Lie detection system based on AC excitation multi-frequency skin electric signals and method thereof

Abstract

The invention discloses a lie detection system and method based on AC excitation multi-frequency skin electric signals, and belongs to the technical field of psychological and physiological detection and signal processing. According to the system, a multi-frequency AC excitation signal is applied to human skin, skin complex impedance is measured by adopting a four-electrode method, and the problems of electrode polarization and baseline drift of direct current excitation are solved. A high-dimensional feature vector is formed by extracting the amplitude, phase and differential ratio of multi-frequency impedance and Cole-Cole relaxation model parameters, and an individualized baseline is established. Feature changes are analyzed through a machine learning algorithm, and lie judgment is achieved. The method has the advantages of strong anti-interference capability, abundant information dimensions, high measurement accuracy and good safety.

Description

Technical Field

[0001] This invention relates to the field of psychophysiological detection and signal processing technology, and in particular to a lie detection system and its implementation method based on AC excitation and using multi-frequency electrodermal signals for measurement. Background Technology

[0002] Lie detection, also known as psychophysiological testing, is a technique that assesses an individual's psychological state (such as tension or deception) by recording changes in their physiological responses to specific stimuli. Currently, mainstream polygraphs primarily measure three physiological indicators: pulse / heart rate, respiratory rate, and Galvanic Skin Response (GSR).

[0003] Among these, electrodermal activity (EDA) is one of the most sensitive and rapid indicators reflecting the excitability of the sympathetic nervous system. Traditional EDA measurements mostly employ direct current (DC) excitation, measuring skin conductance level (SCL) and skin conductance response (SCR) by applying a constant low voltage (or measuring voltage changes under constant current). However, the DC excitation method has some inherent limitations:

[0004] 1. Electrode polarization effect: Long-term DC current can cause electrochemical polarization at the electrode-skin interface, generating polarization voltage, which causes the measurement signal to drift, the signal-to-noise ratio to decrease, and affects the stability of long-term measurement.

[0005] 2. Limited information dimension: The DC method mainly reflects the overall conductivity change after the skin's sweat gland ducts are filled with fluid, and cannot obtain richer frequency domain information in the skin impedance.

[0006] 3. Poor comfort and safety: DC current may cause skin irritation or even burns, affecting the subject's natural state.

[0007] To address the limitations of the DC method, some studies have attempted to use AC excitation. AC excitation, by applying an alternating current at a specific frequency, can overcome electrode polarization and measure the complex impedance characteristics of the skin. However, current applications of AC excitation in lie detection are still immature, typically using only a single frequency and failing to fully utilize the rich physiological and psychological information contained in skin impedance at different frequencies. Different frequencies of current penetrate the skin to varying depths, and the physiological mechanisms of the response also differ (such as extracellular fluid and cell membrane properties). Therefore, measurements using a single frequency may miss crucial information.

[0008] The present invention aims to overcome the above-mentioned shortcomings of the prior art and proposes a lie detection system and method based on AC-excited multi-frequency electrodermal signals. By measuring multi-band impedance, richer and more stable feature vectors are extracted, thereby improving the accuracy and reliability of lie detection and solving problems such as DC baseline drift and security. Summary of the Invention

[0009] This application provides a lie detection system based on AC-excited multi-frequency electrodermal signals, comprising:

[0010] A hardware system is constructed based on a main controller, a digital-to-analog converter (DAC), a voltage-controlled current source (VCCS), an instrumentation amplifier, a filter, and an analog-to-digital converter (ADC). A hardware multi-frequency AC excitation signal module determines the AC excitation signal; the skin complex impedance is determined based on the multi-frequency AC excitation signal; a multi-dimensional feature vector is determined based on the skin complex impedance; an individualized multi-frequency impedance baseline is determined based on a baseline calibration module; and a lie detection result is generated based on a lie analysis and judgment module.

[0011] In some embodiments, the hardware mainly includes a main controller (such as an ARM or DSP chip), a digital-to-analog converter (DAC), a voltage-controlled current source (VCCS), an instrumentation amplifier, a filter, and an analog-to-digital converter (ADC).

[0012] In some embodiments, a set of sinusoidal AC excitation current signals with specific frequencies and low amplitudes is generated based on a hardware multi-frequency AC excitation signal generation module. The frequency group covers multiple frequency bands (e.g., low frequency 0.1Hz-1kHz, mid frequency 1kHz-100kHz, high frequency 100kHz-1MHz), and preferably includes at least three different frequency points.

[0013] In some implementation examples, the skin complex impedance is determined based on a multi-frequency AC excitation signal; a multi-band AC excitation is determined based on a set of sinusoidal AC excitation current signals of specific frequency and low amplitude; the AC excitation is used to determine the application of the excitation signal to the human skin surface through a pair of excitation electrodes; the voltage response signal generated by synchronous measurement through a pair of measuring electrodes is determined based on the AC excitation and skin response; and the voltage response signal is processed by amplification, filtering, and analog-to-digital conversion (ADC) circuitry to convert it into a high-quality digital signal.

[0014] In some implementation examples, a multi-dimensional feature vector is determined based on the complex impedance; the skin complex impedance (including impedance magnitude |Z| and phase angle θ) at each excitation frequency is determined using digital lock-in amplification or a fast Fourier transform (FFT) algorithm; time-domain and frequency-domain features are determined based on the multi-frequency impedance data to form a feature vector; the features include:

[0015] Fundamental frequency characteristics: impedance amplitude and phase at each frequency point.

[0016] Differential characteristics: impedance amplitude ratio and phase difference between different frequency points.

[0017] Relaxation features: Feature parameters (such as R∞, R0, α, τ) obtained by fitting a Cole-Cole model.

[0018] In some implementation examples, the lie judgment result is determined based on the feature vector; a classification model is determined using machine learning algorithms (such as support vector machine SVM, random forest, neural network) or pattern recognition algorithms; the currently measured feature vector is compared with a baseline pattern or norm database according to the classification model; and an index representing the probability of deception is calculated or a judgment result of "honesty" or "deception" is directly given based on the comparison result.

[0019] In some implementation examples, an individualized multi-frequency impedance baseline is determined based on a baseline calibration module; the skin impedance measured by the synchronous multi-frequency AC excitation signal is determined by asking the subject a neutral question; and the individualized multi-frequency impedance baseline is determined based on the skin impedance.

[0020] In some implementation examples, the lie detection result is determined based on the lie analysis and judgment module; a feature vector is determined based on the current question; the feature vector is compared with a multi-frequency impedance baseline; and the lie detection result is determined based on the comparison result.

[0021] The beneficial effects of this invention are as follows:

[0022] 1. Overcoming polarization effects: By adopting AC excitation, the electrode polarization problem in DC measurement is fundamentally eliminated, improving signal stability and measurement accuracy.

[0023] 2. Rich information dimensions: Multi-frequency measurement can obtain the dispersion characteristics of the skin, revealing information at different physiological levels (extracellular fluid, cell membrane), and providing far more feature quantities than single DC or single-frequency AC measurement, providing more sufficient data support for accurate judgment.

[0024] 3. Strong anti-interference capability: Through technologies such as digital phase-locked amplification, it can effectively suppress power frequency interference and environmental noise, and extract weak effective signals.

[0025] 4. High accuracy: By using multi-dimensional features for machine learning pattern recognition, false judgments caused by individual differences and random factors are reduced, significantly improving the accuracy and reliability of lie detection.

[0026] 5. Enhanced safety: The use of low-amplitude AC excitation signals completely avoids the risks of skin electrolysis, stinging, or burns that may be caused by DC excitation, significantly improving the safety of the testing process and the comfort of the test subject, and ensuring that the test is conducted under conditions that meet human safety standards. Attached Figure Description

[0027] Figure 1This is a block diagram of the overall hardware structure of the lie detection system of the present invention.

[0028] Figure 2 This is a flowchart of the lie detection method of the present invention.

[0029] Figure 3 A schematic diagram of the application of multi-frequency AC excitation signals and the acquisition of response signals (four-electrode method).

[0030] Figure 4 This is a schematic diagram of feature extraction from multi-frequency impedance data (exemplary Cole-Cole plot). Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0032] Step S1: The hardware mainly includes a main controller (such as an ARM or DSP chip), a digital-to-analog converter (DAC), a voltage-controlled current source (VCCS), an instrumentation amplifier, filters, and an analog-to-digital converter (ADC). See below for details. Figure 1 .

[0033] S11: The main controller, employing a high-performance ARM Cortex-M7 series microcontroller (such as STMicroelectronics' STM32H7 series) or digital signal processor (DSP) (such as TI's TMS320C5500 series). This controller is responsible for the coordinated control of the entire system, including generating digital excitation signals, executing algorithms, processing data, and outputting results.

[0034] S12: Signal generation link,

[0035] Digital-to-analog converter (DAC): Select a high-resolution (e.g., 16-bit), high-refresh-rate DAC (e.g., Analog Devices' AD5541A). Controlled by the main controller, it generates digital sine wave waveform data.

[0036] Voltage Controlled Current Source (VCCS): Employing a Howland current pump circuit or a VCCS design based on a precision operational amplifier, the voltage signal output from the DAC is converted into a high-precision, high-output-impedance AC constant current source signal, ensuring that the current amplitude applied to the skin is constant and unaffected by changes in skin contact impedance.

[0037] S13: Signal Acquisition and Conditioning Link

[0038] Electrode configuration: A four-electrode method (or Kelvin connection) is used, such as... Figure 3As shown. Ag / AgCl electrodes or gold-plated electrodes are used to reduce polarization effects. The two outer electrodes (E1, E2) are excitation electrodes connected to the VCCS output terminal; the two inner electrodes (M1, M2) are measurement electrodes used to detect the voltage drop on the skin surface.

[0039] Instrumentation amplifier: Select an instrumentation amplifier with high input impedance, high common-mode rejection ratio (CMRR>120dB), and low noise (such as Analog Devices' AD8421). Used to amplify the small differential voltage signal between the measuring electrodes and suppress common-mode interference (such as 50 / 60Hz power frequency interference).

[0040] Anti-aliasing filter: Use a multi-order (e.g., 4th order) low-pass Bessel filter or Butterworth filter, with the cutoff frequency set to more than twice the highest excitation frequency to prevent frequency aliasing during sampling.

[0041] Analog-to-digital converter (ADC): Select a synchronous sampling, high resolution (24-bit), high sampling rate Σ-Δ ADC (such as TI's ADS1299 series) to convert the conditioned analog voltage signal into a high-precision digital signal for processing by the main processor.

[0042] For a description of the measurement principle, see Figure 3 This diagram illustrates the principle of the four-electrode method for measuring skin impedance:

[0043] 1. The two outer E electrodes (E1 and E2) are the excitation electrodes, and a known alternating current signal (I) is applied.

[0044] 2. The two inner M electrodes (M1 and M2) are measuring electrodes, used to detect the generated voltage difference (V).

[0045] 3. Skin impedance (Z = V / I) can be calculated by measuring the ratio of voltage to current.

[0046] 4. The skin tissue structure consists of three layers: the epidermis, the dermis, and the subcutaneous tissue, each with different electrical properties.

[0047] 5. Advantages of the four-electrode method: It eliminates the influence of electrode-skin contact impedance, thus improving measurement accuracy.

[0048] 6. This method is widely used in fields such as bioimpedance measurement and skin conductance studies.

[0049] Step S2:

[0050] In some implementations, a set of sinusoidal AC excitation current signals with specific frequencies and low amplitudes is generated based on a multi-frequency AC excitation signal generation module. The frequency group covers multiple frequency bands (e.g., low frequency 0.1Hz-1kHz, mid frequency 1kHz-100kHz, high frequency 100kHz-1MHz), and preferably includes at least three different frequency points.

[0051] Step S21: The main controller generates corresponding sine wave digital samples in memory according to the preset frequency sequence (e.g., 10Hz, 100Hz, 1kHz, 10kHz, 100kHz) through direct digital frequency synthesis (DDS) or lookup table method.

[0052] Step S22: The digital sample stream is continuously sent to the DAC via DMA (Direct Memory Access) and converted into an analog voltage signal.

[0053] Step S23: The analog voltage signal drives the VCCS to generate a corresponding low-amplitude (e.g., 10μA RMS) multi-frequency sinusoidal AC constant current signal.

[0054] Step S24: Application method: Time division multiplexing is used to sequentially and cyclically apply current signals of various frequencies; or, if hardware resources permit, multiple frequency components are applied simultaneously by superimposing sine waves of different frequencies (frequency division multiplexing).

[0055] Step S3:

[0056] In some implementation examples, the skin complex impedance is determined based on a multi-frequency AC excitation signal; a multi-band AC excitation is determined based on a set of sinusoidal AC excitation current signals of specific frequency and low amplitude; the AC excitation is used to determine the application of the excitation signal to the human skin surface through a pair of excitation electrodes; the voltage response signal generated by synchronous measurement through a pair of measuring electrodes is determined based on the AC excitation and skin response; and the voltage response signal is processed by amplification, filtering, and analog-to-digital conversion (ADC) circuitry to convert it into a high-quality digital signal.

[0057] Step S31: Determine the multi-frequency band covered AC excitation based on a set of sinusoidal AC excitation current signals with specific frequencies and low amplitudes.

[0058] Step S311: Electrode placement. Securely attach a pair of excitation electrodes (E1, E2) to the subject's skin surface (usually located on the fingers, palm, or wrist). Ag / AgCl electrodes are preferred to ensure good electrical contact and low polarization effect.

[0059] Step S312: Apply signal by connecting the AC constant current signal output terminal generated by VCCS to the pair of excitation electrodes.

[0060] Step S313: Application mode, Time Division Multiplexing (TDM) mode: The system applies current signals of various frequencies to the skin sequentially and cyclically.

[0061] Step S32: Determine the application of the AC excitation to the human skin surface via a pair of excitation electrodes based on the AC excitation.

[0062] Step S321: Electrode placement. In the skin region between the excitation electrodes (E1, E2), closer to the current path, attach another pair of measurement electrodes (M1, M2). These four electrodes form a Kelvin connection.

[0063] Step S33: Determine the voltage response signal generated by synchronous measurement through a pair of measuring electrodes based on the AC excitation and skin response.

[0064] Step S331: The excitation current generates a voltage distribution within the skin tissue that is proportional to the complex impedance. The measuring electrodes (M1, M2) simultaneously detect the voltage difference (i.e., the response signal) across this segment of skin tissue. Since the measuring electrodes draw almost no current, the influence of electrode-skin contact impedance on the measurement results is completely eliminated.

[0065] Step S34: Based on the voltage response signal, determine that it has been processed by amplification, filtering and analog-to-digital conversion (ADC) circuitry to convert it into a high-quality digital signal.

[0066] Step S341: Initial amplification. The voltage response signal detected by the measuring electrodes is very weak (typically in the microvolt to millivolt range). It is first input to an instrumentation amplifier (IA). This amplifier must have extremely high input impedance (>1GΩ), high common-mode rejection ratio (CMRR>120dB), and low noise characteristics to accurately amplify the small differential voltage and strongly suppress common-mode interference such as 50 / 60Hz power frequency.

[0067] Step S342: Anti-aliasing filtering. The amplified signal is fed into an anti-aliasing filter (AAF). This filter is typically a multi-order (e.g., 4th order) active low-pass filter (such as a Bessel or Butterworth type). Its cutoff frequency is carefully set to be higher than the highest excitation frequency in the system (at least satisfying the Nyquist sampling theorem) to filter out all high-frequency noise that may cause aliasing while preserving the phase characteristics of the signal.

[0068] Step S343: Analog-to-digital conversion. The conditioned, clean analog voltage signal is sampled by a high-resolution ADC (such as a 24-bit Σ-Δ ADC) and converted into a digital signal. The ADC sampling rate must be set appropriately, much higher than the cutoff frequency of the anti-aliasing filter, to ensure that the digitization process is not distorted. For multi-channel systems, a synchronous sampling ADC should be used to ensure time consistency of data from all channels.

[0069] Step S344: Data output: The high-quality digital signal obtained by conversion is sent to the main controller (such as ARM or DSP) for subsequent digital signal processing (DSP), namely complex impedance calculation.

[0070] Step S4:

[0071] In some implementation examples, a multi-dimensional feature vector is determined based on the complex impedance; the skin complex impedance (including impedance amplitude |Z| and phase angle θ) at each excitation frequency is determined by digital lock-in amplification or fast Fourier transform (FFT) algorithm; and time-domain and frequency-domain features are determined based on the multi-frequency impedance data to form a feature vector.

[0072] Step S41: Accurately calculate the complex impedance at a single frequency point (data basis). Input: Time-domain voltage response signal V(t) acquired by the ADC and known excitation current amplitude I; Core algorithm: Digital Lock-in Amplifier (DLIA).

[0073] Step S411: Reference signal generation. For each excitation frequency f_i, generate a pair of strictly orthogonal reference signals in the digital domain: Ref_sin=sin(2πf_i t) and Ref_cos=cos(2πf_i t).

[0074] Step S412: Correlation detection, multiply the acquired response signal V(t) by the two reference signals respectively.

[0075] I_component=V(t)*Ref_sin

[0076] Q_component=V(t)*Ref_cos

[0077] Step S413: Low-pass filtering. Perform low-pass filtering (or integration) on the above two product results to extract the DC component that is only related to the excitation frequency f_i, and filter out noise and higher harmonics. The filtered result is the real part (R) and imaginary part (X) of the response voltage signal at the frequency f_i.

[0078] R = mean(I_component) * (2 / G) (G is the total system gain)

[0079] X = mean(Q_component) * (2 / G)

[0080] Step S414: Complex impedance calculation,

[0081] Impedance magnitude: |Z(f_i)|=sqrt(R^2+X^2) / I (unit: Ω)

[0082] Impedance phase: θ(f_i) = atan2(X,R) (unit: radians or degrees) (Note: the atan2 function provides complete 360° phase information)

[0083] Step S415: Output, obtain the precise complex impedance value Z_i = |Z_i|∠θ_i for each frequency point f_i.

[0084] Step S42: Determine the time-domain and frequency-domain characteristics based on the multi-frequency impedance data to form a feature vector.

[0085] Based on the set of complex impedance values ​​[Z(f1), Z(f2), ..., Z(fn)] obtained in step one at discrete frequency points, features are extracted from three dimensions: Step S421, Base Frequency Features.

[0086] Description: The most direct characteristic, reflected directly in the impedance response at each frequency point.

[0087] Extraction method: Direct reading.

[0088] Includes the following features:

[0089] |Z|_f1,|Z|_f2,...,|Z|_fn: Impedance amplitude at each frequency point.

[0090] θ_f1,θ_f2,...,θ_fn: Impedance phase at each frequency point.

[0091] Physiological significance: The amplitude generally reflects the conductivity of the skin (related to sweat gland activity and electrolyte concentration), while the phase reflects the cell membrane's capacity.

[0092] Sexual characteristics.

[0093] Step S422, Differential / Ratio Features

[0094] Description: By comparing response values ​​at different frequency points, the dispersion characteristics of the skin are highlighted, reducing the impact of individual absolute differences.

[0095] Extraction method: Combine and calculate the fundamental frequency features.

[0096] Includes the following features:

[0097] Amplitude ratio: |Z|_flow / |Z|_fhigh, for example, |Z|_10Hz / |Z|_100kHz. Low-frequency currents penetrate deeper, and this ratio reflects the differences in electrical properties of tissues at different depths.

[0098] Phase difference: θ_fi - θ_fj, for example, θ_1kHz - θ_10kHz. Reflects the variation in capacitive phase delay at different frequencies.

[0099] Frequency difference: |Z|_fi - |Z|_fj.

[0100] Step S423, Relaxation Model Features

[0101] Description: This is the highest-order, most information-condensed feature. By establishing a physical model of biological tissue impedance, several parameters with clear physiological significance are fitted.

[0102] Model: Cole-Cole model (a classic model of biological tissue impedance).

[0103] Its formula is: Z(ω)=R∞+(R0-R∞) / [1+(jωτ)^α]

[0104] Extraction method: Using optimization algorithms such as nonlinear least squares, the experimentally measured data points [f_i,|Z_i|,θ_i] are fitted onto the Cole-Cole model.

[0105] Includes features (fitting parameters):

[0106] R0: Zero-frequency resistance. It represents the resistance when the current frequency approaches 0, mainly reflecting the resistance of extracellular fluid.

[0107] R∞: Infinite frequency resistance. It represents the resistance when the current frequency approaches infinity, reflecting the total resistance of intracellular and extracellular fluids connected in parallel.

[0108] τ(Tau): Central relaxation time. It is closely related to the capacitance characteristics of the cell membrane, τ = R * C, where R is the path resistance and C is the cell membrane capacitance.

[0109] α (Alpha): Relaxation time distribution coefficient (0 < α ≤ 1). Describes the uniformity of organizational structure. α = 1 indicates only one relaxation time (very uniform organization), while the smaller the α, the wider the relaxation time distribution (the less uniform the organization).

[0110] Step S43: Feature Vector Fusion (Final Output)

[0111] All the above features are combined into a one-dimensional vector, which is then used as the input to the machine learning model.

[0112] Eigenvector F:

[0113] F = [|Z|_f1,|Z|_f2,...,|Z|_fn,θ_f1,θ_f2,...,θ_fn,(fundamental frequency characteristics)]

[0114] |Z|_ratio1,|Z|_ratio2,...,θ_diff1,θ_diff2,...,(Difference features)

[0115] R0,R∞,τ,α](Relaxation model characteristics)

[0116] Example: For a system using 5 frequencies (10Hz, 100Hz, 1kHz, 10kHz, 100kHz), its feature vector might include:

[0117] 5 amplitudes + 5 phases = 10 fundamental frequency characteristics.

[0118] Calculate 3 amplitude ratios (e.g., 10Hz / 100kHz, 100Hz / 100kHz, 1kHz / 100kHz) and 2 phase differences = 5 differential features.

[0119] Four Cole-Cole parameters (R0, R∞, τ, α).

[0120] Total number of features: 10 + 5 + 4 = 19 dimensions.

[0121] Step S5: In some implementation examples, an individualized multi-frequency impedance baseline is determined based on the baseline calibration module.

[0122] Step S51: Determine the skin impedance measured by the synchronous multi-frequency AC excitation signal based on the neutral question posed to the subject, and repeat steps S2-S5.

[0123] Step S52: Determine an individualized multi-frequency impedance baseline based on the skin impedance.

[0124] Step S521: Data Preprocessing and Screening - Preprocess the multiple sets of multi-frequency skin complex impedance data obtained synchronously by posing a neutral question in step S61. Remove abnormal data segments caused by obvious motion artifacts, transient poor contact, or other interference to ensure the stability and reliability of the baseline data. - After preprocessing, retain multiple complete cycles of multi-frequency impedance data within the valid time period (such as a stable period after the neutral question is posed and answered).

[0125] Step S522: Baseline value calculation - for each excitation frequency point f i Calculate its impedance magnitude |Z(f) i )| and phase θ(f i The baseline value for this frequency point is calculated as follows: The arithmetic mean or median of all periodic data within the valid time period is typically used.

[0126] Impedance amplitude baseline: |Z| baseline (fi )=mean(|Z1(f i )|,|Z2(f i )|,...,|Z n (f i )|)

[0127] Impedance phase baseline: θ baseline (f i )=mean(θ1(f i ),θ2(f i ),...,θ n (f i ))–

[0128] This operation is performed on all frequency points in the frequency sequence one by one, and finally a personalized reference complex impedance curve (including amplitude frequency curve and phase frequency curve) covering all excitation frequencies is obtained.

[0129] Step S523: Baseline Feature Extraction and Storage

[0130] To further optimize subsequent comparison efficiency, a set of baseline eigenvectors F can be pre-calculated and stored based on the aforementioned benchmark complex impedance curve. baseline .

[0131] F baseline The calculation method is exactly the same as step S53, which includes:

[0132] Baseline impedance amplitude and phase (fundamental frequency characteristics) at a frequency point

[0133] The calculated baseline difference characteristics (such as amplitude ratio and phase difference)

[0134] The baseline relaxation characteristic parameters (R0) obtained by fitting the reference complex impedance curve to the Cole-Cole model baseline ,R ∞baseline ,τ baseline ,α baseline )

[0135] Use the complete multi-frequency impedance baseline data and / or the calculated baseline eigenvector F baseline It is stored in the system's non-volatile memory as a benchmark for comparison in subsequent question-and-answer testing phases to determine whether a lie has occurred.

[0136] Step S524: Baseline validity verification

[0137] The system can automatically calculate the standard deviation or coefficient of variation of the baseline data. If the data fluctuation at a certain frequency point exceeds the preset threshold (such as coefficient of variation > 10%), the operator will be prompted to recalibrate the baseline to ensure the reliability of the baseline.

[0138] Step S6:

[0139] In some implementation examples, the lie detection result is determined based on the lie analysis and judgment module; a feature vector is determined based on the current question; the feature vector is compared with a multi-frequency impedance baseline; and the lie detection result is determined based on the comparison result.

[0140] Step S61: Feature Vector Input and Preprocessing. Under the current query, the real-time feature vector F obtained in step S4 (feature extraction and fusion) will be processed. current Input to the lie analysis and judgment module; for F current Perform necessary preprocessing, such as standardization or normalization, to eliminate dimensional differences between features and ensure their comparability in machine learning models. Preprocessing parameters (such as mean and standard deviation) should be calculated based on the training dataset and stored in the system.

[0141] Step S62: Baseline Comparison and Feature Difference Calculation. The pre-stored individualized multi-frequency impedance baseline feature vector F is retrieved. baseline (Obtained via step S523); Calculate the current feature vector F. current With baseline eigenvector F baseline The difference eigenvector ΔF = F current -F baseline This difference vector ΔF intuitively reflects the change in the subject's multidimensional physiological response relative to their own calm baseline under the current questioning stimulus, and is the core input for lie detection.

[0142] Step S63: Machine Learning Model Determination

[0143] The difference feature vector ΔF is input into a pre-trained machine learning classification model.

[0144] Model type: The model can be a Support Vector Machine (SVM), Random Forest, Neural Network, or other applicable classification algorithms.

[0145] Model Training: The model is trained offline using a large amount of known sample data of "honesty" and "deception" responses (their differential feature vectors and corresponding labels) to learn the mapping relationship between deceptive behavior and specific physiological response patterns.

[0146] Model output: The model outputs a classification result (such as "honest" or "deceive") and a probability score or confidence level (such as a probability of deceiving of 0.85).

[0147] Step S64: Decision and Threshold Judgment

[0148] The system makes a final judgment based on the probability score output by the model and a preset decision threshold.

[0149] For example: if the probability of deception is greater than 0.7, it is judged as "deception"; if the probability of deception is less than 0.3, it is judged as "honesty"; and if it is in between, it is judged as "uncertain".

[0150] The thresholds can be adjusted according to the requirements of the false alarm rate and false negative rate in the actual application scenario.

[0151] Step S75: Result Generation and Output

[0152] By combining the judgment results within all valid stimulus windows for the current problem (there may be multiple measurements for a single problem), the final lie detection result for that problem is generated.

[0153] The result can be in the following format:

[0154] Binary judgment: Honesty / Deception.

[0155] Probability value: Probability of being deceived: 82%.

[0156] Confidence score: deception score: 0.82 (on a scale of 0-1).

[0157] The results, along with the corresponding original signal segments and characteristic change trend graphs, are sent to the output module for display, storage, or report printing.

Claims

1. A polygraph system based on AC excitation multi-frequency skin electric signals, characterized in that, The application comprises: a signal generation unit for generating a set of sinusoidal AC excitation current signals with specific frequencies and low amplitudes; a signal application unit comprising a pair of excitation electrodes for applying the AC excitation current signals to the surface of human skin; a signal acquisition unit comprising a pair of measurement electrodes, an instrumentation amplifier, a filter and an analog-to-digital converter (ADC) for acquiring, amplifying, filtering and digitizing the voltage response signals generated on the skin by the excitation current; a signal processing unit for receiving the digitized voltage response signals and calculating the skin complex impedance at each excitation frequency based on digital lock-in amplification or fast Fourier transform algorithm; a feature extraction unit for extracting a multi-dimensional feature vector containing fundamental frequency features, differential features and relaxation features based on the skin complex impedance; a baseline calibration unit for acquiring the multi-frequency impedance data of the subject in a neutral state and establishing an individualized multi-frequency impedance baseline; a lie detection unit for comparing the currently extracted multi-dimensional feature vector with the individualized multi-frequency impedance baseline and outputting a lie detection result through a machine learning classification model.

2. The polygraph system based on AC excitation multi-frequency skin electric signal of claim 1, wherein, The signal generation unit comprises a digital-to-analog converter (DAC) and a voltage-controlled current source (VCCS) for converting the digital signals generated by the main controller into high-precision AC constant current signals.

3. The polygraph system of claim 1, wherein, The signal application and acquisition unit adopts a four-electrode method, in which the excitation electrodes and the measurement electrodes are arranged separately to eliminate the influence of electrode-skin contact impedance on the measurement results.

4. The polyurethane composition of claim 1, wherein the polyol is a polyether polyol. The set of excitation signals with specific frequencies covers low, medium and high frequency bands and contains at least three different frequency points.

5. The polyurethane composition of claim 1, wherein the polyol is a polyether polyol. The relaxation features are obtained by Cole-Cole model fitting, including zero-frequency resistance (R0), infinite-frequency resistance (R∞), central relaxation time (τ) and relaxation time distribution coefficient (α).

6. The polyurethane composition of claim 1, wherein the polyol is a polyether polyol. The machine learning classification model is a support vector machine (SVM), a random forest or a neural network model.

7. A polygraph method based on AC excitation multi-frequency skin electric signals, characterized in that, The application comprises the following steps: generating a set of sinusoidal AC excitation current signals with specific frequencies and low amplitudes; applying the excitation current signals to the surface of human skin through a pair of excitation electrodes; synchronously acquiring the voltage response signals on the skin through a pair of measurement electrodes and amplifying, filtering and analog-to-digital converting the signals; calculating the skin complex impedance at each frequency based on the excitation current and voltage response signals according to digital signal processing algorithm; extracting a multi-dimensional feature vector based on the skin complex impedance; establishing an individualized multi-frequency impedance baseline; comparing the current multi-dimensional feature vector with the multi-frequency impedance baseline, making a judgment through a machine learning model and outputting a lie detection result.

8. The method of lie detection according to claim 7, wherein, The step of extracting a multi-dimensional feature vector comprises: extracting the impedance amplitude and phase at each frequency point as fundamental frequency features; calculating the impedance amplitude ratio and phase difference between different frequency points as differential features; fitting the complex impedance data through Cole-Cole model to obtain relaxation feature parameters.

9. The method of lie detection according to claim 7, wherein, The step of establishing an individualized multi-frequency impedance baseline comprises: asking the subject a neutral question and synchronously acquiring his / her skin impedance data, and calculating the mean or median of the multi-frequency impedance after preprocessing as the baseline.

10. The method of lie detection according to claim 7, wherein, The input to the decision step by the machine learning model is a difference vector of the current feature vector and the baseline feature vector. The input to the decision step by the machine learning model is a difference vector of the current feature vector and the baseline feature vector.

Images