A matrix audio life detection method and system

Abstract

The present application relates to the technical field of acoustic wave detection and acoustic positioning, and specifically discloses a matrix audio life detection method and system. The technical scheme of the present application synchronously collects multi-channel signals by deploying a matrix audio sensor array, applies adaptive spatial decorrelation processing and extracts passive noise components to calibrate the sound field transfer function matrix, performs sparse deconvolution beamforming to obtain high-resolution sound source spectrum, and acquires azimuth parameters according to acoustic fingerprints, constructs an adaptive spatial filter constrained by spatial wave-trap to perform collaborative filtering, time-frequency spectrum subtraction and dynamic gain compensation. The present application solves the problem of insufficient positioning accuracy and difficulty in extracting weak signals caused by environmental noise, structure reflected sound waves and multipath effect interference of existing devices, and improves the high-sensitivity identification of vital signs at a buried depth, the precise azimuth determination capability, the decision-making efficiency and the search and rescue success rate within the golden search and rescue time.

Description

Technical Field

[0001] This invention relates to the field of acoustic wave detection and acoustic positioning technology, and in particular to a matrix-type audio life detection method and system. Background Technology

[0002] In emergency rescue scenarios such as earthquakes and building collapses, the rapid and accurate detection of life signs in trapped individuals is crucial for improving rescue efficiency and survival rates. Audio life detection technology, with its advantages of being non-contact, having strong penetration, and adapting to complex environments, has become an important means of life search in rubble environments. Existing audio life detection equipment mostly uses single-point pickup or simple linear array acquisition methods, which are easily affected by environmental noise, structurally reflected sound waves, and multipath effects in complex rubble environments, leading to insufficient accuracy in locating target sound sources and difficulty in extracting weak vital signs. Furthermore, existing systems lack effective means for the collaborative processing and spatial filtering of multi-channel audio data, making it difficult to achieve high-sensitivity identification and accurate location determination of weak breathing sounds, heartbeats, and other vital signs deep underground. This restricts the decision-making efficiency and success rate of rescuers within the golden search and rescue time. There is an urgent need to improve the environmental adaptability and target resolution capabilities of audio life detection systems through array signal processing and multi-channel collaborative detection technologies. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a matrix-based audio life detection method and system.

[0004] In a first aspect, the present invention provides a matrix-based audio life detection method, the technical solution of which is as follows: A matrix-type audio sensor array is deployed in a distributed manner in the detection area to form a two-dimensional planar array element layout. Multi-channel ambient audio signals of the ruins are synchronously acquired through the matrix-type audio sensor array. An adaptive spatial decorrelation process based on the environmental noise cross-spectrum matrix is ​​applied to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, while passive noise components are extracted from the multi-channel ruin environmental audio signal. Using the passive noise component, the acoustic field transfer function matrix corresponding to the buried medium within the detection area is calibrated in real time; Joint sparse deconvolution beamforming is performed on the multi-channel denoised intermediate signal and the sound field transfer function matrix to map the multi-channel denoised intermediate signal onto a preset sound source spatial grid, thereby obtaining a high-resolution sound source spatial spectrum distribution that varies with time and frequency. Based on the acoustic fingerprint parameters of vital signs, candidate vital sign sound sources are screened from the spatial spectrum distribution of the high-resolution sound sources, and the spatial orientation parameters of the candidate vital sign sound sources are extracted. Based on the spatial orientation parameters and the two-dimensional planar array element layout, an adaptive spatial filter constrained by spatial notch filtering is constructed to perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal, thereby separating the enhanced vital sign target signal. Time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation are applied to the enhanced vital sign target signal to recover the weak signal components attenuated due to burial depth and generate detection result output.

[0005] Secondly, the present invention provides a matrix-type audio life detection system, the technical solution of which is as follows: The deployment module is used to deploy a matrix audio sensor array in a distributed manner in the detection area, forming a two-dimensional planar array element layout, and synchronously acquire multi-channel ruin environment audio signals through the matrix audio sensor array. The extraction module is used to apply adaptive spatial decorrelation processing based on the environmental noise cross-spectrum matrix to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, and at the same time extract passive noise components from the multi-channel ruin environmental audio signal. The calibration module is used to calibrate the acoustic field transfer function matrix corresponding to the buried medium in the detection area in real time using the passive noise component; The mapping module is used to perform joint sparse deconvolution beamforming on the multi-channel denoised intermediate signal and the sound field transfer function matrix, and to map the multi-channel denoised intermediate signal to a preset sound source space grid to obtain a high-resolution sound source space spectrum distribution that varies with time and frequency. The filtering module is used to filter candidate vital sign sound sources in the high-resolution sound source spatial spectrum distribution based on the acoustic fingerprint parameters of vital signs, and to extract the spatial orientation parameters of the candidate vital sign sound sources. The filtering module is used to construct an adaptive spatial filter constrained by spatial notch filtering based on the spatial orientation parameters and the two-dimensional planar array element layout, and to perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal to separate the enhanced vital sign target signal. The generation module is used to apply time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation to the enhanced vital sign target signal, to recover the weak signal components attenuated due to burial depth, and to generate the detection result output.

[0006] The technical solution of this invention deploys a matrix-type audio sensor array to synchronously acquire multi-channel signals, applies adaptive spatial decorrelation processing and extracts passive noise components to calibrate the sound field transfer function matrix, performs sparse deconvolution beamforming to obtain a high-resolution sound source spectrum and obtains azimuth parameters based on acoustic fingerprints, and constructs an adaptive spatial filter constrained by spatial notch filtering for collaborative filtering, time-spectrum subtraction and dynamic gain compensation. This solves the problems of insufficient positioning accuracy and difficulty in extracting weak signals caused by interference from environmental noise, structural reflected sound waves and multipath effects in existing equipment, thereby improving the high-sensitivity identification of vital signs in deep burial sites, accurate azimuth determination capabilities, decision-making efficiency within the golden search and rescue time, and search and rescue success rate.

[0007] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0009] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating an embodiment of a matrix-based audio life detection method according to the present invention. Detailed Implementation

[0010] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0011] Figure 1 The diagram illustrates a flowchart of an embodiment of a matrix-based audio life detection method provided by the present invention, executed by a control terminal. Figure 1 As shown, this matrix-based audio life detection method includes the following steps: S1. A matrix audio sensor array is deployed in a distributed manner in the detection area to form a two-dimensional planar array element layout. Multi-channel environmental audio signals of the ruins are synchronously acquired through the matrix audio sensor array.

[0012] The detection area refers to the area of ​​ruin space selected for life detection operations in disaster emergency rescue scenarios. It is defined by search and rescue personnel based on the building collapse pattern, estimated burial depth, and search and rescue priorities. For example, after the earthquake in City A on January 1, 2026, in the ruins of a partially collapsed 7-story residential building, search and rescue personnel designated an irregular area of ​​about 200m², including the fractured area of ​​the top precast slab, the middle brick-concrete masonry accumulation area, and the bottom soil and rock mixed filling area, as the detection area, which served as the working surface for deploying a matrix-type audio sensor array.

[0013] Among them, a matrix audio sensor array refers to a collection of audio pickup hardware consisting of multiple unidirectional pickup elements arranged according to a predetermined spatial geometric relationship, used to synchronously perceive the spatial distribution of the sound field. For example, in the aforementioned residential building ruins detection area, rescuers deployed a matrix audio sensor array composed of 32 unidirectional miniature microphones. Each microphone is connected to a portable control terminal via a waterproof quick-connect connector. The control terminal synchronously collects multi-channel environmental audio signals from the ruins and transmits them back to the on-site data processing unit in real time. A two-dimensional planar array element layout refers to the two-dimensional geometric arrangement of the pickup elements in the matrix audio sensor array on the surface projection of the detection area. For example, in the aforementioned residential building ruins detection area, rescuers deployed 32 unidirectional miniature microphones along the horizontal projection surface of the ruins into a two-dimensional planar array element layout, with the array element coordinate grid covering a 4m × 5m rectangular operating area.

[0014] Among them, the multi-channel ruin environment audio signal refers to the multi-channel discrete time sequence that carries the sound field information of the ruin environment, which is synchronously collected by all the pickup array elements in the matrix audio sensor array within the same time period. For example, in the above-mentioned residential building ruin detection area, 32 unidirectional miniature pickups synchronously record the weak breathing rhythm signals emitted by the buried people, the mechanical vibrations at the search and rescue site and the noise of distant traffic during the 120s acquisition period, forming 32 channels of 16-bit quantized, 48000Hz sampled multi-channel ruin environment audio signal.

[0015] S2. Apply adaptive spatial decorrelation processing based on the environmental noise cross-spectrum matrix to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, and extract passive noise components from the multi-channel ruin environmental audio signal.

[0016] Among them, the environmental noise cross-spectrum matrix refers to the matrix obtained by estimating the frequency domain cross-power spectral density of the time period containing only environmental noise in the multi-channel ruin environmental audio signal. For example, in the above-mentioned residential building ruin detection area, the control terminal intercepts an 8-second silent segment in the multi-channel ruin environmental audio signal where the search and rescue team stops knocking and the generator is briefly shut down. The cross-power spectrum between all pairs of the 32 array elements in the silent segment is calculated to obtain a set of 32 rows and 32 columns of environmental noise cross-spectrum matrix at each frequency point.

[0017] The adaptive spatial decorrelation processing refers to the signal processing step of using the environmental noise cross-spectrum matrix to perform spatial domain projection calculation on the multi-channel ruin environmental audio signal, and adjusting the decorrelation weights in real time according to the statistical characteristics of the noise field to suppress the diffused environmental noise. For example, in the aforementioned residential building ruin detection area, the control terminal performs spatial projection of each frame of multi-channel ruin environmental audio signal through the noise subspace projection matrix obtained by the eigenvalue decomposition of the environmental noise cross-spectrum matrix of the corresponding frequency, suppressing the isotropic diffused noise caused by distant traffic and wind in real time, filtering out the components corresponding to the environmental noise subspace in the original multi-channel ruin environmental audio signal, and reducing the environmental background noise power in the multi-channel denoised intermediate signal after adaptive spatial decorrelation processing by about 14dB compared to before processing.

[0018] Among them, the multi-channel denoising intermediate signal refers to the multi-channel transition signal obtained after adaptive spatial decorrelation processing of the multi-channel ruin environment audio signal, which significantly reduces the ambient noise of the diffuse field. For example, in the above-mentioned residential building ruin detection area, after the control terminal performs adaptive spatial decorrelation processing on 32 multi-channel ruin environment audio signals, the 32 data are named "Channel 1 Denoising Frame Sequence" to "Channel 32 Denoising Frame Sequence". The multi-channel denoising intermediate signal provides input for subsequent sound field transfer function calibration and joint sparse deconvolution beamforming.

[0019] Passive noise components refer to the residual noise sequences separated from the original signals during the adaptive spatial decorrelation processing of multi-channel ruin environment audio signals. These noise sequences are randomly excited by scatterers within the detection area and do not contain sound source information related to the vital signs of the target. For example, in the aforementioned residential building ruin detection area, the control terminal projects 32 multi-channel ruin environment audio signals onto the noise subspace to obtain a time series of 32 projection coefficients. The projection coefficient sequences together constitute the passive noise components. The passive noise components record the random sound energy distribution of non-human sources, such as the slight movement of concrete fragments and the remnants of pneumatic drills.

[0020] S3. Using the passive noise component, calibrate the acoustic field transfer function matrix corresponding to the buried medium in the detection area in real time.

[0021] The burial medium refers to the solid layered structure of the ruins that covers the trapped personnel within the detection area and attenuates and scatters the sound waves of vital signs. For example, in the aforementioned residential building ruins detection area, based on on-site investigation and prior information from ground-penetrating radar, the burial medium consists of approximately 0.6m thick precast reinforced concrete slabs, approximately 1.5m thick brick-concrete masonry blocks, and approximately 0.8m thick soil-rock mixture filling layer from top to bottom.

[0022] The sound field transfer function matrix refers to the matrix that describes the frequency response relationship of the sound propagation path between each candidate location in the sound source space and each element of the matrix audio sensor array. For example, in the aforementioned residential building ruins detection area, the control terminal uses the spatial covariance characteristics of the passive noise component and combines it with the finite element sound propagation model to estimate the sound velocity and density of the three-layer buried medium online, and calibrates in real time the sound field transfer function matrix from 100 preset spatial grid points to 32 array elements, covering a frequency range of 50Hz to 1000Hz.

[0023] The principle of using passive noise components for sound field transfer function matrix calibration in S3 is as follows: the passive noise components inside the buried medium, which are randomly excited by the micro-motion of the scatterer, have spatial incoherence. The difference in the received signals of the passive noise components on each element of the matrix audio sensor array is uniquely determined by the sound propagation characteristics of the buried medium itself. By calculating the principal eigenvector of the cross-spectral matrix between each element of the passive noise components, the sound field transfer function information equivalent to the impulse response function of the buried medium can be extracted, realizing online passive sound field calibration without actively transmitting calibration signals, and avoiding the resonance interference of the buried medium excited by the active sound source, which affects the detection sensitivity of the vital signs sound source.

[0024] S4. Perform joint sparse deconvolution beamforming on the multi-channel denoised intermediate signal and the sound field transfer function matrix to map the multi-channel denoised intermediate signal onto a preset sound source spatial grid, thereby obtaining a high-resolution sound source spatial spectrum distribution that varies with time and frequency.

[0025] The preset sound source spatial grid refers to a set of discrete spatial coordinate points predefined within the detection area, used to discretize the continuous sound source space into a finite number of candidate sound source locations. For example, in the aforementioned residential building ruins detection area, the control terminal presets approximately 1500 spatial grid points within a three-dimensional cuboid search space of 4m long, 5m wide, and 0 to 3m deep, with a grid spacing of 0.2m, forming the preset sound source spatial grid for sound source spatial discretization search. The high-resolution sound source spatial spectrum distribution refers to a three-dimensional spectrum obtained through joint sparse deconvolution beamforming operations, reflecting the changes in sound source intensity with spatial location, time, and frequency on the preset sound source spatial grid. For example, in the aforementioned residential building ruins detection area, the control terminal solves a joint objective function to map the multi-channel denoised intermediate signal onto the preset sound source spatial grid, obtaining a high-resolution sound source spatial spectrum distribution updated every 0.5s. The grid where the buried person is located exhibits a continuous strong spectral peak in the 0.2Hz to 0.6Hz breathing frequency band.

[0026] S5. Based on the acoustic fingerprint parameters of vital signs, candidate vital sign sound sources are screened in the spatial spectrum distribution of the high-resolution sound source, and the spatial orientation parameters of the candidate vital sign sound sources are extracted.

[0027] Among them, the acoustic fingerprint parameters of vital signs refer to the set of acoustic feature patterns extracted in advance from a large number of real rescue cases, which are used to distinguish the sound source of human vital signs from the sound source of mechanical vibration noise in the spatial spectrum distribution of sound sources. For example, in the above-mentioned residential building ruins detection area, the acoustic fingerprint parameters of vital signs include the narrow band energy spectrum peak sequence that appears regularly every 2s to 5s in the 0.2Hz to 0.6Hz range formed by the breathing rhythm of the buried person, and the periodic low frequency pulse envelope feature corresponding to the heartbeat in the 35Hz to 90Hz range. The control terminal uses the acoustic fingerprint parameters of vital signs to distinguish the vital sign pattern of the buried person from the continuous narrow band interference pattern of the generator at the search and rescue site from the high resolution sound source spatial spectrum distribution. Candidate vital sign sound sources refer to the spatial sound source locations of trapped personnel that are initially determined after matching and screening the acoustic fingerprint parameters of vital signs in the high-resolution sound source spatial spectrum distribution. For example, in the above-mentioned residential building ruins detection area, the control terminal marks two spatial grid points in the high-resolution sound source spatial spectrum distribution that have a regular low-frequency pulse envelope at a frequency of 65Hz and an envelope repetition period of about 0.85s as candidate vital sign sound sources, and number them as candidate source A and candidate source B, respectively.

[0028] Each row of the high-resolution sound source spatial spectrum distribution matrix S corresponds to the sound source intensity sequence along the time dimension of a spatial grid point in the preset sound source spatial grid. Each column of matrix S corresponds to the sound source spatial spectrum distribution of all spatial grid points at a given time. When the spatial grid point is located at the location of the buried personnel, the time-frequency peaks in the corresponding row of matrix S exhibit a periodic structure that matches the acoustic fingerprint parameters of vital signs. The occurrence time and duration of the time-frequency peaks change regularly with the breathing rhythm or heartbeat pulse. This periodic structure is quantified and matched and tracked in subsequent steps using time-frequency sparsity features.

[0029] Among them, the spatial orientation parameters refer to a set of coordinate values ​​and corresponding azimuth and elevation angles, which are output by the joint sparse deconvolution beamforming operation and describe the specific geometric position of the candidate vital sign sound source in the detection area. For example, in the above-mentioned residential building ruins detection area, the control terminal extracts the spatial orientation parameters of candidate source A as follows: with the array center as the origin, the horizontal direction is 20° east of north, the distance is 2.3m, and the burial depth is about 1.8m. The spatial orientation parameters of candidate source B are as follows: with the array center as the origin, the horizontal direction is 45° west of south, the distance is 3.1m, and the burial depth is about 2.4m.

[0030] S6. Based on the spatial orientation parameters and the two-dimensional planar array element layout, construct an adaptive spatial filter constrained by spatial notch filtering, and perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal to separate the enhanced vital sign target signal.

[0031] Spatial notch constraint refers to the constraint condition in adaptive spatial filter design that forces the filter to form a response zero or deep null in a specified non-target location. For example, in the aforementioned residential building ruins detection area, when constructing an adaptive spatial filter to enhance candidate source A, the control terminal sets a spatial notch constraint to form a response null exceeding 30dB in the location of candidate source B and the location of a running generator, preventing noise from candidate source B and the generator from leaking into the signal channel of candidate source A. An adaptive spatial filter is a spatial filter that calculates the optimal weight vector in real time based on spatial orientation parameters and array layout, used to separate the sound source signal from the multi-channel denoised intermediate signal. For example, in the aforementioned residential building ruins detection area, for candidate source A, the control terminal calculates a 32-dimensional weight vector by solving a covariance matrix weighted optimization problem with sparse penalty terms and spatial notch constraints. This weight vector is then applied to the multi-channel denoised intermediate signal to obtain an enhanced vital sign target signal output by the adaptive spatial filter.

[0032] The enhanced vital signs target signal refers to the single-channel vital signs time-domain waveform separated after multi-channel collaborative spatial filtering by an adaptive spatial filter, which has a significantly improved signal-to-interference-plus-noise ratio. For example, in the aforementioned residential building ruins detection area, after the control terminal performs spatial filtering on the location of candidate source A, the enhanced vital signs target signal exhibits a clear periodic inspiratory-expiratory wave envelope with a respiratory envelope width of approximately 0.7s and an adjacent envelope interval of approximately 2.8s. The output signal-to-interference-plus-noise ratio is improved by approximately 11dB compared to before filtering.

[0033] S7. Apply time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation to the enhanced vital sign target signal to recover the weak signal components attenuated due to burial depth and generate detection result output.

[0034] Among them, time-frequency domain adaptive spectrum subtraction refers to the process of adaptively calculating the gain function of the enhanced vital sign target signal in the short-time Fourier transform domain based on the real-time estimated residual noise power spectrum, and subtracting the residual noise spectrum from the signal spectrum. For example, in the above-mentioned residential building ruins detection area, the control terminal uses the average amplitude spectrum corresponding to the non-breathing frames in the enhanced vital sign target signal as the adaptively updated noise spectrum estimate, subtracts the noise spectrum estimate from the amplitude spectrum containing the target frame, and outputs the clean spectrum after time-frequency domain adaptive spectrum subtraction. The 35Hz low-frequency pulse corresponding to the heartbeat is clearly preserved in the clean spectrum. Dynamic playback gain compensation refers to the process of calculating the frequency-related amplification factor in real time based on the burial depth of the candidate vital sign sound source and performing graded gain amplification on the signal after time-frequency adaptive spectral subtraction. For example, in the aforementioned residential building ruins detection area, the control terminal applies a dynamic gain compensation of 6dB per octave to the time-frequency component after spectral subtraction in the frequency range of 50Hz to 500Hz based on the burial depth of candidate source A of approximately 1.8m, thereby restoring the breathing sound signal, which has been attenuated by approximately 18dB due to penetrating three layers of burial medium, to above the audible threshold.

[0035] The detection result output refers to the final output from the control terminal to the search and rescue commanders, which presents the processed vital signs information to the search and rescue commanders in the form of a visual interface, audio broadcast, and data report. For example, in the aforementioned residential building ruins detection area, the control terminal generates the detection result output, draws a cross-sectional diagram of the detection area with the array center as the zero point on the display screen, marks the location of candidate source A, the burial depth value, and real-time breathing waveform segments, and plays back the gain-compensated breathing rhythm audio through headphones. Based on the detection result output, the search and rescue commanders dispatch the demolition team to the corresponding location of candidate source A to carry out the rescue.

[0036] The technical solution of this embodiment deploys a matrix audio sensor array to synchronously acquire multi-channel signals, applies adaptive spatial decorrelation processing and extracts passive noise components to calibrate the sound field transfer function matrix, performs sparse deconvolution beamforming to obtain a high-resolution sound source spectrum and obtains azimuth parameters based on acoustic fingerprints, and constructs an adaptive spatial filter constrained by spatial notch filtering for collaborative filtering, time-spectrum subtraction and dynamic gain compensation. This solves the problems of insufficient positioning accuracy and difficulty in extracting weak signals caused by interference from environmental noise, structural reflected sound waves and multipath effects in existing equipment, thereby improving the high-sensitivity identification of vital signs in deep burial sites, accurate azimuth determination capabilities, decision-making efficiency within the golden search and rescue time, and search and rescue success rate.

[0037] In one alternative approach, the two-dimensional planar array elements of the matrix audio sensor array are arranged as a non-uniformly spaced concentric ring array that extends along the surface projection plane of the detection area. The circumferential spacing between array elements on adjacent rings decreases as the ring radius increases, in order to ensure the uniformity of spatial sampling in all azimuth angles.

[0038] Among them, the non-uniformly spaced concentric ring array refers to a matrix audio sensor array whose two-dimensional planar array element layout consists of multiple rings centered at the same point, and the arc length spacing between adjacent array elements on the same ring decreases as the ring radius increases. For example, in the aforementioned residential building ruins detection area, a non-uniformly spaced concentric ring array consisting of 3 concentric rings is deployed, with the inner ring having a radius of 0.6m and an arc length spacing of 0.25m between array elements, the middle ring having a radius of 1.2m and an arc length spacing of 0.18m between array elements, and the outer ring having a radius of 2.0m and an arc length spacing of 0.12m between array elements, ensuring that the spatial sampling density is approximately the same in all azimuth angles.

[0039] In the above-mentioned optional methods, a non-uniformly spaced concentric ring array layout is further adopted, so that the circumferential spacing of the array elements on adjacent rings decreases as the radius increases, ensuring the uniformity of spatial sampling in all azimuth angles and improving the array's reception consistency for sound sources in different directions.

[0040] In one alternative approach, the main axis direction of each element in the matrix audio sensor array is tilted outward by a preset angle according to the center normal vector of the two-dimensional planar array element layout, so that the directional sensitive area covers a preset depth range of the buried medium within the detection area.

[0041] The preset angle refers to the pre-defined tilt angle between the main axis direction of the array element and the normal direction of the two-dimensional planar array element layout. This angle, determined through simulation analysis, ensures the directional sensitive area covers a preset depth range. For example, in the aforementioned residential building ruins detection area, the main axis direction of all unidirectional miniature microphones is tilted outwards from the mounting surface at a preset angle of 25°, ensuring the array's directional sensitive area covers a burial depth range of 0.5m to 3.0m, matching the estimated depth of trapped personnel on-site. The preset depth range refers to the burial depth interval pre-determined based on prior information about the thickness of the burial medium at the search and rescue site, which is the focus of the detection system's search. For example, in the aforementioned residential building ruins detection area, based on the ground-penetrating radar's measurement results of the total thickness of the reinforced concrete layer, brick-concrete masonry layer, and soil-rock mixed layer, the search and rescue personnel set the preset depth range to be between 0.5m and 3.0m from the ruins surface.

[0042] In the above-mentioned optional methods, the main axis of the array element is further tilted outward by a preset angle to expand the coverage of the directional sensitive area, so that the sensor can be aligned with the preset depth range of the buried medium and enhance the ability to capture vital signs signals from deep within.

[0043] In one alternative approach, the step of applying adaptive spatial decorrelation processing based on the environmental noise cross-spectrum matrix to the multi-channel ruin environment audio signal to obtain a multi-channel denoised intermediate signal includes: The multi-channel ruin environment audio signal is divided into frames, the noise cross-spectrum matrix of each frame is estimated, and the noise cross-spectrum matrix is ​​decomposed. The eigenvectors corresponding to the eigenvalues ​​that are less than the rank threshold of the environmental noise diffusion field are selected to form a noise subspace projection matrix. The noise subspace projection matrix is ​​used to spatially project the multi-channel ruin environment audio signal to remove coherent noise components, thereby obtaining the multi-channel denoised intermediate signal.

[0044] The noise cross-spectrum matrix refers to the complex matrix obtained by calculating the cross-power spectral density between each pair of array elements in a frame of noise signal from a multi-channel ruin environment audio signal after framing. For example, in the aforementioned residential building ruin detection area, the control terminal calculates the cross-power spectrum of 32 signals at a resolution of 1 Hz within the 0Hz to 2000Hz range for the 15th frame noise signal containing only mechanical vibration, obtaining a set of noise cross-spectrum matrices at each frequency point. The environmental noise diffusion field rank threshold refers to the eigenvalue threshold used to distinguish the diffusion field noise subspace from the point source signal subspace, which is analytically determined based on the spatial correlation radius of the diffusion field noise and the array geometric parameters. For example, in the aforementioned residential building ruin detection area, the control terminal sets the environmental noise diffusion field rank threshold to 8 based on the upper limit of the half-wavelength spacing of the 32-element array and the 500Hz diffusion field noise coherence radius. After performing eigenvalue decomposition on the noise cross-spectrum matrix, the eigenvectors with eigenvalues ​​ranked from the 9th to the 32nd are retained to form the noise subspace.

[0045] The noise subspace projection matrix refers to the projection operator matrix spanned by the eigenvectors corresponding to small eigenvalues ​​below the rank threshold of the environmental noise diffusion field. For example, in the aforementioned residential building ruins detection area, the control terminal forms a 32-row, 24-column noise subspace projection matrix using 24 eigenvectors at each frequency point whose eigenvalues ​​are less than the rank threshold of the environmental noise diffusion field. The multi-channel ruins environmental audio signal is then projected through the noise subspace projection matrix, and the noise components corresponding to the 24-dimensional subspace are removed.

[0046] In the above-mentioned optional methods, the noise subspace projection matrix is ​​further constructed by estimating the noise cross-spectrum matrix and performing eigenvalue decomposition to remove coherent noise components, thereby reducing the interference of environmental noise on weak vital signs signals and improving the accuracy of subsequent sound source localization and signal extraction.

[0047] In one alternative approach, the step of extracting passive noise components from the multi-channel rubble environment audio signal includes: The multi-channel ruin environment audio signal is projected onto the noise subspace, and the resulting projection coefficient sequence is used as the passive noise component.

[0048] In the above-mentioned optional methods, the projection coefficient sequence is further obtained by projecting the audio signal onto the noise subspace as a passive noise component, which provides a clean noise reference for the real-time calibration of the sound field transfer function matrix and improves the estimation accuracy of the sound field model.

[0049] In an alternative approach, the joint sparse deconvolution beamforming maps the multi-channel denoised intermediate signal to the preset sound source space grid by solving the following optimization problem: in, This represents the intermediate signal matrix for the multi-channel denoising. This represents the sound field transfer function matrix. This represents the high-resolution sound source spatial spectrum distribution matrix to be solved. Represents the residual noise matrix. Describing the Frobenius norm, for Mixture norm used for excitation row sparsity, This represents the weighted nuclear norm, and the weighting vector is determined by the prior distribution of the singular values ​​of the residual noise. and This is the regularization parameter.

[0050] It should be noted that the joint sparse deconvolution beamforming formula transforms the problem of estimating the spatial spectrum distribution of the sound source into a convex optimization problem with multiple constraints. The first term in the expression... For the data fidelity term, the constraint is to minimize the reconstruction error between the multi-channel denoised intermediate signal matrix Y and the forward propagation model constructed from the sound field transfer function matrix H and the high-resolution sound source spatial spectrum distribution matrix S, as well as the residual noise matrix N. The second term... for The mixed norm penalty term is first calculated for each row of matrix S. The norm measures the energy intensity of each spatial grid point in the time dimension, and then applies it to the cross-row... Norm vector application The norm excites row sparsity, causing the spatial spectrum distribution of the sound source to exhibit non-zero values ​​only in the grid row containing the real source and approach zero in the other rows. (Third term) The weighted norm penalty term is applied by weighting the singular value vectors of the residual noise matrix N. The norm constrains the low-rank structure of the residual noise, and the weighting vector is determined by the prior distribution of each singular value of the residual noise, ensuring that the penalty intensity matches the prior significance of the singular values. The above formula, under the condition of multipath propagation in complex ruins, utilizes the sparse prior of the spatial distribution of the sound source and the low-rank characteristics of the residual noise to deconvolve and separate the high-resolution spatial spectrum distribution of the sound source from the intermediate signal of the multi-channel denoising, while simultaneously suppressing coherent interference pseudo-peaks introduced by scattering and reflection from the burial medium.

[0051] In the above-mentioned optional methods, further optimization is achieved by combining sparse deconvolution beamforming to solve the problem. By utilizing the hybrid norm to constrain the row sparsity of the sound source spatial spectrum and the residual noise weighted suppression, the resolution of the sound source spatial spectrum is improved, and high-precision sound source spatial mapping is realized.

[0052] In one alternative approach, the step of screening candidate vital sign sound sources in the high-resolution sound source spatial spectrum distribution based on vital sign acoustic fingerprint parameters includes: The high-resolution sound source spatial spectrum distribution matrix along the time dimension A sliding window is used for truncation, and the time-frequency sparsity features of the spatial spectrum within each sliding window are calculated. The time-frequency sparsity features are matched and tracked with the periodic energy fluctuation patterns represented in the acoustic fingerprint parameters of vital signs. Spatial grid points with matching degrees exceeding the adaptive threshold are marked as candidate vital sign sound sources.

[0053] Among them, the time-frequency sparsity feature refers to the characteristic quantity that measures the proportion of significantly non-zero elements in the high-resolution sound source spatial spectrum distribution matrix S after expanding it into a time-frequency two-dimensional matrix within a sliding window. For example, in the above-mentioned residential building ruins detection area, the control terminal performs a 10s sliding window truncation on the high-resolution sound source spatial spectrum distribution matrix S. In the third sliding window, the spatial grid point 53 shows two significant spectral peaks with an interval of about 0.86s in the 60Hz to 90Hz range, while the energy of the other time-frequency points is close to zero. The control terminal calculates that the time-frequency sparsity feature of grid point 53 is high sparsity, which is highly matched with the heartbeat pulse pattern represented in the acoustic fingerprint parameters of vital signs. Periodic energy fluctuation patterns refer to the time-frequency two-dimensional feature templates contained in the acoustic fingerprint parameters of vital signs, which describe the rhythmic pattern of the spectral energy caused by breathing or heartbeat alternating in strength and weakness on the time axis. For example, in the aforementioned residential building ruins detection area, the standard respiratory periodic energy fluctuation pattern stored in the acoustic fingerprint parameters of vital signs is characterized by an energy peak occurring every 3 seconds at a frequency of 0.3Hz, with the energy rapidly dropping below the noise floor between peaks. The control terminal performs normalized cross-correlation matching between the measured time-frequency sparsity characteristics of the spatial spectrum within the sliding window and the periodic energy fluctuation pattern. A successful match is determined when the cross-correlation peak exceeds 0.75.

[0054] The adaptive threshold refers to the judgment threshold dynamically adjusted based on the background noise level of the spatial spectrum within the current sliding window during the matching and tracking process. For example, in the aforementioned residential building ruins detection area, the control terminal calculates the average spectral energy of all unmarked spatial grid points within the 5th sliding window as the estimated background noise. The adaptive threshold is set to the estimated background noise plus 6 dB. Spatial grid points within the sliding window with a matching degree exceeding the adaptive threshold are marked as candidate vital sign sound sources, avoiding mismarking of knocking vibrations at the search and rescue site as candidate vital sign sound sources due to background noise fluctuations. A spatial grid point refers to a discretized candidate sound source unit indexed by three-dimensional coordinates within a preset sound source spatial grid. For example, in the aforementioned residential building ruins detection area, the spatial grid point index in the preset sound source spatial grid is represented as (row number i, column number j, layer number k). The spatial grid point (15, 21, 6) corresponds to a horizontal coordinate of 15° east of north of the array center, a distance of 2.1 m, and a burial depth of 1.8 m. The control terminal outputs high-resolution sound source spatial spectrum and spatial orientation parameters for the spatial grid points.

[0055] In the above-mentioned optional methods, the spatial spectrum is further extracted by a sliding window and the time-frequency sparsity features are calculated. These features are then matched and tracked with the periodic energy fluctuation patterns. Candidate vital sign sound sources are then selected based on an adaptive threshold, thereby improving the accuracy and reliability of vital sign identification.

[0056] In one alternative approach, the steps of constructing an adaptive spatial filter constrained by spatial notch filtering include: The target steering vector is determined based on the spatial orientation parameters. An array manifold is generated based on the two-dimensional planar array element layout. Multiple interference notch steering vectors at non-target orientations are selected to form a notch constraint matrix. The optimal weight vector of the adaptive spatial filter is obtained by solving the following constraint optimization problem: in, This represents the weight vector of the adaptive spatial domain filter. This represents the covariance matrix of the intermediate signal in the multi-channel denoising. This represents the notch constraint matrix. Represents the constraint response vector. for Norms are used to enhance the sparsity of weight vectors. For sparse regularization parameters, This indicates the conjugate transpose.

[0057] It should be noted that the formula for solving the weight vector of the adaptive spatial domain filter introduces weight vector sparsity regularization within the framework of linearly constrained minimum variance beamforming. Objective function The total output power of the filter is suppressed by minimizing the quadratic form of the weight vector w, where the covariance matrix R carries the spatial statistical characteristics of the intermediate signals for multi-channel denoising. Minimizing the output power under target orientation constraints is equivalent to minimizing the contribution of interference and noise from non-target orientations. (Penalty term) Apply weight vector Norm: In the excitation weight vector, the weight coefficients corresponding to most array elements approach zero. The sparsified weight vector retains only the array elements that significantly contribute to the target azimuth for beamforming, thereby improving the filter's robustness to array position disturbances and transfer function mismatch. Equality constraints The forced filter forms precise response zeros in the interference subspace spanned by the column vectors of the notch constraint matrix C. The notch constraint matrix consists of steering vectors from the non-target strong interference azimuth, and the constraint response vector f is set to all zeros to form deep nulls in the interference azimuth. Based on the spatial azimuth parameters of the candidate vital sign sound sources and the distribution of strong interference, the above formula calculates an adaptive spatial filter weight vector with sparse array element utilization characteristics and the ability to create deep nulls in multiple interference azimuths. Applying the filter to the multi-channel denoising intermediate signal achieves directional enhancement of the vital sign target signal and spatially selective suppression of the interference source.

[0058] In the above-mentioned optional methods, non-target azimuth interference is further suppressed by constructing a notch constraint matrix, and the optimal weight vector is solved by sparse regularization to enhance the receiving strength of the target azimuth signal, suppress sidelobe interference, and improve the signal-to-noise ratio of the vital signs target signal.

[0059] In one alternative approach, the multiple interference notch steering vectors at the non-target azimuth are obtained by selecting strong interference azimuths in the high-resolution sound source spatial spectrum distribution, excluding the grid point where the candidate vital sign sound source is located.

[0060] Among them, the interference notch steering vector refers to the steering vector calculated based on the coordinates of the strong interference source grid points (excluding candidate vital sign sound sources) and array geometry in the preset sound source spatial grid when constructing the notch constraint matrix. For example, in the above-mentioned residential building ruins detection area, the control terminal selects the five strong interference grid points with the top five spectral peak intensities (excluding candidate sources A and B) in the high-resolution sound source spatial spectrum distribution. Based on the azimuth and elevation angles of each strong interference grid point and the layout of the two-dimensional planar array elements, five interference notch steering vectors are calculated respectively. The interference notch steering vectors are used as column vectors to form the notch constraint matrix, so that the adaptive spatial filter forms a response zero point in the strong interference azimuth.

[0061] In the above-mentioned optional methods, by further selecting the strong interference direction other than the candidate vital sign sound source in the high-resolution sound source spatial spectrum distribution as the notch guide vector, the notch constraint is made to better fit the actual interference distribution, thereby improving the adaptability of spatial filtering to the real interference environment.

[0062] This invention provides an embodiment of a matrix-type audio life detection system, which includes: The deployment module is used to deploy a matrix audio sensor array in a distributed manner in the detection area, forming a two-dimensional planar array element layout, and synchronously acquire multi-channel ruin environment audio signals through the matrix audio sensor array. The extraction module is used to apply adaptive spatial decorrelation processing based on the environmental noise cross-spectrum matrix to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, and at the same time extract passive noise components from the multi-channel ruin environmental audio signal. The calibration module is used to calibrate the acoustic field transfer function matrix corresponding to the buried medium in the detection area in real time using the passive noise component; The mapping module is used to perform joint sparse deconvolution beamforming on the multi-channel denoised intermediate signal and the sound field transfer function matrix, and to map the multi-channel denoised intermediate signal to a preset sound source space grid to obtain a high-resolution sound source space spectrum distribution that varies with time and frequency. The filtering module is used to filter candidate vital sign sound sources in the high-resolution sound source spatial spectrum distribution based on the acoustic fingerprint parameters of vital signs, and to extract the spatial orientation parameters of the candidate vital sign sound sources. The filtering module is used to construct an adaptive spatial filter constrained by spatial notch filtering based on the spatial orientation parameters and the two-dimensional planar array element layout, and to perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal to separate the enhanced vital sign target signal. The generation module is used to apply time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation to the enhanced vital sign target signal, to recover the weak signal components attenuated due to burial depth, and to generate the detection result output.

[0063] The technical solution of this embodiment deploys a matrix audio sensor array to synchronously acquire multi-channel signals, applies adaptive spatial decorrelation processing and extracts passive noise components to calibrate the sound field transfer function matrix, performs sparse deconvolution beamforming to obtain a high-resolution sound source spectrum and obtains azimuth parameters based on acoustic fingerprints, and constructs an adaptive spatial filter constrained by spatial notch filtering for collaborative filtering, time-spectrum subtraction and dynamic gain compensation. This solves the problems of insufficient positioning accuracy and difficulty in extracting weak signals caused by interference from environmental noise, structural reflected sound waves and multipath effects in existing equipment, thereby improving the high-sensitivity identification of vital signs in deep burial sites, accurate azimuth determination capabilities, decision-making efficiency within the golden search and rescue time, and search and rescue success rate.

[0064] Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0065] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0066] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and do not imply a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0067] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A matrix-based audio life detection method, characterized in that, The method includes: A matrix-type audio sensor array is deployed in a distributed manner in the detection area to form a two-dimensional planar array element layout. Multi-channel ambient audio signals of the ruins are synchronously acquired through the matrix-type audio sensor array. An adaptive spatial decorrelation process based on the environmental noise cross-spectrum matrix is ​​applied to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, while passive noise components are extracted from the multi-channel ruin environmental audio signal. Using the passive noise component, the acoustic field transfer function matrix corresponding to the buried medium within the detection area is calibrated in real time; Joint sparse deconvolution beamforming is performed on the multi-channel denoised intermediate signal and the sound field transfer function matrix to map the multi-channel denoised intermediate signal onto a preset sound source spatial grid, thereby obtaining a high-resolution sound source spatial spectrum distribution that varies with time and frequency. Based on the acoustic fingerprint parameters of vital signs, candidate vital sign sound sources are screened from the spatial spectrum distribution of the high-resolution sound sources, and the spatial orientation parameters of the candidate vital sign sound sources are extracted. Based on the spatial orientation parameters and the two-dimensional planar array element layout, an adaptive spatial filter constrained by spatial notch filtering is constructed to perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal, thereby separating the enhanced vital sign target signal. Time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation are applied to the enhanced vital sign target signal to recover the weak signal components attenuated due to burial depth and generate detection result output.

2. The matrix-based audio life detection method according to claim 1, characterized in that, The two-dimensional planar array element layout of the matrix audio sensor array is a non-uniformly spaced concentric ring array that unfolds along the surface projection plane of the detection area. The circumferential spacing of the array elements on adjacent rings decreases as the ring radius increases, so as to ensure the uniformity of spatial sampling in all azimuth angles.

3. The matrix-type audio life detection method according to claim 2, characterized in that, In the matrix audio sensor array, the main axis direction of each element is tilted outward at a preset angle according to the center normal vector of the two-dimensional planar array element layout, so that the directional sensitive area covers the preset depth range of the buried medium in the detection area.

4. The matrix-type audio life detection method according to claim 1, characterized in that, The step of applying adaptive spatial decorrelation processing based on the environmental noise cross-spectral matrix to the multi-channel ruin environment audio signal to obtain a multi-channel denoised intermediate signal includes: The multi-channel ruin environment audio signal is divided into frames, the noise cross-spectrum matrix of each frame is estimated, and the noise cross-spectrum matrix is ​​decomposed. The eigenvectors corresponding to the eigenvalues ​​that are less than the rank threshold of the environmental noise diffusion field are selected to form a noise subspace projection matrix. The noise subspace projection matrix is ​​used to spatially project the multi-channel ruin environment audio signal to remove coherent noise components, thereby obtaining the multi-channel denoised intermediate signal.

5. The matrix-type audio life detection method according to claim 4, characterized in that, The step of extracting passive noise components from the multi-channel ruin environment audio signal includes: The multi-channel ruin environment audio signal is projected onto the noise subspace, and the resulting projection coefficient sequence is used as the passive noise component.

6. The matrix-based audio life detection method according to claim 1, characterized in that, The joint sparse deconvolution beamforming maps the multi-channel denoised intermediate signal to the preset sound source space grid by solving the following optimization problem: in, This represents the intermediate signal matrix for the multi-channel denoising. This represents the sound field transfer function matrix. This represents the high-resolution sound source spatial spectrum distribution matrix to be solved. Represents the residual noise matrix. Describing the Frobenius norm, for Mixture norm used for excitation row sparsity, This represents the weighted nuclear norm, and the weighting vector is determined by the prior distribution of the singular values ​​of the residual noise. and This is the regularization parameter.

7. The matrix-type audio life detection method according to claim 6, characterized in that, The step of screening candidate vital sign sound sources in the high-resolution sound source spatial spectrum distribution based on vital sign acoustic fingerprint parameters includes: The high-resolution sound source spatial spectrum distribution matrix along the time dimension A sliding window is used for truncation, and the time-frequency sparsity features of the spatial spectrum within each sliding window are calculated. The time-frequency sparsity features are matched and tracked with the periodic energy fluctuation patterns represented in the acoustic fingerprint parameters of vital signs. Spatial grid points with matching degrees exceeding the adaptive threshold are marked as candidate vital sign sound sources.

8. The matrix-type audio life detection method according to claim 1, characterized in that, The steps for constructing an adaptive spatial filter constrained by spatial notch filtering include: The target steering vector is determined based on the spatial orientation parameters. An array manifold is generated based on the two-dimensional planar array element layout. Multiple interference notch steering vectors at non-target orientations are selected to form a notch constraint matrix. The optimal weight vector of the adaptive spatial filter is obtained by solving the following constraint optimization problem: in, This represents the weight vector of the adaptive spatial domain filter. This represents the covariance matrix of the intermediate signal in the multi-channel denoising. This represents the notch constraint matrix. Represents the constraint response vector. for Norms are used to enhance the sparsity of weight vectors. For sparse regularization parameters, This indicates the conjugate transpose.

9. The matrix-type audio life detection method according to claim 8, characterized in that, The multiple interference notch steering vectors at the non-target azimuth are obtained by selecting strong interference azimuths in the high-resolution sound source spatial spectrum distribution, excluding the grid point where the candidate vital sign sound source is located.

10. A matrix-type audio life detection system, characterized in that, The system includes: The deployment module is used to deploy a matrix audio sensor array in a distributed manner in the detection area, forming a two-dimensional planar array element layout, and synchronously acquire multi-channel ruin environment audio signals through the matrix audio sensor array. The extraction module is used to apply adaptive spatial decorrelation processing based on the environmental noise cross-spectrum matrix to the multi-channel ruin environmental audio signal to obtain a multi-channel denoised intermediate signal, and at the same time extract passive noise components from the multi-channel ruin environmental audio signal. The calibration module is used to calibrate the acoustic field transfer function matrix corresponding to the buried medium in the detection area in real time using the passive noise component; The mapping module is used to perform joint sparse deconvolution beamforming on the multi-channel denoised intermediate signal and the sound field transfer function matrix, and to map the multi-channel denoised intermediate signal to a preset sound source space grid to obtain a high-resolution sound source space spectrum distribution that varies with time and frequency. The filtering module is used to filter candidate vital sign sound sources in the high-resolution sound source spatial spectrum distribution based on the acoustic fingerprint parameters of vital signs, and to extract the spatial orientation parameters of the candidate vital sign sound sources. The filtering module is used to construct an adaptive spatial filter constrained by spatial notch filtering based on the spatial orientation parameters and the two-dimensional planar array element layout, and to perform multi-channel collaborative spatial filtering on the multi-channel denoised intermediate signal to separate the enhanced vital sign target signal. The generation module is used to apply time-frequency domain adaptive spectral subtraction and dynamic playback gain compensation to the enhanced vital sign target signal, to recover the weak signal components attenuated due to burial depth, and to generate the detection result output.

Images