A method and device for detecting live biological organisms in environments with strong electromagnetic interference.

Abstract

A method and apparatus for detecting biological liveness in environments with strong electromagnetic interference, relating to the field of face recognition, are disclosed. In this method, a first physical field signal of the object to be detected is acquired within a preset first time window; it is determined whether the first physical field signal meets preset physiological behavior conditions; if the first physical field signal meets the preset physiological behavior conditions, it is analyzed to generate multimodal liveness features; if the multimodal liveness features do not meet the preset physiological behavior conditions, a second physical field signal of the object to be detected is acquired within a preset second time window, and it is determined whether the second physical field signal meets preset physiological recovery conditions; if the second physical field signal meets the preset physiological recovery conditions, it is analyzed to generate multimodal liveness features; based on the multimodal liveness features, it is determined whether the object to be detected is a living biological entity. Implementing this application improves the accuracy of biological liveness detection and recognition results.

Description

Technical Field

[0001] This application relates to the field of facial recognition, specifically to a method and device for detecting biological liveness in environments with strong electromagnetic interference. Background Technology

[0002] With the widespread application of biometric technology in key areas such as financial payments, security access control, and identity authentication, liveness detection technology, as a core component in ensuring the security of biometric systems, is becoming increasingly important. Existing liveness detection solutions mainly include: optical image analysis techniques that analyze the texture details of facial images and micro-expression movements such as blinking and lip movements; infrared thermal imaging technology that captures the temperature distribution characteristics of biological organisms; and techniques that detect single bioelectrical signals such as electrocardiograms and electromyograms. These technologies provide a basic means of distinguishing genuine users from spoofing attacks under normal conditions.

[0003] Currently, existing technologies face significant technical bottlenecks when dealing with complex application scenarios. Firstly, in environments with strong electromagnetic interference, such as industrial sites or specific security inspection areas, the imaging quality of optical sensors is severely affected, leading to lower performance and accuracy of image analysis-based liveness detection methods. Secondly, with the continuous upgrading of attack methods, single-physical-dimensional detection methods face significant challenges in discriminative uniqueness and robustness when dealing with high-precision biomimetic attacks. For example, highly realistic silicone masks, composite materials with skin-like dielectric properties, and high-fidelity video playback screens can effectively deceive traditional two-dimensional image, three-dimensional structure, or single electromagnetic property detection technologies, resulting in low accuracy of the final identification results. Summary of the Invention

[0004] This application provides a method and apparatus for detecting living organisms in environments with strong electromagnetic interference, which improves the accuracy of the detection and identification results of living organisms.

[0005] A first aspect of this application provides a method for detecting living organisms in environments with strong electromagnetic interference. The method includes: acquiring a first physical field signal of the target object within a preset first time window, the first physical field signal including an electromagnetic modulation signal, a vital micro-motion signal, and a thermal radiation characteristic signal; determining whether the first physical field signal meets preset physiological behavior conditions; if the first physical field signal meets the preset physiological behavior conditions, analyzing the first physical field signal to generate multimodal living organism features; if the multimodal living organism signal does not meet the preset physiological behavior conditions, acquiring a second physical field signal of the target object within a preset second time window, and determining whether the second physical field signal meets preset physiological recovery conditions; if the second physical field signal meets the preset physiological recovery conditions, analyzing the second physical field signal to generate multimodal living organism features; and determining whether the target object is a living organism based on the multimodal living organism features.

[0006] By employing the above technical solution, signals from three different physical fields—electromagnetic modulation signals, micro-movement signals, and thermal radiation characteristic signals—are simultaneously acquired within a preset first time window. This enables comprehensive perception of the physiological characteristics of the target organism. Each physical field signal reflects the inherent characteristics of a living organism from different dimensions, forming a complementary detection mechanism. Determining whether the first physical field signal meets preset physiological behavior conditions allows for rapid screening of targets that conform to normal physiological characteristics, improving detection efficiency. When the first physical field signal meets the preset physiological behavior conditions, multimodal live organism features are directly generated, shortening the detection time. When the conditions are not met, a supplementary detection in the second time window is initiated. By determining whether the second physical field signal meets the preset physiological recovery conditions, abnormal situations caused by temporary interference are effectively identified, avoiding misjudgments caused by brief physiological respiration or environmental interference. The dual-time-window detection mechanism fully utilizes the self-recovery characteristics of organisms. Even if detection is affected at a certain moment under strong electromagnetic interference, the recovery process of physiological signals can still be accurately captured through delayed detection. Based on multimodal live organism features, comprehensive judgment is made, integrating the complementary information of the three physical field signals, significantly improving the accuracy and robustness of live organism detection in complex electromagnetic environments. This application introduces a multi-physics signal acquisition and analysis method in environments with strong electromagnetic interference. Combined with preset physiological behavior and recovery conditions, it utilizes electromagnetic modulation signals, micro-movement signals, and thermal radiation characteristic signals to generate multimodal live organism features, thereby achieving accurate detection of living organisms in complex environments. This method effectively solves the problems of poor detection performance and insufficient robustness of single-dimensional detection in traditional technologies under strong interference environments, significantly improving the accuracy and anti-interference capability of live organism detection.

[0007] Optionally, the acquisition of the first physical field signal of the object to be detected specifically includes: emitting a swept-frequency electromagnetic wave to the object to be detected using an electromagnetic modulation sensing unit, and receiving a reflected signal modulated by the object to be detected; generating an electromagnetic modulation signal characterizing the complex permittivity spectrum of the object to be detected based on the amplitude ratio and phase difference between the swept-frequency electromagnetic wave and the reflected signal; emitting a linear frequency modulated signal to the object to be detected using a life micro-motion sensing unit and receiving an echo signal; processing the echo signal to obtain a range Doppler spectrum, separating the respiratory component and the heartbeat component from the range Doppler spectrum, and calculating the life micro-motion signal based on the respiratory component and the heartbeat component; scanning the object to be detected using a thermal radiation characteristic sensing unit to acquire the thermal radiation intensity distribution of the object to be detected within a preset band; converting the thermal radiation intensity distribution into a temperature field, recording the spatiotemporal changes of the temperature field, and generating the thermal radiation characteristic signal.

[0008] By employing the aforementioned technical solutions, the electromagnetic modulation sensing unit, through transmitting swept-frequency electromagnetic waves and analyzing the amplitude ratio and phase difference of the reflected signals, can accurately measure the complex permittivity spectrum of the object under test at different frequencies. This spectral information directly reflects the electromagnetic properties of biological tissues. The micro-motion sensing unit transmits linear frequency-modulated signals and obtains the range-Doppler spectrum through echo signal processing, achieving high-precision detection of minute physiological movements. The respiratory and cardiac components separated from the range-Doppler spectrum can accurately reflect the dynamic characteristics of vital signs, even capturing and quantifying minute movements at the millimeter level. The thermal radiation characteristic sensing unit scans and collects the thermal radiation intensity distribution within a preset band and converts it into a temperature field. This not only obtains static temperature distribution information but also captures dynamic thermal characteristics related to blood flow by recording the spatiotemporal changes of the temperature field. This spatiotemporal joint analysis can effectively identify the unique thermophysiological patterns of living organisms. The collaborative work of the three sensing units achieves comprehensive information acquisition in the electromagnetic, motion, and thermal domains, providing a rich and reliable raw data foundation for subsequent multimodal feature analysis.

[0009] Optionally, before acquiring the first physical field signal of the object to be detected, the method further includes: acquiring the background electromagnetic spectrum of the current detection environment, identifying the interference frequency band and interference intensity in the background electromagnetic spectrum; constructing an interference feature matrix based on the interference frequency band and interference intensity; adaptively adjusting the transmission frequency band of the electromagnetic modulation sensing unit using the interference feature matrix to avoid the interference frequency band and generate an anti-interference transmission strategy; and applying adaptive notch filtering to the echo signal of the life micro-motion sensing unit based on the interference feature matrix to suppress the noise component corresponding to the interference frequency band.

[0010] By employing the above technical solutions, the background electromagnetic spectrum of the current detection environment is obtained, and interference frequency bands and interference intensity are identified, enabling real-time perception and quantitative assessment of the electromagnetic environment. This provides accurate environmental parameters for subsequent anti-interference measures. The constructed interference feature matrix comprehensively describes the frequency distribution, power level, and time-varying characteristics of various electromagnetic interference sources in the environment, enabling the system to intelligently identify and predict interference modes. The interference feature matrix is ​​used to adaptively adjust the transmission frequency band of the electromagnetic modulation sensing unit. By actively avoiding interference frequency bands and generating an anti-interference transmission strategy, the measurement accuracy of the electromagnetic modulation signal is ensured to be unaffected by environmental interference, achieving high-quality dielectric constant measurement results even in industrial environments or near communication base stations. An adaptive notch filter based on the interference feature matrix is ​​applied to the echo signal of the life micro-motion sensing unit, accurately suppressing noise components in specific interference frequency bands while protecting useful physiological signal components from being filtered out, significantly improving the detection capability of weak life signals in strong electromagnetic interference environments. This proactive environmental adaptation mechanism enables the entire detection system to maintain stable detection performance in various complex electromagnetic environments.

[0011] Optionally, determining whether the first physical field signal meets the preset physiological behavior conditions specifically includes: determining whether the electromagnetic modulation signal is within the electromagnetic confidence interval, determining whether the thermal radiation characteristic signal is within the thermal radiation confidence interval, and determining whether the vital micro-motion signal is greater than or equal to a preset micro-motion threshold; if it is determined that the electromagnetic modulation signal is within the electromagnetic confidence interval, the thermal radiation characteristic signal is within the thermal radiation confidence interval, and the vital micro-motion signal is greater than or equal to the preset micro-motion threshold, then it is determined that the multimodal data feature meets the preset physiological behavior conditions; if it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the vital micro-motion signal is less than the preset micro-motion threshold, then it is determined that the multimodal data feature does not meet the preset physiological behavior conditions.

[0012] By employing the aforementioned technical solution, multi-dimensional physiological characteristic verification is achieved by simultaneously determining whether the electromagnetic modulation signal is within the electromagnetic confidence interval, whether the thermal radiation characteristic signal is within the thermal radiation confidence interval, and whether the vital micro-motion signal is greater than or equal to a preset micro-motion threshold. Each judgment condition corresponds to a necessary characteristic of a living organism, forming a rigorous logical judgment system. The electromagnetic confidence interval is set based on statistical analysis of a large number of live samples, accurately defining the dielectric characteristic range of normal biological tissues; the thermal radiation confidence interval reflects the normal body temperature distribution and thermal dynamic characteristics of the human body; and the preset micro-motion threshold ensures that the detected object has basic vital signs. Using a logical judgment strategy, the preset physiological behavior condition is only determined to be met when all three conditions are satisfied. This strict judgment standard effectively prevents various deception attacks, as forgeries are unlikely to simultaneously meet the physiological characteristic requirements of all three physical domains. When any condition is not met, it is immediately determined that the preset physiological behavior condition is not met, quickly identifying abnormal situations and triggering subsequent supplementary detection processes. This ensures high detection security and improves the system's response speed through rapid judgment.

[0013] Optionally, determining whether the second physical field signal meets the preset physiological recovery condition specifically includes: if it is determined that the vital micro-motion signal is less than the preset micro-motion threshold, and the electromagnetic modulation signal is within the electromagnetic confidence interval, and the thermal radiation characteristic signal is within the thermal radiation confidence interval, then extract the target vital micro-motion signal from the second physical field signal; determine whether a jump signal greater than or equal to the preset micro-motion threshold appears in the target vital micro-motion signal; if it is determined that a jump signal greater than or equal to the preset micro-motion threshold appears in the target vital micro-motion signal, then determine that the second physical field signal meets the preset physiological recovery condition; if it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the jump signal does not exist in the target vital micro-motion signal, then determine that the second physical field signal does not meet the preset physiological recovery condition.

[0014] By employing the aforementioned technical solution, the system can intelligently identify single-mode abnormalities in specific situations where the vital micro-motion signal is below a preset micro-motion threshold, but the electromagnetic modulation signal and thermal radiation characteristic signal are normal. This situation is typically caused by temporary physiological inhibition or transient environmental interference, rather than true non-living characteristics. Extracting the target vital micro-motion signal from the second physical field signal and determining whether a jump signal occurs fully utilizes the organism's adaptive recovery mechanism. When the interference disappears or the physiological state recovers, vital signs naturally return to normal levels; this dynamic recovery process is unique to living organisms. The detection of jump signals not only focuses on the absolute value of the signal amplitude but, more importantly, captures the dynamic change process from below the threshold to above the threshold. This time-dimensional feature analysis significantly improves the accuracy of identifying living organisms. By continuously monitoring whether the other two signals remain within the confidence interval within the second time window, the consistency and stability of the overall physiological characteristics of the detected object are ensured, effectively preventing deceptive attacks that exploit the time-varying characteristics of signals. This judgment mechanism based on physiological recovery characteristics significantly reduces the false negative rate caused by transient interference, improving the user experience of the system in practical applications.

[0015] Optionally, the step of analyzing the first physical field signal to generate multimodal living features specifically includes: extracting features from the electromagnetic modulation signal to obtain the real part of the dielectric constant and the loss factor features of the object to be detected within a preset frequency band, and combining the real part of the dielectric constant and the loss factor features into an electromagnetic feature vector; extracting features from the vital micro-motion signal to obtain respiratory rate, heart rate, respiratory amplitude, and heart rate variability parameters, and combining the respiratory rate, heart rate, respiratory amplitude, and heart rate variability parameters into a micro-motion feature vector; extracting features from the thermal radiation characteristic signal to obtain the spatial gradient features and temporal fluctuation features of the temperature field, and combining the spatial gradient features and the temporal fluctuation features into a thermal radiation feature vector; and weighting and fusing the electromagnetic feature vector, the micro-motion feature vector, and the thermal radiation feature vector according to preset weights to generate the multimodal living features.

[0016] By employing the above technical solutions, the real part of the dielectric constant and the loss factor are extracted from electromagnetic modulation signals, comprehensively characterizing the electromagnetic response properties of biological tissues. The real part of the dielectric constant reflects the polarization ability and water content of the tissue, while the loss factor characterizes the attenuation characteristics of electromagnetic energy. The electromagnetic feature vector formed by the combination of these two can accurately distinguish the differences in electromagnetic properties of different materials. Respiratory rate, heart rate, respiratory amplitude, and heart rate variability parameters are extracted from vital micro-motion signals. This not only obtains basic physiological frequency information but also reflects the regulatory function of the autonomic nervous system through the heart rate variability parameter. The micro-motion feature vector composed of the combination of these parameters comprehensively describes the static and dynamic characteristics of vital signs. Spatial gradient features and temporal fluctuation features are extracted from thermal radiation characteristic signals. Spatial gradient features reflect the distribution pattern of body surface temperature and blood vessel orientation, while temporal fluctuation features capture the minute temperature changes caused by blood flow pulsation. The thermal radiation feature vector integrates the spatial structure and temporal dynamic information of the thermal field. By using a weighted fusion strategy with preset weights, the multimodal liveness features are optimized and combined according to the reliability and importance of different modal features. The generated multimodal liveness features not only retain the unique information of each modality, but also enhance the overall discrimination ability through feature-level fusion, providing information-rich and robust feature representations for subsequent liveness detection.

[0017] Optionally, determining whether the object to be detected is a living organism based on the multimodal liveness features specifically includes: inputting the multimodal liveness features into a preset liveness discrimination model, which is generated through training on living organism samples and non-living organism samples; classifying and calculating the multimodal liveness features using the liveness discrimination model and outputting a liveness confidence score; determining whether the liveness confidence score is greater than or equal to a preset liveness determination threshold; if the liveness confidence score is greater than or equal to the preset liveness determination threshold, then determining that the object to be detected is a living organism; if the liveness confidence score is less than the preset liveness determination threshold, then determining that the object to be detected is a non-living organism.

[0018] By adopting the above technical solution, the preset liveness detection model is generated based on a large number of biological live and non-live samples. Through deep learning technology, it automatically learns complex feature patterns that distinguish between live and non-live individuals, exhibiting stronger generalization ability and adaptability compared to traditional threshold-based judgment methods. When classifying multimodal liveness features, the liveness detection model can automatically uncover the inherent correlations and complementary relationships between different modal features, extracting deeper discriminative information through nonlinear mapping and feature interaction. The output liveness confidence score provides continuous probabilistic evaluation results, not only giving a binary judgment result but also quantifying the credibility of the judgment, providing important reference for system risk control and decision optimization. The preset liveness detection threshold can be flexibly adjusted according to the security requirements of different application scenarios. In high-security scenarios, the threshold can be increased to reduce the false recognition rate, while in ordinary scenarios, the threshold can be appropriately decreased to improve the user pass rate. By comparing the liveness confidence score with the preset liveness determination threshold, a probability-based scientific decision-making method is achieved. This method fully considers the uncertainties in the detection process, making the determination results more robust and reliable, and effectively improving the accuracy of biological liveness detection in environments with strong electromagnetic interference.

[0019] Secondly, embodiments of this application provide a biological liveness detection device for use in a strong electromagnetic interference environment. The biological liveness detection device for use in a strong electromagnetic interference environment includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the biological liveness detection device for use in a strong electromagnetic interference environment to perform the method described in the first aspect and any possible implementation thereof.

[0020] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a biological liveness detection device applied to a strong electromagnetic interference environment, cause the biological liveness detection device applied to a strong electromagnetic interference environment to perform the method described in the first aspect and any possible implementation thereof.

[0021] Fourthly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a biological liveness detection device applied to a strong electromagnetic interference environment, causes the biological liveness detection device applied to a strong electromagnetic interference environment to perform the method described in the first aspect and any possible implementation thereof.

[0022] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: 1. By combining electromagnetic modulation signals, vital micro-motion signals, and thermal radiation characteristic signals, a multimodal liveness detection method is proposed, effectively improving the system's anti-interference capability in environments with strong electromagnetic interference. Through real-time acquisition and analysis of multi-physics field signals, and by adopting adaptive adjustment and filtering strategies for interference frequency bands, the reliability and accuracy of signal acquisition are ensured.

[0023] 2. Preset physiological behavioral and recovery conditions are introduced to judge and analyze the dynamic changes of signals, thereby enhancing the sensitivity and robustness of detecting living organisms in complex scenarios. This dynamic signal monitoring mechanism can effectively address the problems of signal anomalies or short-term failures, avoiding misjudgments caused by brief physiological respiration or environmental interference, and further improving the stability of detection.

[0024] 3. By weighted fusion of multimodal signal features and training a pre-defined liveness detection model, this method significantly improves the accuracy of distinguishing between living and non-living organisms. Through the calculation of liveness confidence scores and threshold determination, the scheme possesses higher detection accuracy and reliability, effectively addressing complex challenges such as high-precision biomimetic attacks. Attached Figure Description

[0025] Figure 1 This is a schematic flowchart of a biological liveness detection method for use in environments with strong electromagnetic interference, as disclosed in an embodiment of this application. Figure 2 This is another schematic flowchart of a biological liveness detection method disclosed in this application for use in environments with strong electromagnetic interference; Figure 3 This is a schematic diagram of a biological liveness detection device for use in environments with strong electromagnetic interference, provided in an embodiment of this application.

[0026] Explanation of reference numerals in the attached drawings: 301, Central Processing Unit; 302, Read-Only Memory; 303, Random Access Memory; 304, Bus; 305, Input / Output Interface; 306, Input Section; 307, Output Section; 308, Storage Section; 309, Communication Section; 310, Driver; 311, Removable Media. Detailed Implementation

[0027] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0028] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0029] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple system devices refer to two or more system devices, and multiple screen terminals refer to two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0030] This application provides a method for detecting living organisms in environments with strong electromagnetic interference, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a biological liveness detection method for use in environments with strong electromagnetic interference, provided in an embodiment of this application. The method is applied to a device, which is a server. The server can execute a biological liveness detection program for environments with strong electromagnetic interference. The server can be a single server, a server cluster consisting of multiple servers, or a cloud computing service center. The method includes steps S101 to S106, as follows: Step S101: Within a preset first time window, acquire the first physical field signal of the object to be detected. The first physical field signal includes electromagnetic modulation signal, micro-movement signal, and thermal radiation characteristic signal.

[0031] In step S101, the preset first time window refers to a specific duration pre-set for the initial data acquisition; the object to be detected refers to the target that needs to be determined as a living organism, such as a face in front of a security access control or financial device; the first physical field signal refers to the initial multimodal raw data set acquired within the preset first time window; the electromagnetic modulation signal is used to represent the dielectric property disturbance data generated by the object to be detected to the surrounding electromagnetic field; the micro-motion signal is used to represent the surface micro-displacement data caused by the physiological activities of the object to be detected, such as breathing and heartbeat; and the thermal radiation characteristic signal is used to represent the infrared thermal radiation data emitted by the object to be detected itself.

[0032] Specifically, when the object to be detected enters the detection range and triggers the system, the controller sends a synchronous trigger command to the multi-physics field collaborative sensing module, initiating the timing of a preset first time window. Within this time window, the electromagnetic modulation sensing unit, the micro-motion sensing unit, and the thermal radiation characteristic sensing unit simultaneously begin operation, respectively detecting the response of the object to be detected in the three physical dimensions of electromagnetics, motion, and thermal radiation. The analog signal acquired by the electromagnetic modulation sensing unit, the Doppler frequency shift signal acquired by the micro-motion sensing unit, and the heat flux signal acquired by the thermal radiation characteristic sensing unit are synchronously transmitted to the signal conditioning and acquisition module. The signal conditioning and acquisition module performs filtering, amplification, and analog-to-digital conversion processing on these three analog signals, generating three independent digital signal streams, which together constitute the first physical field signal.

[0033] In one possible implementation, acquiring the first physical field signal of the object to be detected specifically includes: emitting a swept-frequency electromagnetic wave to the object to be detected using an electromagnetic modulation sensing unit and receiving a reflected signal modulated by the object to be detected; generating an electromagnetic modulation signal characterizing the complex permittivity spectrum of the object to be detected based on the amplitude ratio and phase difference between the swept-frequency electromagnetic wave and the reflected signal; emitting a linear frequency modulated signal to the object to be detected using a vital micro-motion sensing unit and receiving an echo signal; processing the echo signal to obtain a range Doppler spectrum, separating the respiratory component and the heartbeat component from the range Doppler spectrum, and calculating the vital micro-motion signal based on the respiratory component and the heartbeat component; scanning the object to be detected using a thermal radiation characteristic sensing unit to acquire the thermal radiation intensity distribution of the object to be detected within a preset band; converting the thermal radiation intensity distribution into a temperature field and recording the spatiotemporal changes of the temperature field to generate the thermal radiation characteristic signal.

[0034] Specifically, the server executes the electromagnetic characteristic detection process by controlling the electromagnetic modulation sensing unit. The server first configures a vector network analyzer, setting the frequency sweep range to 100MHz to 6GHz, the step frequency to 10MHz, and the output power to -10dBm to ensure human safety. The server controls a directional antenna to transmit frequency-sweeping electromagnetic waves towards the object being detected, with a dwell time of 50 microseconds at each frequency point, and a complete sweep taking 30 milliseconds. When the electromagnetic wave encounters the biological tissue of the object being detected, due to the dielectric properties of water, protein, and ions in the tissue, the incident wave is partially reflected, resulting in a phase shift. The server receives the reflected signal via a coaxial cable, uses orthogonal demodulation technology to separate the in-phase and quadrature components, and calculates the reflection coefficient S11 parameter for each frequency point. Based on transmission line theory, the server converts the S11 parameter into a complex permittivity, where the real part reflects the polarization capability of the tissue and the imaginary part characterizes energy loss characteristics, ultimately generating a complex permittivity spectrum containing data from 600 frequency points as the electromagnetic modulation signal.

[0035] Simultaneously, the server activates the millimeter-wave radar system of the vital signs sensing unit. The server is configured to operate the radar in the 77GHz band with a bandwidth of 4GHz, using sawtooth wave modulation to generate a linear frequency modulated signal with a modulation period of 40 microseconds to ensure simultaneous detection of slow respiratory movements and rapid heartbeat vibrations. The server transmits a frequency-modulated continuous wave towards the chest region of the target object via its transmitting antenna. When the radar wave encounters the chest wall displaced by breathing and heartbeat, the reflected signal exhibits a Doppler frequency shift. The server mixes the received echo signal with the local oscillator signal to obtain an intermediate frequency signal containing distance and velocity information. Through a two-dimensional fast Fourier transform, the server generates a range-Doppler spectrum, where the horizontal axis represents the target distance and the vertical axis represents the velocity. The server employs an adaptive filtering algorithm to identify the respiratory component (0.1-0.5Hz) and the heartbeat component (0.8-2.0Hz) in the spectrum. Peak tracking technology is used to extract the instantaneous frequency and amplitude of each component, thereby calculating the respiratory depth and heartbeat displacement, forming a sequence of vital signs signals including timestamps.

[0036] In parallel, the server activates the infrared thermal imaging system of the thermal radiation characteristic sensing unit. The server stabilizes the operating temperature of the thermopile array sensor at 25 degrees Celsius and activates the temperature compensation circuit to eliminate the influence of ambient temperature drift. The sensor array contains 32×24 thermistors, each responsible for detecting the radiation intensity in the 8-14 micrometer long-wave infrared band at its corresponding spatial location. The server drives the sensor to scan the surface of the object under test line by line at a rate of 30 frames per second, acquiring a 16-bit radiation intensity value matrix as raw data. Applying Planck's blackbody radiation law and considering the emissivity of biological tissue (0.98), the server converts the radiation intensity values ​​into absolute temperature values. To improve temperature resolution, the server uses a bilinear interpolation algorithm to upsample the temperature matrix to a resolution of 160×120. The server establishes a circular buffer to store five consecutive seconds of temperature field data, extracts the rate of temperature change over time through differential operations, identifies periodic temperature fluctuations caused by blood flow pulsation, and finally generates a four-dimensional thermal radiation characteristic signal containing spatial distribution and temporal evolution information.

[0037] For example, when detecting a seated user, the server simultaneously activates three sensing units. The electromagnetic modulation sensing unit detects a dielectric constant of 68.5 (real part) and 18.2 (imaginary part) at 3GHz, consistent with typical values ​​for human muscle tissue. The vital signs sensing unit detects respiratory movements of the chest wall at a frequency of 0.25Hz and an amplitude of 3.2mm, superimposed with heartbeat vibrations at a frequency of 1.2Hz and an amplitude of 0.8mm. The thermal radiation sensing unit detects an average temperature of 35.8 degrees Celsius on the user's face, with a periodic temperature change of 0.3 degrees Celsius in the nasal region due to respiratory airflow. The server performs time-aligned and synchronized storage of these multi-dimensional physical field signals, providing a complete data foundation for subsequent liveness feature extraction and discrimination.

[0038] Step S102: Determine whether the first physical field signal meets the preset physiological behavior conditions.

[0039] In step S102, the preset physiological behavior condition refers to a predefined logical judgment criterion used to determine whether the subject to be tested exhibits all expected and normal physiological activity signs during the initial collection phase.

[0040] Specifically, after receiving the first physical field signal, the server does not immediately perform complex feature engineering, but instead performs a rapid preliminary assessment. The server calculates whether the average amplitude of the electromagnetic modulation signal is within a preset numerical range representing biological tissue, whether the average temperature of the thermal radiation characteristic signal is within, for example, the normal human body temperature range of 35 to 37 degrees Celsius, and whether the signal energy or frequency domain peak value of the vital micro-movement signal exceeds a minimum energy threshold indicating the presence of respiratory or cardiac activity. The preset physiological behavior condition is set to be met only if all three preliminary assessment results are positive. Based on this judgment, the server outputs a Boolean value to determine whether to proceed to step S103 or step S104.

[0041] In one possible implementation, determining whether the first physical field signal meets preset physiological behavior conditions specifically includes: determining whether the electromagnetic modulation signal is within the electromagnetic confidence interval, determining whether the thermal radiation characteristic signal is within the thermal radiation confidence interval, and determining whether the vital micro-motion signal is greater than or equal to a preset micro-motion threshold; if it is determined that the electromagnetic modulation signal is within the electromagnetic confidence interval, the thermal radiation characteristic signal is within the thermal radiation confidence interval, and the vital micro-motion signal is greater than or equal to the preset micro-motion threshold, then it is determined that the multimodal data feature meets the preset physiological behavior conditions; if it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the vital micro-motion signal is less than the preset micro-motion threshold, then it is determined that the multimodal data feature does not meet the preset physiological behavior conditions.

[0042] Specifically, the server executes a comprehensive judgment process for multimodal data characteristics, and uses a parallel processing architecture to simultaneously analyze whether the three physical field signals meet the corresponding physiological behavior standards.

[0043] To determine the electromagnetic modulation signal, the server first extracts key frequency data from the complex permittivity spectrum, focusing on three frequencies sensitive to biological tissues: 1 GHz, 3 GHz, and 5 GHz. The server calculates the real part of the permittivity at each frequency and compares it to a preset electromagnetic confidence interval. This confidence interval is based on statistical analysis of a large number of live samples. At 1 GHz, the real part of the permittivity of normal human tissue should be in the range of [55-65]; at 3 GHz, it should be in the range of [50-58]; and at 5 GHz, it should be in the range of [45-52]. Only when the permittivity values ​​at all three key frequencies fall within their respective confidence intervals does the server determine that the electromagnetic modulation signal is within the electromagnetic confidence interval.

[0044] To determine the thermal radiation characteristic signal, the server calculates multiple statistical indicators from the temperature field data for comprehensive evaluation. First, the server extracts the average temperature value of the region of interest; the normal surface temperature of the human body should fall within the thermal radiation confidence interval of [32-37] degrees Celsius. The server further analyzes the spatial characteristics of the temperature distribution, calculating the directional consistency of the temperature gradient. Under normal circumstances, it should show a decreasing trend from the center of the face to the edge, with a gradient value ranging from 0.1 to 0.5 degrees Celsius / cm. The server also needs to verify the temporal stability of the temperature field by calculating the standard deviation of the temperature over 5 consecutive seconds, ensuring it is less than 0.3 degrees Celsius to eliminate interference from external heat sources. Furthermore, the server analyzes the temperature time series using Fourier transform to identify the blood flow pulsation characteristic peak at 0.2-0.3 Hz, a physiological phenomenon unique to living organisms. Only when the average temperature, spatial distribution, temporal stability, and pulsation characteristics all meet the standards does the server determine that the thermal radiation characteristic signal is within the thermal radiation confidence interval.

[0045] For determining vital signs, the server employs a multi-level threshold detection strategy. First, the server performs amplitude detection on the separated respiratory and heartbeat components. The preset micro-motion threshold includes two sub-thresholds: respiratory amplitude should be greater than or equal to 2 mm, and heartbeat amplitude should be greater than or equal to 0.5 mm. The server uses a peak detection algorithm to identify local maxima and minima of the signal and calculates the peak-to-peak value as the motion amplitude. To improve detection reliability, the server uses a sliding window method to calculate the average amplitude within a 3-second time window, avoiding instantaneous noise interference. The server also verifies the periodicity of the motion by calculating the regularity index of respiration and heartbeat using an autocorrelation function. The autocorrelation coefficient for normal respiration should be greater than 0.8, and the autocorrelation coefficient for heartbeat should be greater than 0.7. Only when the amplitudes of both respiration and heartbeat exceed the corresponding thresholds and exhibit obvious periodicity is the server determined that the vital signs are greater than or equal to the preset micro-motion threshold.

[0046] The server uses a logical AND operation to synthesize the three judgment results. The server sets three Boolean flags in memory, corresponding to the judgment status of electromagnetic modulation signals, thermal radiation characteristic signals, and vital micro-motion signals, respectively. The server outputs a judgment result indicating that the multimodal data characteristics meet preset physiological behavior conditions only if all three flags are true. If any flag is false, the server immediately terminates the current judgment process, outputs a result indicating that the conditions are not met, and records which specific modality's signal failed to meet the criteria for subsequent anomaly analysis and processing strategy selection.

[0047] For example, during the detection process, the server found that the real part of the dielectric constant of the electromagnetic modulation signal of a certain object under test was 54.3 at the 3GHz frequency, falling within the confidence interval of [50-58]. The thermal radiation characteristic signal showed an average temperature of 35.2 degrees Celsius, a temperature gradient of 0.25 degrees Celsius / cm, and detected blood flow pulsation at 0.25Hz, with all indicators being normal. However, the vital signs signal showed a respiratory amplitude of only 1.5 mm, below the threshold of 2 mm, possibly due to the user being in a shallow breathing state or being affected by electromagnetic interference. Based on the logic and operation rules, the server determined that the multimodal data characteristics did not meet the preset physiological behavior conditions and needed to enter the supplementary detection process in the second time window.

[0048] Step S103: If it is determined that the first physical field signal meets the preset physiological behavior conditions, then the first physical field signal is analyzed to generate multimodal live features.

[0049] In step S103, the multimodal liveness feature refers to a structured data set, usually a numerical vector, which is composed of multiple key feature vectors that are deeply extracted from various physical field signals and can accurately describe the liveness attributes.

[0050] Specifically, if the judgment result of step S102 is that the preset physiological behavior conditions are met, it indicates that the initially acquired signal is complete and normal. At this time, the server will perform a refined analysis on the complete first physical field signal. For example, the server will demodulate the electromagnetic modulation signal to calculate a more accurate estimate of the dielectric constant; perform Fourier transform or wavelet analysis on the vital micro-motion signal to extract multiple features such as respiratory rate, heart rate, and respiratory harmonic ratio; and analyze the thermal radiation characteristic signal to calculate features such as the highest facial temperature, average temperature, temperature gradient, and dynamic temperature changes in the nasal cavity region. The server arranges all these calculated values ​​in a predetermined order, combines them into a high-dimensional multimodal live feature vector, and passes this multimodal live feature to step S106 for final judgment.

[0051] In one possible implementation, the analysis of the first physical field signal to generate multimodal living features specifically includes: extracting features from the electromagnetic modulation signal to obtain the real part of the dielectric constant and the loss factor features of the object to be detected within a preset frequency band, and combining the real part of the dielectric constant and the loss factor features into an electromagnetic feature vector; extracting features from the vital micro-motion signal to obtain respiratory rate, heart rate, respiratory amplitude, and heart rate variability parameters, and combining the respiratory rate, heart rate, respiratory amplitude, and heart rate variability parameters into a micro-motion feature vector; extracting features from the thermal radiation characteristic signal to obtain the spatial gradient features and temporal fluctuation features of the temperature field, and combining the spatial gradient features and the temporal fluctuation features into a thermal radiation feature vector; and weighting and fusing the electromagnetic feature vector, the micro-motion feature vector, and the thermal radiation feature vector according to preset weights to generate the multimodal living features.

[0052] Specifically, the server performs in-depth analysis and feature extraction on three physical field signals through a specially designed feature engineering pipeline, ensuring that the extracted features can fully reflect the essential characteristics of living organisms.

[0053] For feature extraction of electromagnetic modulation signals, the server first divides the complex dielectric constant spectrum into frequency bands, dividing the complete spectrum from 100MHz to 6GHz into low-frequency (100MHz-1GHz), mid-frequency (1GHz-3GHz), and high-frequency (3GHz-6GHz) bands. Within each band, the server selects 10 equally spaced characteristic frequency points and calculates the real part of the dielectric constant at each point. The real part of the dielectric constant reflects the polarization ability of biological tissues and is directly related to the water content and electrolyte concentration within the tissue. The server fits the curves of the real part of the dielectric constant versus frequency within each frequency band using the least squares method, extracting the slope and intercept of the curves as dispersion features. For loss factor features, the server calculates the loss tangent (the ratio of the imaginary to the real part of the dielectric constant) at each characteristic frequency point; this parameter characterizes the degree of electromagnetic energy attenuation in the tissue. The server also calculates Cole-Cole relaxation parameters, including static dielectric constant, optical frequency dielectric constant, relaxation time, and distribution parameters; these parameters accurately describe the dielectric relaxation behavior of biological tissues. Ultimately, the server combines the real part of the dielectric constant at 30 frequency points, 3 sets of dispersion parameters, 30 loss factor values, and 4 Cole-Cole parameters into a 67-dimensional electromagnetic eigenvector.

[0054] For feature extraction of vital micro-motion signals, the server employs a multi-scale analysis method to extract physiological features at different levels. First, the server performs peak detection on the separated respiratory signals, identifies the start and end times of each respiratory cycle, calculates the time interval between adjacent respiratory cycles, and obtains the average respiratory rate (breaths / minute) and its standard deviation through statistical analysis. The server calculates the total chest cavity displacement for each respiratory cycle through integral calculations and takes the average of multiple cycles as the respiratory amplitude feature. For heartbeat signals, the server uses the autocorrelation function method to accurately extract the instantaneous heart rate and calculates the heart rate time series over 5 minutes. Based on the heart rate time series, the server calculates several heart rate variability parameters: time-domain parameters include SDNN (standard deviation of all normal heartbeat intervals), RMSSD (root mean square of the difference between adjacent heartbeat intervals), and pNN50 (percentage of adjacent heartbeat intervals with a difference greater than 50ms); frequency-domain parameters are obtained through power spectrum analysis, including low-frequency power (0.04-0.15Hz), high-frequency power (0.15-0.4Hz), and the low-frequency / high-frequency power ratio. The server also extracts nonlinear parameters, such as sample entropy and approximate entropy, to reflect the complexity of heart rate. Finally, an 11-dimensional micro-motion feature vector is constructed, including the mean and standard deviation of respiratory rate, respiratory amplitude, instantaneous heart rate, and six heart rate variability parameters.

[0055] Feature extraction of thermal radiation signals focuses on capturing the spatial distribution patterns and temporal evolution of the temperature field. The server first calculates the spatial gradient of the temperature field data, using the Sobel operator to calculate the temperature gradients in the horizontal and vertical directions, and then calculates the gradient magnitude and direction. The server divides the facial region into five sub-regions: forehead, eyes, nose, cheeks, and chin, and calculates the average gradient magnitude and gradient direction entropy for each region. The gradient direction entropy reflects the degree of order in the temperature distribution; a living face will exhibit a specific gradient pattern due to the distribution of blood vessels. The server also calculates the spatial autocorrelation function of the global temperature distribution and extracts the correlation length as a measure of the continuity of the temperature field. For temporal fluctuation features, the server performs spectral analysis on the temperature time series of each pixel, extracting the average power spectral density of the 0.2-0.4Hz blood flow pulsation frequency band. The server calculates the temporal autocorrelation function of temperature fluctuations and extracts the autocorrelation time as a temperature stability index. Furthermore, the server extracts the statistical moment features of the temperature field, including skewness and kurtosis, reflecting the asymmetry and sharpness of the temperature distribution. Finally, a 15-dimensional thermal radiation feature vector is generated, which includes the gradient magnitude of 5 regions, the entropy of 5 gradient directions, the spatial correlation length, the pulsating power spectral density, the autocorrelation time, the skewness, and the kurtosis.

[0056] In the feature fusion stage, the server employs an adaptive weighting strategy to generate the final multimodal liveness features. The server first normalizes the three feature vectors, mapping each feature to the [0, 1] interval to eliminate the influence of dimensional differences. Preset weights are determined based on the reliability of each modality under different environmental conditions: in normal environments, the electromagnetic feature weight is 0.3, the micro-motion feature weight is 0.4, and the thermal radiation feature weight is 0.3; in environments with strong electromagnetic interference, the server automatically reduces the micro-motion feature weight to 0.2 and increases the electromagnetic and thermal radiation feature weights to 0.4. The server performs feature-level fusion by calculating a weighted sum, but not a simple linear combination. Instead, it uses a kernel function mapping method to map the feature vectors to a high-dimensional space before fusion to capture the nonlinear relationships between features. The final generated multimodal liveness feature is a 93-dimensional comprehensive feature vector containing all the key information extracted from the three physical field signals.

[0057] like Figure 2 As shown, in one possible implementation, before acquiring the first physical field signal of the object to be detected, the method further includes steps S201-S204, as follows: Step S201: Obtain the background electromagnetic spectrum of the current detection environment, and identify the interference frequency bands and interference intensity in the background electromagnetic spectrum.

[0058] In step S201, the server performs a full-band scan of the current detection environment. The server controls the receiver to scan the frequency range from 100MHz to 6GHz in 10MHz increments, stopping at each frequency point for 100ms to measure the power spectral density. The server uses a Fast Fourier Transform (FFT) algorithm to perform spectral analysis on the acquired time-domain signal to obtain the power spectral density value at each frequency point. To improve measurement accuracy, the server performs 10 measurements at each frequency point and calculates the average value, while recording the standard deviation of the power spectral density to evaluate signal stability. The server uses a pre-defined interference identification algorithm to mark frequency bands with power spectral density exceeding -80dBm as potential interference bands. The server further analyzes the spectral characteristics of these frequency bands, identifying interference source types such as WiFi signals (2.4GHz and 5GHz bands), mobile communication signals (900MHz and 1800MHz bands), and industrial heating equipment (915MHz and 2450MHz) based on spectral width, center frequency, and modulation characteristics. The server stores the identified interference frequency bands and their corresponding interference intensities in an environmental electromagnetic database.

[0059] Step S202: Construct an interference feature matrix based on the interference frequency band and the interference intensity.

[0060] In step S202, the server constructs a multi-dimensional interference feature matrix based on the identified interference information. The server creates an N×M matrix structure, where N represents the number of identified interference frequency bands, and M represents the feature dimension of each band. For each interference frequency band, the server extracts eight key features: center frequency, bandwidth, peak power, average power, power fluctuation range, duty cycle, modulation type, and time stability. For example, for detected WiFi interference, the server records its center frequency as 2.437 GHz, bandwidth as 20 MHz, peak power as -45 dBm, average power as -52 dBm, power fluctuation range as ±5 dB, duty cycle as approximately 60%, modulation type as OFDM, and time stability as persistent. The server also calculates the correlation between interference frequency bands to identify possible intermodulation interference and harmonic interference. The server performs dimensionality reduction on the interference feature matrix using principal component analysis (PCA) to extract the most prevalent interference patterns, providing a simplified decision-making basis for subsequent anti-interference strategy formulation.

[0061] Step S203: Adaptively adjust the transmission frequency band of the electromagnetic modulation sensing unit using the interference feature matrix to avoid the interference frequency band and generate an anti-interference transmission strategy.

[0062] In step S203, the server intelligently adjusts the operating parameters of the electromagnetic modulation sensing unit based on the interference feature matrix. The server first analyzes the "clean" frequency bands within the 100MHz-6GHz band, i.e., continuous frequency bands with interference intensity below -90dBm. The server uses a spectrum hole detection algorithm to identify available frequency bands with at least 50MHz of continuous bandwidth. Based on the distribution of available frequency bands, the server re-plans the frequency sweeping scheme of the electromagnetic modulation sensing unit. For example, when strong WiFi interference is detected in the 2.4-2.5GHz range, the server adjusts the original 2.45GHz measurement point to 2.3GHz or 2.6GHz. The server not only adjusts the frequency point location but also optimizes the transmission power allocation strategy, appropriately increasing the transmission power in adjacent frequency bands with strong interference to improve the signal-to-noise ratio, while ensuring that the total transmission power complies with electromagnetic compatibility standards. The anti-interference transmission strategy generated by the server includes: a main measurement frequency list (30 frequencies to avoid interference), the transmission power of each frequency (adaptively adjusted from -10dBm to 0dBm), the frequency sweep order (prioritizing the measurement of frequency bands with less interference), and the integration time (extended to 200 milliseconds near the interference frequency band). The server sends these strategy parameters to the RF front-end of the electromagnetic modulation sensing unit through the control interface.

[0063] Step S204: Apply adaptive notch filtering to the echo signal of the life micro-motion sensing unit based on the interference feature matrix to suppress the noise component corresponding to the interference frequency band.

[0064] In step S204, the server performs precise interference suppression processing on the life micro-motion sensing unit. The server first analyzes the interference characteristic matrix to determine which interference frequency bands affect the intermediate frequency (IF) signal of the radar operating frequency band (typically 24 GHz or 77 GHz). The server designs an adaptive notch filter bank, with each notch filter targeting a specific interference frequency band. The server employs an infinite impulse response (IIR) filter structure, dynamically adjusting the notch frequency and notch depth by calculating the filter coefficients in real time. For example, when harmonic interference from industrial heating equipment at 2.45 GHz is detected affecting the 24.5 GHz radar signal, the server sets a notch filter with a center frequency of 50 MHz, a bandwidth of 2 MHz, and a notch depth of 40 dB during the IF processing stage. The server uses an adaptive algorithm to continuously monitor the filtering effect, evaluating the filtering performance by calculating the improvement in signal-to-noise ratio before and after filtering. If new interference is detected or the interference characteristics change, the server updates the filter parameters within 200 milliseconds. The server also implements a protection mechanism to ensure that notch filtering does not damage useful vital micro-signals. Spectrum analysis verifies that respiratory (0.1-0.5Hz) and heartbeat (0.8-2Hz) signal components are not attenuated.

[0065] Step S104: If it is determined that the multi-physics field signal does not meet the preset physiological behavior conditions, then within the preset second time window, the second physical field signal of the object to be detected is collected, and it is determined whether the second physical field signal meets the preset physiological recovery conditions.

[0066] In step S104, the preset second time window refers to an additional preset duration for supplementing signal acquisition, which is immediately activated after determining that the preset physiological behavior conditions are not met; the second physical field signal refers to the physical field signal acquired again within the preset second time window, the main goal of which is to capture the previously missing physiological signals; the preset physiological recovery condition refers to a logical criterion for determining whether the subject to be tested has recovered from a transient abnormal physiological state to a normal physiological state within the preset second time window.

[0067] Specifically, if the judgment result of step S102 is that the preset physiological behavior conditions are not met, it is usually due to the lack of vital micro-motion signals. The server will immediately initiate a compensation process, entering a preset second time window of, for example, 3 seconds. During this period, the server instructs the front-end detection terminal to continuously collect the second physical field signal, especially focusing on monitoring the data stream of vital micro-motion signals. The server will calculate the short-term energy of the signal in real time. The preset physiological recovery condition is set as follows: within the preset second time window, the energy of the vital micro-motion signal jumps from a state below the initial judgment minimum energy threshold to a state exceeding a preset recovery energy threshold. Once this jump is detected, it is determined that the preset physiological recovery condition is met. If the preset second time window ends and the energy has not jumped, it is determined that the condition is not met.

[0068] In one possible implementation, determining whether the second physical field signal meets the preset physiological recovery condition specifically includes: if it is determined that the vital micro-motion signal is less than the preset micro-motion threshold, and the electromagnetic modulation signal is within the electromagnetic confidence interval, and the thermal radiation characteristic signal is within the thermal radiation confidence interval, then extracting the target vital micro-motion signal from the second physical field signal; determining whether a jump signal greater than or equal to the preset micro-motion threshold appears in the target vital micro-motion signal; if it is determined that a jump signal greater than or equal to the preset micro-motion threshold appears in the target vital micro-motion signal, then determining that the second physical field signal meets the preset physiological recovery condition; if it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the jump signal does not exist in the target vital micro-motion signal, then determining that the second physical field signal does not meet the preset physiological recovery condition.

[0069] Specifically, after the server detects an abnormality in the micro-movement signal of life within the first time window, it immediately initiates a special detection process for physiological recovery characteristics. This process is designed based on the characteristic that organisms will naturally recover normal physiological activities after being subjected to brief disturbances.

[0070] The server first categorizes and analyzes the detection results from the first time window to determine if they fall under a specific case of single-mode anomaly. When the server detects that the vital signs signal is below a preset threshold, but the electromagnetic modulation signal and thermal radiation characteristic signal are normal, it determines that this may be a recoverable anomaly caused by temporary physiological inhibition or external vibration interference. For example, a user may experience sleep apnea due to tension, or the radar signal may be interfered with by surrounding mechanical vibrations. In this case, the server activates the enhanced acquisition mode of the second time window, increasing the radar sampling rate from 100Hz to 500Hz and enabling adaptive beamforming technology to precisely focus the detection beam on the chest region of the subject, minimizing environmental interference.

[0071] Within the second time window, the server performs refined processing on the collected target vital micro-motion signals. The server employs short-time Fourier transform technology, using a window length of 200 milliseconds and a step size of 50 milliseconds to perform time-frequency analysis on the signal, generating a high-time-resolution spectrogram. The server tracks the energy variation trajectories of the 0.1-0.5Hz respiratory band and the 0.8-2.0Hz heartbeat band within the spectrogram. By calculating the band energy within each time window, the server constructs time series of respiratory and heartbeat amplitudes. The server pays particular attention to the rising edge characteristics of these time series, using a first-order difference algorithm to calculate the rate of change of amplitude between adjacent time points.

[0072] For identifying surge signals, the server employs multiple judgment criteria. First, the server defines a surge signal as a dynamic process in which the amplitude of a micro-movement rapidly increases from below a threshold and stably exceeds the threshold within one second. The server uses a sliding window detection algorithm to identify monotonically increasing segments in the amplitude time series. When the respiratory amplitude increases from 1 mm to 3 mm within one second, or the heart rate amplitude increases from 0.3 mm to 0.8 mm, the server marks it as a candidate surge event. The server further verifies the stability of the signal after the surge, requiring that the micro-movement amplitude remain above the threshold for at least two seconds after the surge, and that the coefficient of variation is less than 20%. Only when both the rapid increase and stable maintenance conditions are met does the server confirm the detection of a valid surge signal.

[0073] The server monitors the persistence of electromagnetic modulation signals and thermal radiation characteristic signals in parallel. Within a second time window, the server updates the detection values ​​of these two signals every 500 milliseconds to ensure they remain within normal confidence intervals. If an abnormal drift in the dielectric constant of the electromagnetic modulation signal is detected, such as a sudden jump from 55 to 45 at the 3GHz frequency point, or if the thermal radiation signal shows a rapid temperature drop of more than 2 degrees Celsius, the server will immediately terminate the physiological recovery determination process, as such changes do not conform to normal physiological characteristics.

[0074] The server uses a state machine model to manage the entire decision-making process. The state machine includes three states: "Waiting for Recovery," "Recovering," and "Recovery Completed." The initial state is "Waiting for Recovery." When a micro-motion signal is detected to begin rising, the server transitions to the "Recovering" state. Only after confirming that the rising signal is valid and the other two signals remain normal does the server transition to the "Recovery Completed" state. If the "Recovery Completed" state is not reached within the 7-second second time window, or if other abnormalities are detected, the server directly outputs a decision result indicating that the preset physiological recovery conditions are not met.

[0075] For example, during the first time window detection, a user might have just entered from a cold outdoor environment, resulting in shallow breathing or, due to nervousness and breath-holding, micro-movements of only 1.2 mm. The server detects normal electromagnetic modulation signals (dielectric constant of 55.5 at 3GHz) and normal thermal radiation signals (facial temperature of 34.8 degrees Celsius and gradually rising), classifying it as a recoverable anomaly. In the first 3 seconds of the second time window, the user gradually adapts to the indoor environment, and their breathing deepens. The server detects that the breathing amplitude increases from 1.2 mm to 3.5 mm within 1.5 seconds, and the heart rate increases from 0.4 mm to 0.9 mm, remaining stable for the next 4 seconds. Simultaneously, the electromagnetic and thermal radiation signals remain normal, and the facial temperature stabilizes at 35.6 degrees Celsius. The server confirms the detection of a valid surge signal, determines that the second physical field signal meets the preset physiological recovery conditions, and allows entry into the subsequent liveness feature extraction process.

[0076] Step S105: If it is determined that the second physical field signal meets the preset physiological recovery conditions, the second physical field signal is analyzed to generate multimodal live features.

[0077] Specifically, if the server determines in step S104 that the second physical field signal meets the preset physiological recovery conditions, it indicates that the subject under test has only temporarily held its breath and then resumed normal breathing. To make a final decision, the server needs to generate a complete and valid set of multimodal liveness features. At this point, the server will call upon the electromagnetic modulation signal and thermal radiation characteristic signal from the first physical field signal that was collected and verified in step S101, and combine them with the vital micro-motion signals from the second physical field signal that has been confirmed as valid and collected within the preset second time window in step S104. The server performs the same refined analysis and feature extraction algorithm as in step S103 on this set of spliced ​​and verified data, thereby generating a high-quality multimodal liveness feature vector with the same structure and dimensions, and then passes this multimodal liveness feature to step S106.

[0078] Step S106: Based on the multimodal liveness features, determine whether the object to be detected is a living organism.

[0079] Specifically, after receiving the multimodal liveness feature vector from step S103 or S105, the server performs the final classification decision. The server has a pre-trained classifier deployed within it, such as a support vector machine, decision tree, or logistic regression classifier. The server inputs the multimodal liveness feature vector into this classifier. The classifier calculates the input vector based on its internal decision boundary or discriminant function and outputs a final classification label. This label is the final judgment result for the object to be detected: biological liveness or non-biological liveness. The server then outputs this judgment result to the upper-layer application system to complete subsequent business processes such as access control or payment confirmation.

[0080] To further illustrate this application, an example of the specific implementation of steps S101-S106 is given below. Suppose a user, feeling slightly nervous, unconsciously holds their breath for the first two seconds while using facial recognition access control. The server acquires the first physical field signal within a preset first time window. During initial assessment, it finds that the user's electromagnetic modulation signal and thermal radiation characteristic signal are within the normal range for a living organism, but the energy of the vital micro-motion signal is extremely low, failing to meet the preset physiological behavior conditions. At this point, the server does not immediately reject the user but initiates a compensation process, entering a preset second time window of 3 seconds for continuous acquisition. In the second second, the user relaxes and exhales, and the server immediately detects a significant jump in the energy of the vital micro-motion signal, meeting the preset physiological recovery conditions. Subsequently, the server combines the electromagnetic and thermal radiation signals acquired in the first acquisition with the effective respiratory signal acquired in the second acquisition to perform comprehensive feature extraction, generating a complete multimodal liveness feature vector. Finally, the server sends this feature vector to the decision system, ultimately correctly determining that the user is a living organism and sending an opening command to the access control system.

[0081] In one possible implementation, determining whether the object to be detected is a living organism based on the multimodal liveness features specifically includes: inputting the multimodal liveness features into a preset liveness discrimination model, which is generated based on live biological samples and non-live biological samples; classifying and calculating the multimodal liveness features using the liveness discrimination model and outputting a liveness confidence score; determining whether the liveness confidence score is greater than or equal to a preset liveness determination threshold; if the liveness confidence score is greater than or equal to the preset liveness determination threshold, then determining that the object to be detected is a living organism; if the liveness confidence score is less than the preset liveness determination threshold, then determining that the object to be detected is a non-living organism.

[0082] Specifically, the server performs the liveness detection task through a deep learning inference engine deployed on a GPU cluster, which loads a pre-set liveness detection model trained on large-scale data.

[0083] The pre-designed liveness detection model's architecture fully considers the heterogeneity and complementarity of multimodal features. The server-side model is a multi-branch fusion network, comprising three sub-networks specifically processing different modal features and a decision fusion layer. The electromagnetic feature branch uses a one-dimensional convolutional neural network to specifically learn the frequency domain patterns of the dielectric constant spectrum; the micro-motion feature branch uses a Long Short-Term Memory (LSTM) network to capture the temporal dependencies of physiological signals; and the thermal radiation feature branch uses a graph convolutional network to model the spatial topology of the temperature field. The outputs of the three branches are fused in a high-level semantic space, and the contribution of each modality is adaptively adjusted through an attention mechanism. This model is trained on a dataset containing 50,000 live biological samples and 30,000 non-live samples, with the non-live samples encompassing various deception methods such as silicone masks, 3D printed models, and highly realistic robots.

[0084] After receiving the 93-dimensional multimodal liveness features, the server first performs feature preprocessing. The server splits the feature vectors into electromagnetic feature sub-vectors (67 dimensions), micro-motion feature sub-vectors (11 dimensions), and thermal radiation feature sub-vectors (15 dimensions) according to their modal origin, and inputs them into the corresponding network branches. During forward propagation, the server maintains the numerical stability of the features through batch normalization to prevent gradient vanishing or exploding. After each branch network extracts high-level semantic features, the server calculates cross-modal attention weights at the fusion layer. These weights reflect the credibility of each modality in the current detection scenario. For example, when strong electromagnetic interference is detected, the model automatically reduces the weight of micro-motion features and increases the influence of electromagnetic and thermal radiation features.

[0085] The liveness confidence score is calculated using a probabilistic output method. The server uses a softmax activation function in the last layer of the model, mapping the network output to probability values ​​in the interval [0, 1]. This probability value represents the likelihood that the detected object is a living organism, i.e., the liveness confidence score. To improve the robustness of the discrimination, the server employs an ensemble learning strategy, simultaneously running five structurally similar but parameter-different model replicas. Each replica uses different data augmentation strategies and initialization parameters during training. The server calculates the weighted average of the five model outputs as the final liveness confidence score, with the weights dynamically adjusted based on the performance of each model on the validation set.

[0086] The preset liveness detection threshold was determined based on rigorous statistical analysis and practical application requirements. Before deployment, the server system underwent large-scale threshold optimization experiments, analyzing the confidence score distribution of 10,000 independent test samples and plotting Receiver Operating Characteristic (ROC) curves. The server selected the threshold corresponding to the equal error rate (EER) point as a benchmark, where the false acceptance rate and false rejection rate are equal. Considering security requirements, the server increased the threshold by 10% from the EER, setting it to 0.85. This means that the system only determines a living organism when the liveness confidence score reaches or exceeds 0.85. The server also implemented a dynamic threshold adjustment mechanism, automatically adjusting the threshold according to the security level of the application scenario; in high-security scenarios, the threshold can be increased to 0.95.

[0087] The server's decision-making process incorporates multiple verification mechanisms. When the calculated liveness confidence score approaches a threshold (e.g., between 0.83 and 0.87), the server initiates a secondary verification process. The server analyzes the feature activation map within the model to check which features contribute most to the final score. If a feature of a particular modality is found to be abnormally prominent or abnormally weak, the server generates an interpretability report to help identify potential attack patterns. The server also maintains a decision log database, recording detailed information for each decision, including feature values ​​for each modality, intermediate layer activations, attention weights, and the final score, for subsequent model optimization and anomaly detection.

[0088] For example, when detecting a highly realistic silicone mask, the multimodal liveness features extracted by the server showed: the real part of the dielectric constant in the electromagnetic features was 48.5 at 3 GHz, slightly lower than that of normal skin tissue; the micromotion features were completely absent, with both respiratory and heartbeat signal amplitudes at 0; the thermal radiation features showed an overly uniform temperature distribution, lacking temporal fluctuations caused by blood flow pulsation. The server input these features into the liveness detection model, with the electromagnetic feature branch outputting 0.3, the micromotion feature branch outputting 0.0, and the thermal radiation feature branch outputting 0.2. After calculation by the fusion layer, the final liveness confidence score was only 0.15, far below the preset liveness detection threshold of 0.85. Therefore, the server determined that the object to be detected was not a living organism and rejected it from the liveness detection.

[0089] Conversely, for a real user, even with a slight fever due to a mild cold, the features extracted by the server still showed: normal electromagnetic characteristics (dielectric constant 54.8 at 3GHz), clear micro-motion characteristics (18 breaths / minute respiration, 80 beats / minute heart rate), and although the temperature was slightly elevated, the thermal radiation characteristics still showed a normal blood flow pulsation pattern. The model calculated a liveness confidence score of 0.92, exceeding the threshold, and the server determined it to be a living biological entity. This indicates that the system has good tolerance for normal physiological changes and can effectively identify various spoofing attacks.

[0090] The following describes a biological liveness detection device applied in a strong electromagnetic interference environment from the perspective of hardware processing, according to an embodiment of this invention. Please refer to [link / reference needed]. Figure 3 This is a schematic diagram of a biological liveness detection device applied in a strong electromagnetic interference environment according to an embodiment of this application.

[0091] It should be noted that, Figure 3 The structure of the biological liveness detection device shown is merely an example and should not impose any limitations on the functionality and scope of application of the embodiments of the present invention.

[0092] like Figure 3As shown, a biological liveness detection device for use in environments with strong electromagnetic interference includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a Read-Only Memory (ROM) 302 or a program loaded from a storage section 308 into a Random Access Memory (RAM) 303, such as executing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for device operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0093] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0094] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0095] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0097] Specifically, a biological liveness detection device for use in a strong electromagnetic interference environment according to this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the biological liveness detection method for use in a strong electromagnetic interference environment provided in the above embodiment.

[0098] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the biological liveness detection device for use in a strong electromagnetic interference environment described in the above embodiments; or it may exist independently and not assembled into the biological liveness detection device for use in a strong electromagnetic interference environment. The storage medium carries one or more computer programs, which, when executed by a processor of the biological liveness detection device for use in a strong electromagnetic interference environment, cause the biological liveness detection device for use in a strong electromagnetic interference environment to implement the biological liveness detection method for use in a strong electromagnetic interference environment provided in the above embodiments.

Claims

1. A method for detecting living organisms in a strong electromagnetic interference environment, characterized in that, The method includes: Within a preset first time window, the first physical field signal of the object to be detected is collected. The first physical field signal includes electromagnetic modulation signal, micro-movement signal of life and thermal radiation characteristic signal. Determine whether the first physical field signal meets the preset physiological behavior conditions; If it is determined that the first physical field signal meets the preset physiological behavior conditions, then the first physical field signal is analyzed to generate multimodal live features; If it is determined that the multi-physics field signal does not meet the preset physiological behavior conditions, then within the preset second time window, the second physical field signal of the object to be detected is collected, and it is determined whether the second physical field signal meets the preset physiological recovery conditions. If it is determined that the second physical field signal meets the preset physiological recovery conditions, then the second physical field signal is analyzed to generate multimodal live cell features; Based on the multimodal liveness characteristics, it is determined whether the object to be detected is a living organism.

2. The method of claim 1, wherein, The acquisition of the first physical field signal of the object to be detected specifically includes: An electromagnetic modulation sensing unit is used to transmit swept electromagnetic waves to the object under test and to receive the reflected signal modulated by the object under test. Based on the amplitude ratio and phase difference between the swept electromagnetic wave and the reflected signal, an electromagnetic modulation signal characterizing the complex permittivity spectrum of the object to be detected is generated. A life micro-motion sensing unit is used to transmit a linear frequency modulated signal to the object to be detected and receive the echo signal. The echo signal is processed to obtain a range Doppler spectrum, and the respiratory component and the heartbeat component are separated from the range Doppler spectrum. The vital micro-motion signal is calculated based on the respiratory component and the heartbeat component. The object to be detected is scanned using a thermal radiation characteristic sensing unit to collect the thermal radiation intensity distribution of the object to be detected within a preset wavelength band. The thermal radiation intensity distribution is converted into a temperature field, and the spatiotemporal changes of the temperature field are recorded to generate the thermal radiation characteristic signal.

3. The method of claim 2, wherein, Before acquiring the first physical field signal of the object to be detected, the method further includes: Obtain the background electromagnetic spectrum of the current detection environment, and identify the interference frequency bands and interference intensity in the background electromagnetic spectrum; Based on the interference frequency band and the interference intensity, an interference feature matrix is ​​constructed; The transmission frequency band of the electromagnetic modulation sensing unit is adaptively adjusted using the interference feature matrix to avoid the interference frequency band and generate an anti-interference transmission strategy. Based on the interference feature matrix, an adaptive notch filter is applied to the echo signal of the life micro-motion sensing unit to suppress the noise components corresponding to the interference frequency band.

4. The method of claim 1, wherein, The determination of whether the first physical field signal meets the preset physiological behavior conditions specifically includes: Determine whether the electromagnetic modulation signal is within the electromagnetic confidence interval, determine whether the thermal radiation characteristic signal is within the thermal radiation confidence interval, and determine whether the life micro-motion signal is greater than or equal to a preset micro-motion threshold. If it is determined that the electromagnetic modulation signal is within the electromagnetic confidence interval, the thermal radiation characteristic signal is within the thermal radiation confidence interval, and the vital micro-motion signal is greater than or equal to the preset micro-motion threshold, then it is determined that the multimodal data features satisfy the preset physiological behavior conditions. If it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the vital micro-motion signal is less than the preset micro-motion threshold, then it is determined that the multimodal data features do not meet the preset physiological behavior conditions.

5. The method of claim 1, wherein, The determination of whether the second physical field signal meets the preset physiological recovery conditions specifically includes: If it is determined that the micro-motion signal of life is less than the preset micro-motion threshold, and the electromagnetic modulation signal is within the electromagnetic confidence interval, and the thermal radiation characteristic signal is within the thermal radiation confidence interval, then the target micro-motion signal of life in the second physical field signal is extracted. Determine whether a jump signal greater than or equal to the preset micro-motion threshold appears in the target life micro-motion signal; If it is determined that a jump signal greater than or equal to the preset micro-motion threshold appears in the target life micro-motion signal, then it is determined that the second physical field signal satisfies the preset physiological recovery condition. If it is determined that the electromagnetic modulation signal is not within the electromagnetic confidence interval, or the thermal radiation characteristic signal is not within the thermal radiation confidence interval, or the target life micro-motion signal does not contain the jump signal, then it is determined that the second physical field signal does not meet the preset physiological recovery condition.

6. The method of claim 1, wherein, The step of analyzing the first physical field signal to generate multimodal liveness features specifically includes: Feature extraction is performed on the electromagnetic modulation signal to obtain the real part of the dielectric constant and the loss factor features of the object to be detected within a preset frequency band, and the real part of the dielectric constant and the loss factor features are combined into an electromagnetic feature vector. Feature extraction is performed on the vital micro-motion signals to obtain respiratory rate, heart rate, respiratory amplitude and heart rate variability parameters, and the respiratory rate, heart rate, respiratory amplitude and heart rate variability parameters are combined into a micro-motion feature vector; Feature extraction is performed on the thermal radiation characteristic signal to obtain the spatial gradient features and temporal fluctuation features of the temperature field, and the spatial gradient features and the temporal fluctuation features are combined into a thermal radiation feature vector. The electromagnetic feature vector, the micro-motion feature vector, and the thermal radiation feature vector are weighted and fused according to preset weights to generate the multimodal liveness feature.

7. The method of claim 6, wherein, The step of determining whether the object to be detected is a living organism based on the multimodal liveness features specifically includes: The multimodal liveness features are input into a preset liveness discrimination model, which is generated based on training on biological live samples and non-live samples. The liveness detection model is used to classify and calculate the multimodal liveness features, and the liveness confidence score is output. Determine whether the liveness confidence score is greater than or equal to a preset liveness determination threshold; If the liveness confidence score is determined to be greater than or equal to the preset liveness determination threshold, then the object to be detected is determined to be a living organism. If the liveness confidence score is determined to be less than the preset liveness determination threshold, then the object to be detected is determined to be a non-living organism.

8. A biological liveness detection device for use in environments with strong electromagnetic interference, characterized in that, The biological liveness detection device for use in strong electromagnetic interference environments includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors invoke the computer instructions to cause the biological liveness detection device for use in strong electromagnetic interference environments to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on a biological liveness detection device applied to a strong electromagnetic interference environment, the biological liveness detection device applied to a strong electromagnetic interference environment performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on a biological liveness detection device applied to a strong electromagnetic interference environment, the biological liveness detection device applied to a strong electromagnetic interference environment performs the method as described in any one of claims 1-7.

Images