A multi-target life detection method and system based on radar signals

Abstract

The application relates to the technical field of life detection, and discloses a multi-target life detection method based on radar signals, which comprises the following steps: step one, collecting a multi-target mixed signal received by a radar and pre-processing the signal; step two, reconstructing a phase space of the pre-processed signal to obtain a high-dimensional phase space trajectory; step three, constructing an adaptive dictionary and performing sparse representation on the signal; and step four, separating the preliminary vital sign signals of each target from the mixed signal through a sparse optimization algorithm.The technical scheme of fusing the multi-frequency radar signal and deep learning is adopted, so that the technical effect of high-precision separation of multi-target vital sign signals in a complex electromagnetic environment is achieved.Compared with the technical scheme of relying on a single frequency band or fixed threshold classification in the prior art, the technical scheme solves the problems of serious signal aliasing and low separation precision in the dense obstacle or multi-target mixed scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of life detection, in particular to a multi-target life detection method and system based on radar signals. BACKGROUND

[0002] Radar signal processing technology has important application value in the field of vital sign monitoring, especially in natural disaster rescue, medical health care and military reconnaissance scenes, which can provide non-contact efficient detection means. By analyzing the vital sign signals (such as breathing frequency, heartbeat rate) in the radar reflection echo, remote detection and positioning of the target can be realized. However, with the complication of application scenarios, especially in multi-target dense areas and complex electromagnetic environments, the existing technology faces serious signal aliasing, low positioning accuracy and other problems, which is difficult to meet the actual needs.

[0003] At present, the mainstream radar life detection technology is mainly based on synthetic aperture radar (SAR) and Doppler radar principle. These technologies extract vital sign signals by analyzing the reflected echo, and show good performance in single target or simple environment. For example, Doppler radar can effectively extract breathing and heartbeat signals by detecting the frequency shift caused by target motion; SAR technology improves the resolution through synthetic aperture processing to realize the detection of stationary targets. In addition, existing technologies usually use Fourier transform or fixed threshold classifier to process signals to realize target separation and positioning.

[0004] The existing technology is particularly prominent in the problem of signal aliasing in complex electromagnetic environment. In multi-target dense areas, radar signals are easily affected by multipath effect and noise interference, resulting in aliasing of vital sign signals of multiple targets. The processing ability of traditional methods (such as Fourier transform) for nonlinear and non-stationary signals is limited, which cannot effectively separate the spectrally overlapping signals; and the fixed threshold classifier has high misjudgment rate in time-varying environment, which is difficult to accurately distinguish the vital sign signals of different individuals. These problems seriously limit the application effect of existing technology in complex scenes, and a new technical solution is needed to effectively solve the problem of signal aliasing. SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides a multi-target life detection method and system based on radar signals, which can effectively solve the problem of serious aliasing of multi-target vital sign signals in complex electromagnetic environment.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a multi-target life detection method based on radar signals, comprising the following steps:

[0007] Step one, collect the mixed signal of multiple targets received by the radar, and pretreat the signal;

[0008] Step two, reconstruct the phase space of the pre-processed signal to obtain high-dimensional phase space trajectory;

[0009] Step three, construct an adaptive dictionary and perform sparse representation on the signal;

[0010] Step four, separate the preliminary vital signs of each target from the mixed signal through a sparse optimization algorithm;

[0011] Step five, use a nonlinear dynamic model to correct the preliminary separation signal to eliminate residual aliasing and obtain the final separation signal;

[0012] Step six, based on the final separation signal, calculate the target position and extract vital signs information;

[0013] Step seven, output the target position information and vital sign signal.

[0014] Preferably, the pre-processing in step one includes filtering the signal to remove high-frequency noise and baseline drift, and normalizing the signal.

[0015] Preferably, the adaptive dictionary in step three is constructed by the K-SVD algorithm, specifically including: optimizing the dictionary atoms and sparse coefficients by minimizing the reconstruction error of the linear combination of the signal and the dictionary atoms.

[0016] Preferably, the sparse optimization algorithm in step four is the orthogonal matching pursuit algorithm, specifically including: selecting dictionary atoms by iteration to gradually approach the optimal sparse representation of the signal.

[0017] Preferably, the nonlinear dynamic model in step five is used to describe the coupling relationship between signals, specifically including: eliminating residual aliasing in the preliminary separation signal through a correction function.

[0018] Preferably, the target position calculation in step six is implemented by the time reversal localization algorithm, specifically including: calculating the spatial position of the target by constructing the time reversal channel response function.

[0019] A multi-target life detection system based on radar signal, comprising:

[0020] A signal acquisition module for acquiring a mixed signal of multiple targets received by the radar, the output end of which is connected to a signal preprocessing module;

[0021] A signal preprocessing module for filtering and normalizing the signal, the output end of which is connected to a phase space reconstruction module;

[0022] A phase space reconstruction module for reconstructing the phase space of the pre-processed signal, the output end of which is connected to a dictionary construction module;

[0023] a dictionary construction module for constructing an adaptive dictionary and sparsely representing the signal, which is connected to the signal separation module;

[0024] a signal separation module for separating the preliminary vital sign signals of each target through a sparse optimization algorithm, which is connected to the signal correction module;

[0025] a signal correction module for eliminating residual aliasing using a nonlinear dynamic model, which is connected to the target positioning module;

[0026] a target positioning module for calculating the target position and extracting vital sign information, which is connected to the result output module;

[0027] a result output module for outputting the target position information and vital sign signals

[0028] Preferably, the signal separation module includes an orthogonal matching pursuit algorithm unit for gradually approaching the optimal sparse representation of the signal by iteratively selecting dictionary atoms.

[0029] Preferably, the target positioning module includes a time reversal positioning algorithm unit for calculating the spatial position of the target by constructing a time reversal channel response function.

[0030] The present application provides a multi-target life detection method and system based on radar signals. It has the following beneficial effects:

[0031] 1. The present application adopts a technical solution of multi-band radar signal and deep learning fusion, achieving high-precision separation of multi-target vital sign signals in complex electromagnetic environments. Compared with the technical solution of relying on a single frequency band or fixed threshold classification in the prior art, it solves the problems of serious signal aliasing and low separation precision in dense obstacle or multi-target mixed scenes.

[0032] 2. The present application adopts a technical solution of time reversal positioning algorithm and multi-sensor fusion, achieving sub-meter positioning accuracy at a long distance. Compared with the technical solution of traditional radar positioning error in the prior art, it solves the problem of insufficient positioning accuracy in dense occlusion environment.

[0033] 3. The present application adopts a technical solution of dynamic adaptive filtering and noise suppression model, achieving stable operation of the system in strong noise background. Compared with the technical solution of fixed filtering or single noise processing in the prior art, it solves the problems of high false alarm rate and poor system stability in high noise environment.

[0034] 4. This invention employs a layered processing framework and GPU-accelerated computing, achieving an average processing latency of less than 2 seconds. Compared to existing technologies with complex processing flows and long response times, this invention solves the problem of failing to meet real-time requirements in scenarios with high demands. Attached Figure Description

[0035] Fig. 1 This is a schematic diagram of the method flow of the present invention;

[0036] Fig. 2 This is a schematic diagram of the system flow of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see the appendix Figs. 1-2 This invention provides a multi-target life detection method and system based on radar signals, comprising:

[0039] Signal acquisition and preprocessing, as the foundational steps of the entire technical solution, must ensure the quality and consistency of the input signal to provide a reliable data foundation for subsequent key processing such as phase space reconstruction and sparse representation.

[0040] In this embodiment, the specific implementation of signal acquisition and preprocessing is as follows:

[0041] A multi-band composite radar system is employed for signal acquisition, with the X-band (8-12 GHz) used for high-resolution surface target detection and the L-band (1-2 GHz) for obstacle penetration detection. The radar transmitter utilizes multiple-input multiple-output (MIMO) technology to transmit time-frequency-space coded pulse signals, with an adjustable pulse width ranging from 10 ns to 1 μs and an adaptive pulse repetition frequency (PRF) range from 1 kHz to 10 MHz to meet detection requirements in various scenarios. The receiver is equipped with a 16-channel MIMO array, achieving a noise figure of less than 2 dB and a dynamic range greater than 80 dB, ensuring the capture of weak vital signs even in environments with strong interference.

[0042] The signal preprocessing includes filtering and normalization. For high-frequency noise suppression, a band-pass filter with a passband range of 0.1 Hz to 5 Hz is used to retain the effective frequency band of the respiratory and heartbeat signals. Baseline drift removal is achieved through wavelet transform, for example, using db4 wavelet basis function for 5-layer decomposition to remove low-frequency drift components in the signal. Normalization is achieved by the following formula:

[0043]

[0044] wherein, is the original signal; is the signal mean; is the signal standard deviation, is the normalized signal. This processing makes the signal amplitudes of different channels and frequency bands uniform to the same order of magnitude, avoiding amplitude deviation problems in subsequent processing.

[0045] To adapt to complex electromagnetic environments, adaptive noise cancellation technology can be further introduced. For example, by collecting environmental noise signals through a reference channel, the least mean square algorithm is used to update the filter coefficients in real time, and the noise components are subtracted from the main signal. Its update formula is:

[0046]

[0047] wherein, is the filter weight vector, is the step factor, is the error signal, is the reference noise signal. Through such methods, the signal-to-noise ratio can be significantly improved, providing a more pure input signal for subsequent processing.

[0048] Radar system parameters can be dynamically adjusted according to actual scenarios. For example, when penetrating concrete walls, the L-band is preferred, while when high-resolution positioning is needed, the X-band is switched to. In addition, the signal sampling rate is set to 2GSPS (gigasamples per second) to ensure complete capture of high-frequency signals. The preprocessed signal is transmitted to the processing unit through a high-speed optical fiber interface to avoid signal attenuation and interference during transmission.

[0049] In some embodiments, for the special needs of multi-target scenarios, a preliminary screening of blind source separation technology can be introduced in the preprocessing stage. For example, through independent component analysis, the mixed signal is roughly separated, and potential vital sign signal components are screened out, and then processed by subsequent modules for refinement. Such extended technology can further improve the robustness of the system in multi-target dense scenarios.

[0050] ​​After the signal acquisition and preprocessing are completed, the application further converts the one-dimensional time series signal into a trajectory in a high-dimensional space through phase space reconstruction technology, so as to provide more abundant structural information for subsequent signal separation and feature extraction. Phase space reconstruction is a key step for solving the problem of nonlinear signal aliasing, and the core is to map the dynamic characteristics of the signal to a high-dimensional space through embedding theorem, so as to reveal the internal law of the signal.

[0051] In the embodiment, the specific implementation mode of phase space reconstruction is as follows:

[0052] Firstly, the delay time is determined by the autocorrelation function method The calculation formula of the autocorrelation function is as follows:

[0053] ;

[0054] Wherein, x (t) is the preprocessed signal, is the signal length, is the delay time. The first zero point of the autocorrelation function is selected as the optimal delay time to ensure that the trajectory of the reconstructed phase space can be fully unfolded. The determination of embedding dimension m is realized by the false nearest neighbor method. The specific steps include: gradually increasing the embedding dimension, calculating the false nearest neighbor ratio, and stopping increasing when the ratio is less than a preset threshold (for example, 5%). This method can effectively avoid the overfolding or redundancy of the phase space trajectory.

[0055] In some embodiments, according to Takens embedding theorem, the preprocessed signal is reconstructed in phase space to obtain a high-dimensional phase space trajectory

[0056] : ;

[0057] Wherein, m is the embedding dimension, is the delay time. Through phase space reconstruction, the dynamic characteristics of the signal are mapped to a high-dimensional space, which provides a more intuitive geometric structure for subsequent signal separation.

[0058] In order to further improve the reconstruction effect, the local projection method can be introduced to smooth the phase space trajectory. For example, the high-dimensional trajectory is projected to a low-dimensional space through the local linear embedding (LLE) algorithm, which retains the main dynamic characteristics of the signal and removes the trajectory jitter caused by noise. The objective function is:

[0059] ;

[0060] ;

[0061] ​​wherein, is a projection point in a low-dimensional space; is a local weight coefficient; is a number of near neighbor points.

[0062] In some embodiments, for a multi-target scene in a complex electromagnetic environment, a multi-scale phase space reconstruction technique can be further introduced. The signal is decomposed into multiple scale components through wavelet transform, and phase space reconstruction is performed respectively, and finally the reconstruction results of each scale are fused. This method can effectively capture the multi-scale characteristics of the signal and improve the accuracy of subsequent signal separation.

[0063] In some embodiments, to verify the effect of phase space reconstruction, the chaotic characteristics of the signal can be evaluated by calculating the maximum Lyapunov exponent. The Lyapunov exponent is calculated by the Wolf algorithm:

[0064] ;

[0065] wherein, is the distance between two points on the phase space trajectory, is the initial distance. If the Lyapunov exponent is positive, it indicates that the signal has chaotic characteristics and is suitable for processing using nonlinear dynamics methods.

[0066] The results of phase space reconstruction can be displayed through visualization technology. The high-dimensional trajectory is projected into a three-dimensional space to observe its geometric structure and dynamic evolution law. Such visualization means not only helps to understand the signal characteristics, but also provides intuitive basis for the optimization of subsequent algorithm parameters.

[0067] Through the above phase space reconstruction technology, the present application can convert a one-dimensional time series signal into a trajectory in a high-dimensional space, laying a solid foundation for subsequent signal separation and feature extraction.

[0068] After completing the phase space reconstruction, the present application converts the high-dimensional phase space trajectory into a sparse coefficient combination by constructing an adaptive dictionary and performing sparse representation of the signal, providing a mathematical basis for subsequent signal separation. This step aims to utilize the sparsity characteristics of the signal to convert the complex aliasing problem into an analyzable sparse optimization problem, thereby effectively separating the multi-target vital sign signals.

[0069] In some embodiments, the K-SVD algorithm is used to construct an adaptive dictionary . This algorithm minimizes the signal reconstruction error by alternately optimizing the dictionary atoms and sparse coefficients. Its objective function is:

[0070] ;

[0071] wherein, is the high-dimensional trajectory after phase space reconstruction, Let be the dictionary matrix to be optimized. It is a sparse coefficient vector. As a sparsity constraint, it controls the maximum number of non-zero elements in the coefficients. This represents the total number of signal samples.

[0072] In some embodiments, dictionary initialization may use a random matrix or predefined basis functions (such as Gabor basis, wavelet basis). Initial Dictionary The atoms are generated by a random Gaussian distribution and optimized iteratively. Each iteration consists of two stages: sparse encoding and dictionary update.

[0073] Sparse coding stage: fixed dictionary The sparse coefficients are solved using the Orthogonal Matching Pursuit (OMP) algorithm. ;

[0074] Dictionary update phase: Fixed sparsity coefficient Update dictionary atoms column by column The combination of atoms and their corresponding coefficients is optimized through singular value decomposition (SVD).

[0075] In some embodiments, the dictionary atomic update formula is:

[0076] ;

[0077] in, This indicates the removal of the current atom. The residual matrix after It is the Frobenius norm. This represents the row vector corresponding to the sparse coefficients. It is decomposed using SVD. The updated atoms Pick The first column, the sparsity coefficients are updated to .

[0078] In some embodiments, to improve the dictionary's adaptability to vital sign signals, prior knowledge of physiological characteristics can be introduced. Typical waveforms of respiration and heartbeat signals (such as sine waves and Morlet wavelets) can be used as initial dictionary atoms, which are then fine-tuned using the K-SVD algorithm. Such methods can accelerate algorithm convergence and improve the interpretability of sparse representations.

[0079] In some embodiments, to address the complexity of multi-objective scenarios, a block-based dictionary construction strategy can be employed. For example, the phase space trajectory can be divided into multiple sub-blocks, each with its own sub-dictionary, and global optimization can be achieved through joint sparse constraints. The objective function can then be extended as follows:

[0080] ;

[0081] wherein, is a sub-signal, is a corresponding sub-dictionary, is a global sparsity constraint.

[0082] In some embodiments, the result of sparse representation can be evaluated by residual analysis. The reconstructed signal is calculated by the original signal and the mean square error (MSE) is calculated as:

[0083] ;

[0084] When the MSE is lower than a preset threshold (e.g., 10 −4 ), it is determined that the sparse representation reaches the convergence criterion.

[0085] In some embodiments, to further enhance the robustness, a noise-aware sparse representation model can be introduced. For example, a norm constraint of the noise term is added in the objective function:

[0086] wherein, is a regularization parameter, balancing the reconstruction error and sparsity. Such improvement can effectively suppress the influence of noise on sparse coefficients, and improve the stability of signal separation.

[0087] Through the above dictionary construction and sparse representation technology, the present application can convert high-dimensional phase space trajectories into sparse coefficient combinations, providing a mathematical basis for subsequent signal separation, while adapting to the needs of multi-target detection in complex electromagnetic environments.

[0088] Nonlinear dynamics correction: After completing the sparse signal separation, the present application corrects the preliminary separated signals through a nonlinear dynamics model, eliminates residual aliasing, and obtains the final separated signals. This step aims to utilize the nonlinear dynamics characteristics to further optimize the signal separation results and improve the extraction accuracy of multi-target vital sign signals.

[0089] In this embodiment, the specific implementation of nonlinear dynamics correction is as follows:

[0090] In some embodiments, a nonlinear dynamics model is first constructed to describe the coupling relationship between signals. For example, a nonlinear coupling function is used to describe the interaction between target signals, which includes linear, quadratic, and cubic terms, and the coupling coefficients are determined by least squares fitting.

[0091] The preliminary separated signals are corrected using the nonlinear dynamics model. The specific steps are as follows:

[0092] Calculate the coupling signal: according to the nonlinear coupling function, calculate the coupling influence of each target signal on other signals;

[0093] Residual aliasing elimination: Subtract the coupling signal from the preliminary separated signal to obtain the corrected signal. This method can effectively eliminate residual aliasing between signals and improve separation accuracy.

[0094] To improve the correction accuracy, an adaptive coupling coefficient updating mechanism can be introduced. The coupling coefficient is dynamically adjusted by gradient descent method to minimize the correction error. This method can optimize the correction effect in real time according to the signal characteristics.

[0095] For the complexity of multi-target scenarios, a phased correction strategy can be used. The signal is divided into multiple frequency bands, and nonlinear dynamic correction is performed on each frequency band. Finally, the correction results of each frequency band are fused. This method can effectively handle the nonlinear aliasing problem of the signal and improve the separation accuracy. To further enhance the robustness, a noise suppression correction model can be introduced. In the correction process, the norm constraint of the noise term is added to balance the correction error and noise suppression. This method can effectively suppress the influence of noise on the correction result and improve the system stability. To verify the effect of nonlinear dynamic correction, the signal separation accuracy can be evaluated by spectral analysis. For example, the spectral difference between the signals before and after correction is calculated to ensure that the main vital sign frequency components are preserved while eliminating the pseudo-frequency components introduced by aliasing.

[0096] Through the above nonlinear dynamic correction, the present application can further optimize the signal separation result, eliminate residual aliasing, and improve the extraction accuracy of multi-target vital sign signals, providing reliable input for subsequent target positioning and vital sign extraction.

[0097] Target positioning and vital sign extraction: After completing the nonlinear dynamic correction, the present application calculates the target position and extracts the vital sign signal through a target positioning algorithm. This step aims to use the corrected separated signal to achieve high-precision target positioning and vital sign information extraction, providing key data for the final result output.

[0098] In this embodiment, the specific implementation of target positioning and vital sign extraction is as follows:

[0099] In some embodiments, a time reversal positioning algorithm (TIMLE) is used to calculate the target position. This algorithm constructs a time reversal channel response function and combines the time reversal characteristics of multipath signals to accurately calculate the spatial coordinates of the target. The channel impulse response is estimated by the correlation between the received signal and the reference signal, and the target position is finally determined by time reversal processing.

[0100] To improve the positioning accuracy, a multi-sensor fusion technology is introduced. The positioning results of multiple radar nodes are weighted and fused to obtain the optimal target position by dynamically adjusting the weight coefficient according to the signal-to-noise ratio or distance. This method can effectively reduce the positioning error of a single node and improve the system robustness.

[0101] The extraction of vital signs signals is achieved through spectral analysis. The modified separated signals are subjected to fast Fourier transform, and the peak frequency in the spectrum is detected to determine the respiratory and heartbeat frequencies. By setting frequency band thresholds (such as respiratory band 0.1-0.5 Hz and heartbeat band 0.8-2 Hz), noise interference is excluded, and the effectiveness of the extraction results is ensured.

[0102] Adaptive filtering technology is introduced to smooth the spectrum. For example, a Kalman filter is used to dynamically update state estimation, remove noise-induced false frequency components, and retain true vital sign frequencies. This method can adjust the filtering parameters in real time according to the signal characteristics, and adapt to changes in complex electromagnetic environments. To address the complexity of multi-target scenarios, a frequency band extraction strategy is used. The signal is divided into different frequency bands, and the respiratory and heartbeat features are extracted respectively, and the final vital sign information is determined through joint optimization. This method can avoid cross interference between frequency bands and improve the extraction accuracy in multi-target scenarios.

[0103] Through the above target positioning and vital sign extraction technology, the present application can realize high-precision target positioning and vital sign information extraction, providing reliable data support for the final result output.

[0104] Result output and visualization: After completing target positioning and vital sign extraction, the present application presents target position information and vital sign signals to users in an intuitive form through the result output and visualization module. This step aims to convert complex technical processing results into easily understandable visual information to support rescue or care decisions. Through efficient visualization design, users can quickly obtain key information, improving the practicality and user experience of the system.

[0105] In this embodiment, the specific implementation of result output and visualization is as follows:

[0106] Target position information is visualized through two-dimensional or three-dimensional maps. The target position is mapped to the geographic coordinate system to generate a heat map or point distribution map. The heat map visually displays the target dense area by calculating the spatial distribution of target signal strength. In specific implementation, the heat map intensity is calculated based on the target signal strength and position coordinates A smooth hotspot distribution is generated through a Gaussian kernel function. Users can quickly identify target dense areas through color depth, facilitating quick positioning of key areas in the rescue site.

[0107] The vital signs signals are visualized through time series plots or frequency spectra. The respiratory and heartbeat signals are plotted as time series separately, with the main frequency components labeled. The time series plots can visually show the trend of the target's vital signs, making it easier for users to analyze the target's life status. The frequency spectra are generated through Fast Fourier Transform, highlighting the respiratory and heartbeat frequencies, helping users quickly identify the target's physiological characteristics. For example, the respiratory frequency is usually between 0.1 Hz and 0.5 Hz, while the heartbeat frequency is between 0.8 Hz and 2 Hz. By setting frequency band thresholds, noise interference can be excluded, ensuring the effectiveness of the extraction results.

[0108] A dynamic updating mechanism is introduced to ensure the real-time nature of the visualization results. The target position and vital signs signals are updated every fixed time interval (e.g., 1 second) to reflect the latest status of the target. This method can meet the needs of real-time monitoring, especially in scenarios such as disaster rescue or military reconnaissance that require high timeliness. The dynamic updating mechanism is achieved through asynchronous communication between background data processing and front-end display, ensuring a smooth and responsive user interface.

[0109] To address the complexity of multi-target scenarios, a hierarchical visualization strategy is adopted. The targets are divided into multiple layers based on signal strength or distance, and displayed with different colors or markers. By setting signal strength thresholds, high-priority targets and low-priority targets can be distinguished, making it easier for users to quickly locate key targets. Targets with signal strength above the threshold are marked in red, indicating that they need to be rescued first; targets with medium signal strength are marked in yellow, indicating that they need to be further confirmed; and targets with low signal strength are marked in green, indicating that they have low priority. This hierarchical design can help users make quick decisions in complex scenarios.

[0110] Interactive visualization functions are introduced to enhance the flexibility of user operations. Users can view detailed information (such as target distance, vital sign frequency, etc.) by clicking on the target marker, or view historical data by sliding the time axis. This method can meet the diverse needs of users and improve the practicality of the system. Interactive functions are implemented through front-end frameworks (such as React or Vue.js), ensuring a responsive and smooth user interface. In addition, users can adjust the map display range through scaling, dragging, and other operations to further optimize the user experience.

[0111] For multi-target scenarios in complex electromagnetic environments, multi-modal fusion visualization technology is introduced. Radar detection results are fused with other sensor data such as infrared images and video streams to provide more comprehensive target information. This method can compensate for the limitations of single sensors and improve the accuracy of target identification. In disaster rescue scenarios, radar detection results can be overlaid with infrared images taken by unmanned aerial vehicles to help rescue personnel more accurately locate trapped individuals. Multi-modal fusion is achieved through data synchronization and coordinate alignment techniques to ensure consistency in time and space for different sensor data.

[0112] To further enhance the visualization effect, data compression and optimization techniques can be introduced. Principal component analysis (PCA) can be used to reduce data dimensionality and reduce the computational load of front-end rendering, ensuring smooth display in large-scale data scenarios. In addition, WebGL and other graphics rendering technologies can be used to improve the rendering efficiency of maps and charts, providing users with a smoother interactive experience.

[0113] Through the above result output and visualization technology, the invention can convert complex technical processing results into intuitive visual information, providing users with clear and real-time target location and vital sign data support. This efficient visualization design not only improves the practicality of the system but also provides strong support for user decision-making.

[0114] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-target life detection method based on radar signals, characterized in that, The method comprises the following steps: Step one, collecting the mixed signal of multiple targets received by the radar, and pre-processing the signal; The pre-processing includes filtering the signal to remove high-frequency noise and baseline drift, and normalizing the signal; the specific implementation is as follows: A multi-band composite radar system is used for signal acquisition, in which the X-band (8-12 GHz) is used for high-resolution surface target detection, and the L-band (1-2 GHz) is used for obstacle penetration detection; the radar transmitting end adopts the multiple-input multiple-output (MIMO) technology, transmits time-frequency-space three-dimensional coded pulse signals, the pulse width can be adjusted in the range of 10 ns to 1 us, and the pulse repetition frequency (PRF) is adaptively in the range of 1 kHz to 10 MHz, to meet the detection requirements in different scenes; the receiving end is configured with a 16-channel MIMO array, the noise coefficient is less than 2 dB, and the dynamic range is greater than 80 dB, to ensure that the weak vital sign signals can still be captured in a strong interference environment; For high-frequency noise suppression, a band-pass filter with a passband range of 0.1 Hz to 5 Hz is used to retain the effective frequency band of the breathing and heartbeat signals; the baseline drift is eliminated by wavelet transform, that is, a db4 wavelet basis function is selected for 5-layer decomposition to remove the low-frequency drift component in the signal; the normalization is realized by the following formula: where x(t) is the original signal; μ is the signal mean; σ is the signal standard deviation, is the normalized signal; this processing makes the signal amplitude of different channels and frequency bands uniform to the same order of magnitude, avoiding the amplitude deviation problem in subsequent processing; To adapt to complex electromagnetic environments, an adaptive noise cancellation technology is further introduced; that is, the environmental noise signal is collected through a reference channel, the filter coefficient is updated in real time by using the least mean square algorithm, and the noise component is subtracted from the main signal; the update formula is: w(n+1)=w(n)+μe(n)x(n); Wherein, w(n+1) is the filter weight vector, μ is the step factor, e(n) is the error signal, and x(n) is the reference noise signal; by such method, the signal-to-noise ratio is significantly improved, and a purer input signal is provided for subsequent processing; The radar system parameters are dynamically adjusted according to the actual scene; that is, the L-band is preferentially used when penetrating the concrete wall, and the X-band is switched to when high-resolution positioning is required; in addition, the signal sampling rate is set to 2GSPS (gigasamples per second) to ensure complete capture of high-frequency signals; the pre-processed signal is transmitted to the processing unit through a high-speed optical fiber interface to avoid signal attenuation and interference during transmission; For the special needs of multiple target scenes, a preliminary screening of blind source separation technology is introduced in the pre-processing stage; that is, the mixed signal is coarsely separated by independent component analysis, and the potential vital sign signal component is screened out, and then is processed by the subsequent module; Step two, reconstructing the phase space of the pre-processed signal to obtain a high-dimensional phase space trajectory; The phase space reconstruction is realized by Takens embedding theorem, which specifically includes: determining the delay time according to the autocorrelation function method, and determining the embedding dimension by the false nearest neighbor method; the specific implementation is as follows: First, the delay time τ is determined by the autocorrelation function method; the calculation formula of the autocorrelation function R(τ) is: Wherein, x(t) is the pre-processed signal, N is the signal length, and τ is the delay time; The first zero-crossing point of the autocorrelation function is selected as the optimal delay time to ensure that the trajectory of the reconstructed phase space can be fully unfolded; The determination of the embedding dimension m is realized by the false nearest neighbor method; The specific steps include: gradually increasing the embedding dimension, calculating the false nearest neighbor ratio, and stopping increasing when the ratio is less than the preset threshold; According to Takens' embedding theorem, the pretreated signal is reconstructed into a high-dimensional phase space trajectory X(t) by phase space reconstruction. Wherein, m is the embedding dimension, and τ is the delay time; Through phase space reconstruction, the dynamic characteristics of the signal are mapped into a high-dimensional space, providing a more intuitive geometric structure for subsequent signal separation; In order to further improve the reconstruction effect, the local projection method is introduced to smooth the phase space trajectory; That is, through the local linear embedding (LLE) algorithm, the high-dimensional trajectory is projected into a low-dimensional space, retaining the main dynamic characteristics of the signal while removing the trajectory jitter caused by noise; The objective function is: wherein y i is a projection point in a low-dimensional space; w ij is a local weight coefficient; k is a number of near neighbor points; For the multi-target scene in a complex electromagnetic environment, a multi-scale phase space reconstruction technology is further introduced; That is, through wavelet transform, the signal is decomposed into multiple scale components, and phase space reconstruction is performed respectively, and finally the reconstruction results of each scale are fused; This method can effectively capture the multi-scale characteristics of the signal and improve the accuracy of subsequent signal separation; In order to verify the effect of phase space reconstruction, the chaotic characteristics of the signal are evaluated by calculating the maximum Lyapunov exponent; The Lyapunov exponent is calculated by Wolf algorithm: where d i is the distance between two points on the phase space trajectory, d0is the initial distance; if the Lyapunov exponent is positive, it indicates that the signal has chaotic characteristics and is suitable for processing using nonlinear dynamics methods; The results of phase space reconstruction are displayed through visualization technology; That is, the high-dimensional trajectory is projected into a three-dimensional space to observe its geometric structure and dynamic evolution law; Step three, construct an adaptive dictionary and perform sparse representation on the signal; The adaptive dictionary is constructed by K-SVD algorithm, which specifically includes: by minimizing the reconstruction error of the linear combination of the signal and the dictionary atom, the dictionary atom and the sparse coefficient are optimized; The specific implementation is as follows: An adaptive dictionary D is constructed using the K-SVD algorithm; This algorithm minimizes the signal reconstruction error by alternately optimizing the dictionary atom and the sparse coefficient; The objective function is: where X(t) is the high-dimensional trajectory after phase space reconstruction, D is the dictionary matrix to be optimized, α i is the sparse coefficient vector, K is the sparsity constraint, which controls the maximum number of non-zero elements in the coefficient, and N is the total number of signal samples. Step four, separate the preliminary vital signs of each target from the mixed signal through sparse optimization algorithm; The sparse optimization algorithm is the orthogonal matching pursuit algorithm, which specifically includes: by iteratively selecting dictionary atoms, the optimal sparse representation of the signal is gradually approached; The specific implementation is as follows: Dictionary initialization with random matrices or predefined basis functions; initial dictionary D (0) The atoms are generated from a random Gaussian distribution and are gradually optimized by iterative updates; each iteration contains two stages: sparse coding and dictionary update: Sparse coding stage: fixed dictionary D, solve sparse coefficient a by using orthogonal matching pursuit (OMP) algorithm i ; Dictionary update phase: fixing the sparse coefficient a i updating the dictionary atoms d column-wise k optimizing the combination of atoms and their corresponding coefficients by singular value decomposition (SVD); The dictionary atom update formula is: where E k denotes the residual matrix after removing the current atom d k , ‖·‖ F is the Frobenius norm, and a k is the row vector of the corresponding sparse coefficients; by SVD decomposition of E k = U∑V T , the updated atom is taken as the first column of U, and the sparse coefficients are updated as ∑(1,1) x V(:,1) T ; In order to improve the adaptability of the dictionary to the vital signs, physiological feature prior knowledge is introduced, the typical waveforms of the respiratory and heartbeat signals are used as initial dictionary atoms, and then the K-SVD algorithm is used for fine tuning; This speeds up the convergence of the algorithm and improves the interpretability of the sparse representation; In view of the complexity of the multi-target scene, a block dictionary construction strategy is adopted; That is, the phase space trajectory is divided into multiple sub-blocks, sub-dictionaries are constructed respectively, and global optimization is realized through joint sparse constraint; The objective function is extended as: wherein X b (t) is a sub-block signal, D b is a corresponding sub-dictionary, K total is a global sparsity constraint; The results of sparse representation are evaluated by residual analysis; the reconstructed signal is computed The mean square error with the original signal X(t): When the MSE is lower than the preset threshold, it is determined that the sparse representation meets the convergence criteria; Step five, use a nonlinear dynamic model to correct the preliminary separated signal to eliminate residual aliasing and obtain the final separated signal; The nonlinear dynamic model is used to describe the coupling relationship between signals, specifically including: eliminating residual aliasing in the preliminary separated signals through a correction function; the specific implementation is as follows: First, a nonlinear dynamic model is constructed to describe the coupling relationship between signals; that is, a nonlinear coupling function is used to describe the interaction between target signals, which includes linear, quadratic and cubic terms, and the coupling coefficients are determined by least squares fitting; The preliminary separated signals are corrected using the nonlinear dynamic model; the specific steps are as follows: Calculate the coupling signal: according to the nonlinear coupling function, calculate the coupling effect of each target signal on other signals; Eliminate residual aliasing: subtract the coupling signal from the preliminary separated signal to obtain the corrected signal; this method can effectively eliminate the residual aliasing between signals and improve the separation accuracy; To improve the correction accuracy, an adaptive coupling coefficient updating mechanism is introduced; that is, the coupling coefficients are dynamically adjusted by gradient descent method to minimize the correction error; this method can optimize the correction effect in real time according to the signal characteristics; In view of the complexity of multi-target scene, a stage-by-stage correction strategy is adopted; that is, the signal is divided into multiple frequency bands, and nonlinear dynamic correction is performed respectively, and finally the correction results of each frequency band are fused; To further enhance the robustness, a noise suppression correction model is introduced; that is, the norm constraint of noise term is added in the correction process to balance the correction error and noise suppression; To verify the effect of nonlinear dynamic correction, the signal separation accuracy is evaluated by spectral analysis; that is, the spectral difference between the signals before and after correction is calculated to ensure that the main vital sign frequency components are retained while the pseudo-frequency components introduced by aliasing are eliminated; Step six, based on the final separated signal, the target position is calculated and the vital sign information is extracted; The target position calculation is realized by time reversal localization algorithm, specifically including: the spatial position of the target is calculated by constructing the time reversal channel response function; the specific implementation is as follows: The time reversal localization algorithm (TIMLE) is used to calculate the target position; this algorithm constructs the time reversal channel response function, combines the time reversal characteristics of multipath signals, and accurately calculates the spatial coordinates of the target; the channel impulse response is estimated by the correlation between the received signal and the reference signal, and the target position is finally determined by time reversal processing; To improve the positioning accuracy, a multi-sensor fusion technology is introduced; that is, the positioning results of multiple radar nodes are weighted and fused, and the weight coefficient is dynamically adjusted according to the signal-to-noise ratio or distance to obtain the optimal target position; The extraction of vital sign signal is realized by spectral analysis; that is, the modified separated signal is subjected to fast Fourier transform, the peak frequency in the spectrum is detected, and the respiratory frequency and heartbeat frequency are determined; by setting the frequency band threshold, noise interference is excluded to ensure the effectiveness of the extraction result; An adaptive filtering technique is introduced to smooth the spectrum; that is, a Kalman filter is used to dynamically update the state estimation to remove the pseudo-frequency components caused by noise and retain the true vital sign frequency; In view of the complexity of multi-target scene, a frequency band extraction strategy is adopted; that is, the signal is divided into different frequency bands, and the respiratory and heartbeat features are extracted respectively, and the final vital sign information is determined through joint optimization; Step seven, output target position information and vital signs signal.

2. A multi-target life-detection system based on radar signals, according to the method of claim 1, characterized in that, Comprise: Signal acquisition module, for collecting mixed signal of multiple targets received by radar, its output end connects signal preprocessing module; Signal preprocessing module, for filtering and normalizing the signal, its output end connects phase space reconstruction module; Phase space reconstruction module, for reconstructing the phase space of the preprocessed signal, its output end connects dictionary construction module; Dictionary construction module, for constructing adaptive dictionary and sparsely representing the signal, its output end connects signal separation module; Signal separation module, for separating the preliminary vital signs signal of each target through sparse optimization algorithm, its output end connects signal correction module; Signal correction module, for eliminating residual aliasing by using nonlinear dynamic model, its output end connects target positioning module; Target positioning module, for calculating target position and extracting vital signs information, its output end connects result output module; Result output module, for outputting target position information and vital signs signal; The signal separation module includes an orthogonal matching pursuit algorithm unit, for selecting dictionary atoms by iteration, gradually approaching the optimal sparse representation of the signal; The target positioning module includes a time reversal positioning algorithm unit, for calculating the spatial position of the target by constructing the time reversal channel response function.

Images