional lie detection methods often rely on a single detection means, such as common lie detectors based on changes in physiological indicators such as blood pressure and heart rate. These methods only detect the tested person from limited dimensions, and the obtained data information is relatively single, making it difficult to comprehensively and accurately reflect the true psychological state of the tested person. With the continuous development of technology, multi-sensor fusion technology has gradually been applied to multiple fields, bringing new ideas and possibilities for the innovation of lie detection technology.

[0003] The data acquisition dimension of the existing lie detection system is limited, and it is impossible to comprehensively analyze the state of the tested person from multiple physiological perspectives. A single sensor is easily affected by external factors, resulting in inaccurate detection results. For example, factors such as environmental temperature and the physical condition of the tested person may affect the lie detection results based on blood pressure and heart rate. In addition, during the data processing process, the reliability assessment of sensor data is not scientific enough. The existing methods often lack comprehensive consideration of the hardware state of the sensor and the characteristics of historical data, making it difficult to accurately judge the credibility of the data, so that unreliable data may participate in subsequent analysis, affecting the accuracy of lie detection results. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a lie detection data analysis system and method thereof to solve the above-mentioned technical defects.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0006] In the first aspect, the present invention provides a lie detection data analysis system, including:

[0007] A four-dimensional sensing data acquisition module, configured to build a four-dimensional sensing matrix for collecting lie detection data parameters of a target test person, and obtain four-dimensional sensing parameters of the target test person corresponding to each test period. The four-dimensional sensing matrix is integrated by a millimeter-wave radar, a hyperspectral camera, a capacitive array sensor, and a bone conduction microphone according to a certain spatial layout;

[0008] A sensing parameter confidence analysis module, connected to the four-dimensional sensing data acquisition module, configured to calculate and analyze each effective measurement value of each sensor in the four-dimensional sensing matrix during each test period, and obtain the confidence of each sensor during each test period;

[0009] A sensing weight calculation and analysis module, connected to the sensing parameter confidence analysis module, performs comprehensive calculation and analysis based on the noise variance and confidence level of each sensor during each test period to obtain the weight of each sensor during each test period;

[0010] A parameter fusion calculation and analysis module, connected to the sensing weight calculation and analysis module, is used to calculate the final weight of each sensor during each test period and the normalized measurement data of each sensor using a weighted fusion algorithm to obtain the lie detection analysis coefficient of the target test person corresponding to each test period, and analyze the lie detection result determination of the target test person corresponding to each test period based on the lie detection analysis coefficient;

[0011] A sensing confidence optimization analysis module, connected to the sensing parameter confidence analysis module, performs evaluation and optimization based on the confidence level of each sensor during each test period, and performs evaluation and feedback on the lie detection analysis coefficient of the target test person corresponding to each test period;

[0012] A database, respectively connected to the four-dimensional sensing data acquisition module and the sensing confidence optimization analysis module, is used to store the four-dimensional sensing parameters of the target test person corresponding to each test period.

[0013] Further, the four-dimensional sensing parameters of the target test person include laryngeal micro-vibrations, facial blood flow changes, skin conductivity, and fundamental vocal cord frequency.

[0014] Further, the method for the sensing parameter confidence analysis module to calculate the confidence level is: collect no less than 35 effective measurement values of each sensor during each test period, calculate the mean and standard deviation using the mathematical calculation formulas of the mean and standard deviation, select a confidence level of 95%, look up the standard normal distribution table to determine the relevant value, divide the standard deviation by the square root of the number of effective measurement values to obtain the standard error, and calculate the confidence level of each sensor during each test period according to the formula.

[0015] Further, the method for the sensing weight calculation and analysis module to obtain the weight is: calculate the mean of the squared deviation between the measurement data of each sensor and the mean during each test period as the noise variance; obtain the covariance between the measurement data of two sensors through the covariance calculation formula, and then calculate the Pearson correlation coefficient. After summing up the Pearson correlation coefficients, divide by the total number of sensors to obtain the mutual support degree; combine the noise variance and the confidence level to obtain the initial weight, and then combine the mutual support degree to obtain the final weight.

[0016] Further, the parameter fusion calculation and analysis module calculates the lie detection analysis coefficient in the following way: Obtain the lie detection analysis coefficients of N historical target testers corresponding to each test period, calculate their mean and standard deviation, and set them as the normal range of the lie detection analysis coefficients of the target tester corresponding to each test period; Compare the lie detection analysis coefficients of the current target tester corresponding to each test period with the normal range to determine whether the target tester is lying.

[0017] Further, the sensing confidence optimization and analysis module evaluates and optimizes the confidence level in the following way: Use a high-precision power supply monitoring module to monitor the power supply voltage and current fluctuations of the sensor. When the voltage fluctuation exceeds ±5% of the rated voltage or the current fluctuation exceeds ±10% of the rated current, trigger a confidence level degradation, and obtain the dynamically adjusted basic confidence level by decaying 0.05 each time; Fix the sliding window width at 30 seconds, set the window sliding interval according to the sensor data update frequency, calculate the feature matching degree. When the feature matching degree is lower than 0.6, trigger a confidence level degradation, and obtain the confidence level verified by historical data by decaying 0.1 each time.

[0018] Further, set weights for the dynamically adjusted basic confidence level and the confidence level verified by historical data, perform a product calculation and then sum to obtain the final confidence level, and determine whether the confidence level is reasonable by comparing it with the allowable deviation of the confidence level.

[0019] In a second aspect, the present invention provides a method for analyzing lie detection data, including the following steps:

[0020] Use the four-dimensional sensing data acquisition module to build a four-dimensional sensing matrix and collect the four-dimensional sensing parameters of the target tester corresponding to each test period;

[0021] Through the sensing parameter confidence analysis module, calculate and analyze the effective measurement values of each sensor in each test period to obtain the confidence level;

[0022] The sensing weight calculation and analysis module comprehensively calculates and analyzes based on the noise variance and the confidence level to obtain the weight of each sensor in each test period;

[0023] Use the parameter fusion calculation and analysis module to perform weighted fusion calculation on the final weight and the normalized measurement data to obtain the lie detection analysis coefficient and judge the lie detection result;

[0024] With the help of the sensing confidence optimization and analysis module, evaluate and optimize the confidence level and evaluate and feedback the lie detection analysis coefficient;

[0025] Use the database to store the four-dimensional sensing parameters of the target tester corresponding to each test period.

[0026] Further, in the step of collecting the four-dimensional sensing parameters of the target tester corresponding to each test period, it specifically includes using a millimeter-wave radar to obtain laryngeal micro-vibration data, a hyperspectral camera to obtain facial blood flow change data, a capacitive array sensor to obtain skin conductivity data, and a bone conduction microphone to obtain the fundamental frequency of vocal cord vibration data.

[0027] Further, in the step of evaluating and optimizing the confidence level, it includes operations such as monitoring the power supply voltage and current fluctuations to adjust the basic confidence level, calculating the feature matching degree to adjust the confidence level verified by historical data, calculating the final confidence level, and comparing it with the allowable deviation of the confidence level to determine the rationality of the confidence level.

[0028] Advantages of the present invention:

[0029] 1. The system in this application utilizes a four-dimensional sensing matrix constructed by a millimeter-wave radar, a hyperspectral camera, a capacitive array sensor, and a bone conduction microphone to synchronously collect multi-dimensional data such as laryngeal micro-vibration, facial blood flow change, skin conductivity, and the fundamental frequency of vocal cord vibration. These data reflect the state of the tested person from different physiological perspectives, complement each other, and the multi-dimensional data collection makes the description of the state of the tested person more accurate and comprehensive, avoiding the limitations of single-sensor data, and providing a rich and reliable data basis for subsequent accurate lie detection analysis.

[0030] 2. The sensing parameter confidence analysis module in this application, based on the central limit theorem, collects a large number of effective measurement values to calculate the mean, standard deviation, and standard error, and then obtains the confidence level. This process scientifically evaluates the reliability of the data of each sensor during each test period. Through the calculation of the confidence level, the fluctuation range and credibility of the data can be judged. For the sensor data with large measurement value fluctuations and high standard deviation, its confidence level is relatively low, and it will be reasonably considered in subsequent analysis, avoiding the interference of unreliable data on the lie detection result, and ensuring that the data used for analysis has high reliability and stability.

[0031] 3. The proposed sensing confidence optimization analysis module in this application dynamically adjusts the sensor confidence level from two aspects: power supply stability and historical data feature matching. By monitoring the power supply voltage and current fluctuations, and analyzing the feature matching degree between the data in the sliding window and the historical data, it evaluates and optimizes the confidence level in real time. In a complex and changeable environment, such as when the environment is dry and affects the performance of the sensor, it can adjust the confidence level in time to ensure the stable operation of the system. If the power supply is unstable or the data feature matching degree is low, the confidence level is correspondingly reduced, enabling the system to more accurately evaluate the reliability of the sensor data in different environments, improving the adaptability and stability of the system, and further ensuring the accuracy of the lie detection analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present invention will be further described below with reference to the accompanying drawings.

[0033] Figure 1 It is a connection schematic diagram of modules of a lie detection data analysis system according to an embodiment of the present invention;

[0034] Figure 2 It is a step flow chart of a lie detection data analysis system according to an embodiment of the present invention;

[0035] Figure 3 It is an analysis flow chart of a sensing confidence optimization analysis module according to an embodiment of the present invention. Specific embodiments

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of the present invention.

[0037] Please refer to Figure 1 As shown, the present invention provides a lie detection data analysis system, and the specific module distribution is as follows: a four-dimensional sensing data acquisition module, a sensing parameter confidence analysis module, a sensing weight calculation analysis module, a parameter fusion calculation analysis module, a sensing confidence optimization analysis module, and a database. Among them, the connection method between the modules is: the four-dimensional sensing data acquisition module is connected to the sensing parameter confidence analysis module, the sensing parameter confidence analysis module is respectively connected to the sensing weight calculation analysis module and the sensing confidence optimization analysis module, the sensing weight calculation analysis module is connected to the parameter fusion calculation analysis module, and the database is respectively connected to the four-dimensional sensing data acquisition module and the sensing confidence optimization analysis module.

[0038] The four-dimensional sensing data acquisition module builds a four-dimensional sensing matrix for collecting lie detection data parameters of a target test person, and obtains four-dimensional sensing parameters of the target test person corresponding to each test period, including laryngeal micro-vibration, facial blood flow change, skin conductivity, and fundamental frequency of vocal cord vibration.

[0039] It should be further noted that the construction of the four-dimensional sensing matrix for collecting lie detection data parameters of the target tester specifically includes a millimeter-wave radar, a hyperspectral camera, a capacitive array sensor, and a bone conduction microphone. The above four sensors are integrated according to a certain spatial layout to form a compact sensing module. Among them, the antenna of the millimeter-wave radar faces the throat area of the target tester; the lens of the hyperspectral camera is aligned with the face of the target tester; the contact surface of the capacitive array sensor is attached to the skin of the target tester; the sound pickup component of the bone conduction microphone is in close contact with the appropriate part of the skull of the target tester. Each sensor is connected to the data processing unit through a data interface, and the data processing unit synchronously collects and stores the measurement data from different sensors, thereby constructing a four-dimensional sensing matrix containing measurement data of multiple sensors.

[0040] The four-dimensional sensing matrix constructed by the present invention can collect multi-dimensional information simultaneously, providing a rich data source for subsequent data analysis. Different types of sensors complement each other, improving the accuracy and comprehensiveness of the description of the state of the monitored object.

[0041] It should be further noted that the flowchart of the system implementation steps is as Figure 2 shown.

[0042] It should be further noted that the specific acquisition methods of the throat micro-vibration, facial blood flow change image, skin conductivity, and fundamental frequency of vocal cord vibration of the target tester are as follows:

[0043] Use a millimeter-wave radar at 77 GHz to monitor the micro-movement of the throat muscles of the target tester using the reflection principle of electromagnetic waves to obtain the throat micro-vibration data of the target tester; among them, the emitted millimeter-wave signal is reflected after encountering the throat muscles of the target tester, and the reflected signal is received by the radar receiving antenna. By analyzing and processing parameters such as the frequency and phase of the transmitted signal and the received signal, micro-vibrations as low as 0.1 mm can be detected.

[0044] Use a hyperspectral camera to cover the spectral range of 400 - 1000 nm. The light of the target face is focused on the image sensor through an optical lens. The spectroscopic system inside the camera separates the incident light according to the wavelength, so that each pixel can obtain the light intensity information within a specific wavelength range. Use the hyperspectral camera to continuously shoot at a frame rate of 120 frames per second to capture the dynamic changes of the facial blood flow. The data acquisition module records the light intensity data of each pixel at different times in real time to form time-series target facial blood flow change image data; among them, the image sensor consists of multiple pixels, and each pixel can respond to light of different wavelengths.

[0045] The capacitive array sensor is used to contact the skin of the target tester, and a capacitance is formed between the skin of the target tester and the sensing unit. The skin conductivity data of the target tester is obtained by measuring the change in capacitance through the capacitive array sensor. Specifically, the measurement circuit uses an AC excitation method to apply an AC signal of a certain frequency to the capacitive sensing unit, and then measures the voltage change across the capacitance through the detection circuit. After processing steps such as signal amplification, filtering, and analog-to-digital conversion, the skin conductivity measurement value with an accuracy of 0.01 μS is finally obtained.

[0046] A bone conduction microphone is used to collect the fundamental frequency of the vocal cord vibration of the target tester by using the bone conduction principle. Among them, the sound pickup component of the microphone is in contact with the human skull. When the vocal cords vibrate, the vibration is transmitted to the sound pickup component of the microphone through the skull, and the sensitive element inside the sound pickup component converts the mechanical vibration into an electrical signal. After being amplified, filtered, etc., the electrical signal is sampled by the signal acquisition module. The sampling frequency range of the bone conduction microphone is set to 20 - 8000 Hz, and it can effectively collect the fundamental frequency signal generated by the vocal cord vibration.

[0047] It should be noted that the system in this application uses a four-dimensional sensing matrix constructed by a millimeter-wave radar, a hyperspectral camera, a capacitive array sensor, and a bone conduction microphone to synchronously collect multi-dimensional data such as laryngeal micro-vibrations, facial blood flow changes, skin conductivity, and fundamental frequency of vocal cord vibration. These data reflect the state of the tested person from different physiological perspectives, complement each other, and the multi-dimensional data collection makes the description of the state of the tested person more accurate and comprehensive, avoiding the limitations of single-sensor data, and providing a rich and reliable data basis for subsequent accurate lie detection analysis.

[0048] The sensing parameter confidence analysis module calculates and analyzes each effective measurement value of each sensor in the four-dimensional sensing matrix during each test period to obtain the confidence level Ci of each sensor during each test period.

[0049] It should be noted that i = 1, 2,..., m, where i represents the number of each sensor, and m represents the total number of each sensor number.

[0050] As a preferred feasible example, the specific method for calculating and analyzing each effective measurement value of each sensor in the four-dimensional sensing matrix is as follows:

[0051] Collect n effective measurement values of each sensor during each test period, and n ≥ 35 meets the requirements of the central limit theorem. Using the mathematical calculation formulas of the mean and standard deviation, calculate and obtain the mean μi and standard deviation σi of each sensor during each test period respectively. Among them, the mean μi of each sensor during each test period reflects the central tendency of the measurement data, and the standard deviation σi of each sensor during each test period characterizes the degree of dispersion of the measurement data.

[0052] Select the confidence level of the four-dimensional sensing matrix test application scenario to be 95%, corresponding to the significance level α = 0.05. At the same time, check the standard normal distribution table to determine Z α2 = 1.96;

[0053] Divide the standard deviation σi of each sensor in each test period by the square root of the number of effective measurement values of each sensor in each test period, and the calculated result is the standard error WEi of each sensor in each test period. The standard error WEi of each sensor in each test period is used to reflect the fluctuation range of the mean value of each sensor in each test period. Finally, according to the formula Ci = μ ± Z α2 ×WEi to calculate the confidence level Ci of each sensor in each test period.

[0054] It should be noted that the confidence level refers to the estimation interval of the population parameter constructed by the sample statistic. The calculation formula of the confidence level Ci is to determine an interval range containing the population mean at a given confidence level. μ - Z α2 ×WEi calculates the lower limit of the confidence level, and μ + Z α2 ×WEi calculates the upper limit of the confidence level.

[0055] Specifically, the sensing parameter confidence analysis module in this application, based on the central limit theorem, collects a large number of effective measurement values to calculate the mean value, standard deviation, and standard error, and then obtains the confidence level. This process scientifically evaluates the reliability of the data of each sensor in each test period. Through the calculation of the confidence level, the fluctuation range and credibility of the data can be judged. For the sensor data with large fluctuations and high standard deviations in the measurement values, its confidence level is relatively low and will be reasonably considered in subsequent analyses, avoiding the interference of unreliable data on the lie detection results and ensuring that the data used for analysis has high reliability and stability.

[0056] The sensing data weight analysis module comprehensively calculates and analyzes based on the noise variance and confidence level of each sensor in each test period to obtain the weight of each sensor in each test period;

[0057] As a preferred feasible example, the specific method for comprehensively calculating and analyzing the noise variance and confidence level of each sensor in each test period is as follows:

[0058] Calculate the mean of the square of the deviation between the measurement data of each sensor in each test period and the mean value, and use the calculated result as the noise variance θi of each sensor;

[0059] Substitute the standard deviations of the measurement data of the two sensors into the covariance calculation formula for calculation to obtain the covariance between the measurement data of the two sensors. Divide the covariance between the measurement data of the two sensors by the product of the standard deviations of the measurement data of the two sensors to obtain the Pearson correlation coefficient between the measurement data of the two sensors. Sum the Pearson correlation coefficients between the measurement data of each sensor and then divide by the total number of sensors to obtain the mutual support degree Si between the measurement data of each sensor;

[0060] Comprehensively calculate and analyze the noise variance θi of each sensor and the confidence level Ci of each sensor in each test period to obtain the initial weight Wi of each sensor in each test period (1) ;

[0061] Then comprehensively calculate and analyze the initial weight Wi of each sensor in each test period (1) and the mutual support degree Si between the measurement data of each sensor to obtain the final weight Wi of each sensor in each test period (final) ;

[0062] As a preferred feasibility example, the comprehensive calculation and analysis method for the noise variance θi of each sensor and the confidence level Ci of each sensor in each test period is as follows:

[0063] According to the formula Calculate the initial weight Wi of each sensor in each test period (1) ;

[0064] It should be noted that in the calculation formula of the initial weight Wi of each sensor in each test period, Ci / θi reflects the reliability of each sensor. For a sensor with a high confidence level and a small noise variance, this ratio is large, indicating that its data is more reliable; (1) By summing the ratios of the confidence levels and noise variances of all sensors, a normalization effect is achieved; finally, the initial weight Wi of each sensor in each test period is calculated through the combination of the two , and according to the reliability of each sensor, relative weight allocation is performed among all sensors. The higher the reliability of the sensor, the larger its weight Wi (1) , and the greater the proportion it occupies in data fusion. (1)

[0065] The comprehensive calculation and analysis method for the initial weight Wi of each sensor in each test period (1) and the mutual support degree Si between the measurement data of each sensor is as follows:

[0066] According to the formula Calculate the final weight Wi of each sensor in each test period (final) ;​

[0067] It should be noted that the initial weight Wi of each sensor in each test period (final) In the calculation formula of Wi (1) Si takes into account both the reliability of each sensor itself and the relationship between each sensor data and other sensor data; Play a normalization role, so that the final weight Wi (final) The value of is within a reasonable range; finally, the final weight Wi of each sensor in each test period is calculated by combining the two. (final) On the basis of the initial weight distribution, further adjustments are made according to the mutual support between the sensor data, so that the final fusion result not only depends on the reliability of a single sensor, but also comprehensively considers the synergy between the sensor data, thereby obtaining a more accurate and reliable data fusion result.

[0068] Specifically, the sensor data weight analysis module in the present application comprehensively considers the noise variance, confidence and mutual support between data of the sensor to determine the weight. Sensors with small noise variance, high confidence and high mutual support with other sensor data have greater weights in data fusion. This calculation method fully considers the individual characteristics of the sensor and the synergistic relationship between data, making the final fusion result more accurate and reliable. Within a certain test period, if the measurement data of a sensor is highly correlated with the data of multiple other sensors, it means that it can well reflect the overall state of the test subject, and is given a higher weight in the fusion calculation, thereby optimizing the data fusion effect and improving the accuracy of polygraph analysis.

[0069] The parameter fusion calculation and analysis module calculates the final weight Wi of each sensor in each test period. (final) The normalized measurement data Xi of each sensor is calculated by a weighted fusion algorithm to obtain the polygraph analysis coefficient P of the target test person corresponding to each test period, and the polygraph result of the target test person corresponding to each test period is analyzed according to the polygraph analysis coefficient P of the target test person corresponding to each test period;

[0070] As a preferred feasible example, the final weight Wi of each sensor in each test period is (final) The weighted fusion calculation method of each sensor measurement data Xi after normalization is as follows:

[0071] According to the formula Calculate the polygraph analysis coefficient P of the target tester corresponding to each test period;

[0072] By obtaining the lie detection analysis coefficients Pl of N historical target testers corresponding to each test period, and denoting their set as Pl = {pl1, pl2, …, plN}, using the mathematical calculation formulas of the mean and standard deviation, calculate the mean μl and standard deviation σl of the lie detection analysis coefficients Pl of the historical target testers corresponding to each test period, and set [μl - 2σl, μl + 2σl] as the normal range of the lie detection analysis coefficients of the target tester corresponding to each test period;

[0073] Compare and analyze the lie detection analysis coefficient P of the target tester corresponding to each test period with the set normal range of the lie detection analysis coefficients of the target tester corresponding to each test period. If the lie detection analysis coefficient P of the target tester corresponding to each test period falls within the set normal range of the lie detection analysis coefficients of the target tester corresponding to each test period, it indicates that the target tester is not lying. Otherwise, it indicates that the target tester has the possibility of lying.

[0074] Specifically, the parameter fusion calculation and analysis module calculates the lie detection analysis coefficient using a weighted fusion algorithm, and determines the lie detection result by setting the normal range through comparison with historical data. This method is based on the statistical analysis of a large amount of historical data, making the determination of the lie detection result more scientific and reasonable. Comparing the lie detection analysis coefficient of the tested person with the normal range can intuitively judge whether they are lying. When the lie detection analysis coefficient of the tested person exceeds the normal range, it indicates that their physiological state has abnormal changes and there is a possibility of lying, improving the accuracy and reliability of lie detection.

[0075] The sensing confidence optimization analysis module evaluates and optimizes based on the confidence level Ci of each sensor during each test period, and provides an evaluation feedback on the lie detection analysis coefficient P of the target tester corresponding to each test period.

[0076] It should be further noted that the flowchart of the confidence level evaluation and optimization steps is as Figure 3 shown.

[0077] As a preferred feasible example, by obtaining the confidence level Ci of each sensor during each test period, using a high-precision power supply monitoring module, real-time monitor the fluctuations of the sensor supply voltage and current. When the voltage fluctuation exceeds ±5% of the rated voltage or the current fluctuation exceeds ±10% of the rated current, it is determined that the power supply is unstable. Each time an unstable situation occurs, trigger a confidence level downgrade, and each time the confidence level Ci decays by 0.05 to obtain the dynamically adjusted basic confidence level Cb;

[0078] Fix the width of the sliding window at 30 seconds. For sensors with a high data update frequency, set the window sliding interval to 1 second; for sensors with a low data update frequency, set the window sliding interval to 5 seconds. According to the formula Calculate the feature matching degree Ch of each sensor within each test period; where the feature matching degree Ch of each sensor within each test period is used to measure the matching degree between the data features within the sliding window and the historical data features. The closer the value is to 1, the higher the matching degree. g represents the sample number of the sensor measurement data, g = 1, 2, …, k, and k represents the total number of sample numbers of the sensor measurement data within the sliding window, that is, when analyzing 30 - second data using the sliding window, the total number of data points included. e represents the natural constant, xg represents the value of the k - th sample data point within the sliding window, that is, corresponding to the k - th sensor measurement value obtained in the measurement order. μ represents the mean calculated for the historical data within the sliding window, and σ represents the standard deviation calculated for the historical data within the sliding window. The larger the value of σ, the more dispersed the historical data; the smaller the value of σ, the more concentrated the historical data.

[0079] When the feature matching degree Ch of each sensor within each test period is lower than 0.6, trigger confidence degradation. Each time the confidence Ci decays by 0.1 to obtain the confidence Cl obtained from historical data verification.

[0080] It should be noted that in the calculation formula of the feature matching degree Ch of each sensor within each test period, (xg - μ) 2 measures the square of the deviation between the g - th sample data point xg and the mean μ. The larger the square of the deviation, the farther the data point deviates from the mean. By normalizing the square of the deviation and considering the degree of data dispersion, in the normal distribution, the closer the data point is to the mean, the higher the probability of occurrence; conversely, the farther away from the mean, the lower the probability of occurrence. As the exponent of the natural constant e, when the deviation between the data point xg and the mean μ is larger, that is the larger, the smaller the value of Conversely, when the data point is close to the mean, the value of is close to 1, that is, it means that the data points with small deviations from the mean contribute more to the result, which conforms to the expectation that the matching degree is higher when the data features are stable and close to the historical mean. Finally, by summing the exponential function values corresponding to the k data points within the sliding window, the comprehensive matching situation index of these k data points is obtained, and then the average value is calculated, that is, the feature matching degree Ch of each sensor within each test period is obtained. When the feature matching degree Ch is closer to 1, it indicates that the data features within the sliding window are more matched with the historical data features; when the feature matching degree Ch is smaller, it indicates that the data feature differences are larger.

[0081] After calculating the product by setting corresponding weights for the dynamically adjusted basic confidence level and the confidence level verified by historical data, the sum is obtained to get the final confidence level Cfinal of each sensor in each test period; the confidence level Ci of each sensor in each test period is multiplied by floating up and down by 5% to obtain the allowable deviation of the confidence level of each sensor in each test period The final confidence level Cfinal of each sensor in each test period and the allowable deviation of the confidence level of each sensor in each test period Carry out comparative analysis to obtain the number q of the final confidence level Cfinal of each sensor in each test period falling within the allowable deviation of the confidence level According to the calculation result of q / m, determine whether the confidence level Ci of each sensor in each test period is reasonable. If the calculation result of q / m is less than 0.65, it means that the setting of the confidence level Ci of each sensor in each test period is unreasonable. By calculating the average value of the confidence level Ci and the final confidence level Cfinal of each sensor in each test period, it is used as the optimized confidence level Cy of each sensor in each test period. Otherwise, it means that the setting of the confidence level Ci of each sensor in each test period is reasonable

[0082] Specifically, in the environmental monitoring scenario of the target tester, for the sensor affected by large environmental dryness, the weight of the dynamically adjusted basic confidence level is set to 0.7, and the weight of the confidence level verified by historical data is set to 0.3

[0083] It should be noted that the sensing confidence optimization analysis module proposed in this application dynamically adjusts the sensor confidence level from two aspects: power supply stability and historical data feature matching. By monitoring the fluctuations of the power supply voltage and current, and analyzing the feature matching degree between the data in the sliding window and the historical data, the confidence level is evaluated and optimized in real time. In a scenario where the environment is complex and changeable, such as when the environmental dryness affects the performance of the sensor, the confidence level can be adjusted in time to ensure the stable operation of the system. If the power supply is unstable or the data feature matching degree is low, the confidence level is correspondingly reduced, so that the system can more accurately evaluate the reliability of the sensor data in different environments, improving the adaptability and stability of the system, and further ensuring the accuracy of the lie detection analysis

[0084] The database stores the four-dimensional sensing parameters of the target tester corresponding to each test period

[0085] In addition, those skilled in the art will understand that various aspects of the present invention can be illustrated and described by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Accordingly, various aspects of the present invention can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can all be referred to as "data blocks", "modules", "engines", "units", "components", or "systems". In addition, various aspects of the present invention may be embodied as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0086] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0087] The above is a description of the present invention and should not be construed as a limitation thereof. Although several exemplary embodiments of the present invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is a description of the present invention and should not be considered limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A polygraph data analysis system, characterized in that: include: A four-dimensional sensor data acquisition module is used to build a four-dimensional sensor matrix for collecting polygraph data parameters of the target test person, and obtain the four-dimensional sensor parameters of the target test person corresponding to each test period. The four-dimensional sensor matrix is ​​integrated by millimeter wave radar, hyperspectral camera, capacitive array sensor and bone conduction microphone according to a certain spatial layout; A sensor parameter confidence analysis module is connected to the four-dimensional sensor data acquisition module and is used to calculate and analyze each effective measurement value of each sensor in the four-dimensional sensor matrix in each test period to obtain the confidence of each sensor in each test period; A sensor weight calculation and analysis module is connected to the sensor parameter confidence analysis module, and performs comprehensive calculation and analysis based on the noise variance and confidence of each sensor in each test period to obtain the weight of each sensor in each test period; A parameter fusion calculation and analysis module is connected to the sensor weight calculation and analysis module, and is used to calculate the final weight of each sensor in each test period and the normalized measurement data of each sensor using a weighted fusion algorithm to obtain the polygraph analysis coefficient of the target test person corresponding to each test period, and analyze the polygraph result judgment of the target test person corresponding to each test period according to the polygraph analysis coefficient; The sensor confidence optimization analysis module is connected to the sensor parameter confidence analysis module, and performs evaluation and optimization based on the confidence of each sensor in each test period, and evaluates and provides feedback on the polygraph analysis coefficient of the target test person corresponding to each test period.

2. A polygraph data analysis system according to claim 1, characterized in that: The four-dimensional sensing parameters of the target test person include laryngeal micro-tremors, facial blood flow changes, skin conductivity and vocal cord vibration fundamental frequency.

3. A polygraph data analysis system according to claim 1, characterized in that: The sensing parameter confidence analysis module calculates the confidence in the following manner: collect no less than 35 valid measurement values ​​of each sensor in each test period, calculate the mean and standard deviation using mathematical calculation formulas for the mean and standard deviation, select a confidence level of 95%, determine the relevant value by looking up the standard normal distribution table, divide the standard deviation by the square root of the number of valid measurement values ​​to obtain the standard error, and calculate the confidence of each sensor in each test period according to the formula.

4. A polygraph data analysis system according to claim 1, characterized in that: The sensor weight calculation and analysis module obtains the weight in the following manner: the mean of the square of the deviation between the measured data of each sensor and the mean in each test period is calculated as the noise variance; the covariance between the measured data of two sensors is obtained by the covariance calculation formula, and then the Pearson correlation coefficient is calculated, and the sum of the Pearson correlation coefficients is divided by the total number of sensors to obtain the mutual support; The initial weight is obtained by combining the noise variance and confidence, and then the final weight is obtained by combining the mutual support.

5. A polygraph data analysis system according to claim 1, characterized in that: The parameter fusion calculation and analysis module calculates the polygraph analysis coefficient in the following manner: obtain the polygraph analysis coefficients of N historical target testers corresponding to each test period, calculate their mean and standard deviation, and set them as the normal range of the polygraph analysis coefficients of the target testers corresponding to each test period; compare the polygraph analysis coefficients of the current target tester corresponding to each test period with the normal range to determine whether the target tester is lying.

6. A polygraph data analysis system according to claim 1, characterized in that: The sensor confidence optimization analysis module evaluates the optimization confidence in the following manner: a high-precision power supply monitoring module is used to monitor the sensor power supply voltage and current fluctuations. When the voltage fluctuation exceeds ±5% of the rated voltage or the current fluctuation exceeds ±10% of the rated current, the confidence degradation is triggered, and the basic confidence is dynamically adjusted by decaying by 0.05 each time; the fixed sliding window width is 30 seconds, the window sliding interval is set according to the sensor data update frequency, and the feature matching degree is calculated. When the feature matching degree is lower than 0.6, the confidence degradation is triggered, and the confidence is decayed by 0.1 each time to obtain the confidence verified by historical data.

7. A polygraph data analysis system according to claim 6, characterized in that: The final confidence is obtained by multiplying the dynamically adjusted basic confidence and the historical data verification confidence setting weights, and comparing the confidence with the confidence allowable deviation to determine whether the confidence is reasonable.

8. A polygraph data analysis method, characterized in that: The polygraph data analysis system according to any one of claims 1 to 7 is applied, comprising the following steps: Use the four-dimensional sensor data acquisition module to build a four-dimensional sensor matrix and collect the four-dimensional sensor parameters of the target test personnel corresponding to each test period; The effective measurement value of each sensor in each test period is calculated and analyzed through the sensor parameter confidence analysis module to obtain the confidence level; The sensor weight calculation and analysis module performs comprehensive calculation and analysis based on noise variance and confidence to obtain the weight of each sensor in each test period; Use the parameter fusion calculation and analysis module to perform weighted fusion calculation on the final weight and normalized measurement data to obtain the polygraph analysis coefficient and determine the polygraph result; The confidence level is evaluated and optimized with the help of the sensor confidence optimization analysis module, and the polygraph analysis coefficient is evaluated and fed back; The database is used to store the four-dimensional sensing parameters of the target test person corresponding to each test period.

9. A polygraph data analysis method according to claim 8, characterized in that: The step of collecting the four-dimensional sensing parameters of the target test person corresponding to each test period specifically includes using a millimeter-wave radar to obtain laryngeal micro-vibration data, a hyperspectral camera to obtain facial blood flow change data, a capacitive array sensor to obtain skin conductivity data, and a bone conduction microphone to obtain vocal cord vibration fundamental frequency data.

10. A polygraph data analysis method according to claim 8, characterized in that: The steps of evaluating and optimizing the confidence include monitoring the supply voltage and current fluctuations to adjust the basic confidence, calculating the feature matching degree to adjust the historical data to verify the confidence, and calculating the final confidence and comparing it with the confidence allowable deviation to determine the rationality of the confidence.

Images