Lung ultrasonic signal feature identification method based on RF signal

Abstract

The invention discloses a lung ultrasonic signal feature identification method based on an RF signal, and relates to the technical field of feature identification, and the method comprises the steps: collecting an original lung ultrasonic radio frequency signal, extracting the local frequency spectrum energy distribution of the original lung ultrasonic radio frequency signal in a frequency sub-band, and forming a frequency-time energy matrix; calculating a resonance period vector of a local energy peak value according to the frequency-time energy matrix, and constructing a lung texture dynamic period sensing map through multi-scale fusion mapping; the method comprises the following steps: performing cepstrum transformation on an original lung ultrasonic radio-frequency signal to generate an original radio-frequency cepstrum spectrum, constructing a multi-path reflection structure model based on the original radio-frequency cepstrum spectrum, and then aligning periodic structure candidate features in a lung texture dynamic periodic sensing spectrum with the original radio-frequency cepstrum spectrum to obtain a time domain alignment error; fusion modeling of a periodic change mode of an original lung ultrasonic radio-frequency signal under a multi-frequency sub-band is realized through a step of constructing a lung texture dynamic periodic sensing map.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of feature recognition, in particular to a lung ultrasonic signal feature recognition method based on RF signals. BACKGROUND

[0002] In the field of medical ultrasound, radio frequency signals are widely used to extract tissue structure features and analyze tissue state changes due to their high fidelity characteristics of preserving complete acoustic information. The current commonly used method mainly performs frequency domain transformation and time domain processing on RF signals to extract local spectral energy features, reflection delay parameters or cepstrum features, respectively, for constructing response feature representations of lung tissue. For example, the energy distribution of frequency subbands can be analyzed based on short-time Fourier transform, or the delay response in the multi-path reflection structure can be identified using the cepstrum atlas, so as to reflect the complex morphology of gas-tissue interface structure. The conventional lung feature recognition method based on RF signals has been studied and applied in lung lesion recognition, tissue interface change monitoring and layered structure analysis, supporting the further development of RF signals in the field of lung ultrasound assisted diagnosis.

[0003] However, the existing feature recognition strategy based on RF signals still has limitations in two key aspects. On the one hand, the joint modeling mechanism between the frequency dimension and the time dimension is insufficient, making it difficult to effectively capture the dynamic coupling relationship between the periodic structural changes of the lung and the multi-path reflection features. On the other hand, the current processing means are mostly based on single-dimensional features, lacking comprehensive modeling methods that integrate structural features, dynamic difference factors and reflection path delays, which limits the continuous quantitative ability of the occurrence probability of structural instability state. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a lung ultrasonic signal feature recognition method based on RF signals to solve the problems of insufficient dynamic coupling modeling of lung structure periodic changes and multi-path reflection, and difficulty in quantitative probability of structural instability state.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a lung ultrasonic signal feature recognition method based on RF signals, which comprises, acquiring an original lung ultrasonic radio frequency signal and extracting the local spectral energy distribution of the original lung ultrasonic radio frequency signal in the frequency subband to form a frequency-time energy matrix; calculating a resonance period vector of the local energy peak value according to the frequency-time energy matrix, and constructing a lung texture dynamic period perception atlas through multi-scale fusion mapping; The cepstrum transformation is performed on the original lung ultrasonic radio frequency signal to generate an original radio frequency cepstrum spectrum, and a multi-path reflection structure model is constructed based on the original radio frequency cepstrum spectrum; subsequently, the period structure candidate features in the lung texture dynamic period sensing spectrum are aligned with the original radio frequency cepstrum spectrum, and a time domain alignment error is obtained; Based on the period structure candidate features in the lung texture dynamic period sensing spectrum, the time domain alignment error, and the reflection path parameters in the multi-path reflection structure model, a structure-dynamic coupled lung texture feature vector is constructed; Based on the structure-dynamic coupled lung texture feature vector and the reflection delay abnormal information in the multi-path reflection structure model, a structure instability probability function reflecting the lung lesion structure is constructed, and the structure instability probability values of different regions of the lung are calculated.

[0007] As a preferred scheme of the lung ultrasonic signal feature recognition method based on RF signals according to the present application, the local frequency spectrum energy distribution of the original lung ultrasonic radio frequency signal in the frequency subband is extracted to form a frequency-time energy matrix, and the steps are as follows, A radio frequency receiving device is used to collect continuous time domain radio frequency signals as an original lung ultrasonic radio frequency signal sequence; A windowed short-time Fourier transform is performed on the original lung ultrasonic radio frequency signal sequence to construct a time-frequency complex value matrix; A plurality of non-uniform frequency subbands are divided according to the radio frequency band, and the local energy distribution of each time frame in each frequency subband is calculated to form a frequency subband energy matrix through frame-level accumulation; The frequency subband energy matrix is normalized, and the normalized frequency subband energy matrix is interpolated and expanded to generate a frequency-time energy matrix.

[0008] As a preferred scheme of the lung ultrasonic signal feature recognition method based on RF signals according to the present application, the local energy peak value resonance period vector is calculated from the frequency-time energy matrix, and the lung texture dynamic period sensing spectrum is constructed through multi-scale fusion mapping, and the steps are as follows, The time position of the local peak value of each frequency subband energy sequence in the frequency-time energy matrix is extracted, and then based on the time interval between adjacent local peak values, an energy peak value time interval sequence set is generated; The time interval between adjacent peak values in the energy peak value time interval sequence set is counted to form a local resonance period distribution, and the local resonance period distribution is normalized to generate a resonance period vector set; A frequency subband weight model is constructed according to the relative position of the frequency subband, and the resonance period vector set is weighted and fused according to the preset weight and the weight model to generate a fused overall resonance period vector; The response intensity distribution of each frequency sub-band energy sequence in the frequency-time energy matrix is calculated, and the response intensity distribution is uniformly resampled and spatially aligned, and then the integrated overall resonance period vector is combined with the response intensity distribution to construct a lung texture dynamic period perception atlas.

[0009] As a preferred scheme of the identification method of the lung ultrasound signal feature based on the RF signal according to the present application, wherein: the lung texture dynamic period perception atlas takes the frequency sub-band index as the vertical axis, and takes the standard time sampling point sequence after uniform scale resampling as the horizontal axis, and the value of each position in the lung texture dynamic period perception atlas represents the normalized period response intensity value of the corresponding frequency sub-band at the time position.

[0010] As a preferred scheme of the identification method of the lung ultrasound signal feature based on the RF signal according to the present application, wherein: the cepstrum transformation is performed on the original lung ultrasound RF signal to generate an original RF cepstrum atlas, and a multi-path reflection structure model is constructed based on the original RF cepstrum atlas, and then the period structure candidate feature in the lung texture dynamic period perception atlas is aligned with the original RF cepstrum atlas to obtain a time domain alignment error, and the steps are as follows, The original lung ultrasound RF signal is converted into an original lung ultrasound RF signal frequency domain amplitude spectrum, and then inverse Fourier transform is performed to generate a cepstrum sequence, and then the original RF cepstrum atlas is obtained by recombining; The delay structure identification is performed on the original RF cepstrum atlas to obtain the local response position of the original RF cepstrum atlas, and then time clustering and energy aggregation are performed to form a multi-path reflection structure model; The period response feature position and the period value of the frequency sub-band in the lung texture dynamic period perception atlas are recorded and combined to form a period structure candidate feature set; Based on the period structure candidate feature set, the cepstrum response position with the minimum event position deviation of the period value is located in the frequency sub-band corresponding to the original RF cepstrum atlas, and compared with the main path delay in the multi-path reflection structure model to generate a time domain alignment error.

[0011] As a preferred scheme of the identification method of the lung ultrasound signal feature based on the RF signal according to the present application, wherein: based on the period structure candidate feature in the lung texture dynamic period perception atlas, the time domain alignment error, and the reflection path parameter in the multi-path reflection structure model, a structure-dynamic coupled lung texture feature vector is constructed, and the steps are as follows, Based on the period structure candidate feature in the lung texture dynamic period perception atlas and the reflection path parameter in the multi-path reflection structure model, a matching mapping table is constructed; Based on the matching mapping table, the dominant period value is extracted and compared with the corresponding frequency sub-band position in the original RF cepstrum atlas, and the time error is calculated as a dynamic difference measurement factor. Based on the periodic structure candidate features, the reflection path parameters and the dynamic difference measurement factors, a three-dimensional joint vector is constructed, and is sequentially spliced in frequency order to form an original lung texture structure-dynamic coupling vector set; The original lung texture structure-dynamic coupling vector set is normalized, and then a fixed embedding code table is introduced for splicing to form a structure-dynamic coupling lung texture feature vector.

[0012] As a preferred scheme of the method for identifying lung ultrasound signal features based on RF signals, wherein: based on the structure-dynamic coupling lung texture feature vector and the reflection delay abnormal information in the multi-path reflection structure model, a lung lesion structure instability probability function is constructed, and the structure instability probability values of different regions of the lung are calculated, and the steps are as follows, According to the frequency sub-band index, the sub-band features in the structure-dynamic coupling lung texture feature vector are matched and screened with the reflection delay abnormal information corresponding to the sub-band features to obtain a screened fusion feature subset; Based on the screened fusion feature subset, a structure instability probability mapping model is established; The structure instability probability mapping model is parameter trained and regularly optimized using a labeled training sample to generate a trained lung structure instability probability mapping model; The collected structure-dynamic coupling lung texture feature vector is substituted into the trained lung structure instability probability mapping model to construct a lung lesion structure instability probability function, and the structure instability probability values of different regions of the lung are calculated.

[0013] As a preferred scheme of the method for identifying lung ultrasound signal features based on RF signals, wherein: the labeled training sample refers to a sample data set used in the training of the structure instability probability mapping model, wherein each sample data is composed of a group of structure-dynamic coupling lung texture feature vectors and corresponding structure instability state labels.

[0014] In a second aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program is executed by the processor to implement any step of the method for identifying lung ultrasound signal features based on RF signals according to the first aspect of the present application.

[0015] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement any step of the method for identifying lung ultrasound signal features based on RF signals according to the first aspect of the present application.

[0016] The present application has the beneficial effects that: through the step of constructing a lung texture dynamic cycle perception atlas, the fusion modeling of the periodic change mode of the original lung ultrasonic radio frequency signal under a plurality of frequency subbands is realized, the distribution of the periodic structure in different subbands and the dynamic intensity thereof are effectively depicted, and therefore the periodic structure candidate information which is stable and has time-frequency linkage characteristics is provided, thereby providing clear organization cycle atlas support for subsequent structure level recognition. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 The flowchart of the method for recognizing the lung ultrasonic signal features based on RF signals.

[0019] Fig. 2 The flowchart of generating a frequency-time energy matrix.

[0020] Fig. 3 The flowchart of constructing a lung texture dynamic cycle perception atlas.

[0021] Fig. 4 The flowchart of generating a structure-dynamic coupling feature vector. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification.

[0023] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0024] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment which is mutually exclusive with other embodiments.

[0025] REFERENCE Figs. 1-4 For one embodiment of the present application, the embodiment provides a method for recognizing lung ultrasonic signal features based on RF signals, comprising the following steps: S1: Collect the original lung ultrasound radio frequency signal and extract the local spectrum energy distribution of the original lung ultrasound radio frequency signal in the frequency sub-band to form a frequency-time energy matrix.

[0026] Specifically, the following steps are included: S1.1: Use a radio frequency receiving device to collect continuous time-domain radio frequency signals as the original lung ultrasound radio frequency signal sequence.

[0027] Specifically, by setting the working frequency band range of the RF receiving device, the frequency range covered by the collected continuous time-domain RF signal is determined, and then the sampling frequency is set. During the actual acquisition process, the continuous time-domain RF signal within the working frequency band is continuously received through the receiving antenna of the RF receiving device, and the received continuous time-domain RF signal is converted from analog to digital to form an original lung ultrasound RF signal sequence.

[0028] It should be explained that the continuous time-domain RF signal includes a set of signals composed of multiple physical propagation mechanisms and RF reflection paths, mainly including: the reflection signal of the external RF excitation signal by the lung structure, multipath reflection signal, environmental background interference signal and equipment background electronic noise signal.

[0029] It should be explained that the sampling frequency is set in combination with the time-varying characteristics of the continuous time-domain RF signal and the analysis accuracy requirements, and the actual sampling frequency is further set. For example, the sampling frequency can be set to a fixed value between 10MHz and 100MHz to achieve high time resolution sampling of the continuous time-domain RF signal.

[0030] Among them, analog-to-digital conversion refers to controlling the sampling timing through a high-stability clock source, and periodically sampling the continuous time-domain RF signal according to the set sampling frequency: for example, when the sampling frequency is set to 50MHz, periodic sampling is performed every 20 nanoseconds to form discrete sample points with equal time intervals. Subsequently, the instantaneous amplitude of the analog signal corresponding to each discrete sample point is mapped to a digital code value limited by a preset number of quantization bits. Finally, the discrete amplitude codes arranged in chronological order are encoded to form the original lung ultrasound RF signal sequence.

[0031] S1.2: Perform a windowed short-time Fourier transform on the original lung ultrasound RF signal sequence to construct a time-frequency complex matrix.

[0032] Specifically, according to the set frame length parameter and frame shift length, the original lung ultrasound radio frequency signal sequence is divided into a plurality of adjacent or overlapping signal frames, then a windowing operation is performed on each frame of the original lung ultrasound radio frequency signal sequence, for example, a Hamming window or a Hanning window is used to weight the signals in the frame to suppress the spectral leakage effect, and a fast Fourier transform is performed on each windowed signal frame to obtain complex form spectrum coefficients of the signal frame in the frequency domain, and finally, the complex form spectrum coefficients of all signal frames in the frequency domain are sequentially spliced and arranged along the time sequence to form a time-frequency complex value matrix.

[0033] It should be explained that the fast Fourier transform refers to taking the signal frame weighted by the windowing function as the input sequence, using the divide-and-conquer strategy to recursively divide the input sequence into a plurality of subsequences, and using the periodicity and conjugate symmetry of the complex exponential rotation factor to reduce repeated multiplication and addition operations to obtain the complex form spectrum coefficients of the signal frame in the frequency domain.

[0034] S1.3: According to the division of the preset plurality of non-uniform frequency subbands, the local energy distribution of each time frame in each frequency subband is calculated by a short-time energy spectrum calculation method, and frame-level accumulation is performed to form a frequency subband energy matrix.

[0035] Specifically, according to the division of the preset plurality of non-uniform frequency subbands, the amplitude square calculation is performed on each column of complex form spectrum coefficients in the time-frequency complex value matrix to obtain a short-time energy spectrum, and according to the frequency index range corresponding to each non-uniform frequency subband, the spectrum coefficient energy values belonging to each non-uniform frequency subband in the time-frequency complex value matrix are extracted, then the spectrum coefficient energy values are weighted and averaged to obtain the local energy distribution of each information frame in each non-uniform frequency subband, and finally, all the local energy distributions are sequentially accumulated and arranged at the frame level according to the time sequence to form a frequency subband energy matrix.

[0036] The amplitude square calculation formula is as follows, ; Wherein, represents the time frame index, represents the frequency domain index, represents the short-time energy spectrum at the frame and the frequency position, represents the complex form spectrum coefficient in the time-frequency complex value matrix at the frame and the frequency position, represents the real part of the complex number, represents the imaginary part of the complex number. ​​

[0037] It should be explained that the preset plurality of non-uniform frequency subbands are divided based on the energy distribution characteristics of the lung ultrasound radio frequency signal in different depth tissues, and a nonlinear division strategy is adopted: a narrower frequency interval is adopted in the low frequency band (for example: 0.5 MHz-3 MHz) to enhance the resolution ability of the reflection characteristics of the fine structure of the shallow tissue, and a wider frequency interval is adopted in the high frequency band (for example: 3 MHz-10 MHz) to improve the coverage efficiency of the echo energy of the deep tissue, thereby forming a plurality of subbands with different frequency spans.

[0038] S1.4: Normalizing the frequency subband energy matrix, and interpolating and expanding the normalized frequency subband energy matrix to generate a frequency-time energy matrix.

[0039] Specifically, based on each time frame in the frequency subband energy matrix, the minimum value and the maximum value of the corresponding frequency subband energy value are calculated respectively, and the linear normalization method is used to obtain the normalized frequency subband energy value. Subsequently, the normalized frequency subband energy value is used as interpolation input data, interpolation expansion is performed along the frequency dimension, the energy value on the original non-uniform frequency subband is mapped to the set equal-interval target frequency grid, a normalized energy distribution vector with consistent frequency resolution is generated, and finally the frequency-time energy matrix is arranged in time sequence.

[0040] The linear normalization formula is as follows, ; Wherein, denotes the time frame index, denotes the frequency subband index, denotes the traversal index of the frequency subband, denotes the frequency subband energy value at the th time frame and the th frequency subband position in the normalized frequency subband energy matrix, denotes the frequency subband energy value at the th time frame and the th frequency subband position in the frequency subband energy matrix before normalization, denotes the frequency subband energy value at the th time frame and the th frequency subband position in the frequency subband energy matrix, denotes the maximum energy value generated after traversal of the frequency subband energy value in the th time frame, denotes the minimum energy value generated after traversal of the frequency subband energy value in the th time frame.

[0041] It needs to be explained that the interpolation expansion refers to that in the normalized frequency sub-band energy matrix, the normalized energy value under each time frame is filled along the frequency dimension according to the order of the center frequency corresponding to the non-uniform frequency sub-band, and the normalized energy value is rearranged according to the frequency position by comparing with the preset equal-interval target frequency network, to generate a normalized energy distribution vector with consistent frequency resolution.

[0042] Wherein, the preset equal-interval target frequency network is preset by setting the minimum receiving frequency and the maximum receiving frequency of the radio frequency receiving device, and combining the required frequency resolution parameter, for example: when the minimum receiving frequency is 100MHz, the maximum receiving frequency is 200MHz, and the required frequency resolution is 10MHz, the equal-interval target frequency grid constructed is 100MHz, 110MHz, 120MHz, …, 200MHz.

[0043] S2: Calculate the resonance period vector of the local energy peak value according to the frequency-time energy matrix, and construct a lung texture dynamic cycle perception atlas through multi-scale fusion mapping.

[0044] Specifically, it includes the following steps, S2.1: Local maximum value detection operation is performed on each frequency sub-band energy sequence in the frequency-time energy matrix, the time position of the local peak value is extracted, and then the time interval corresponding to the adjacent local peak value is calculated based on the time position, to generate a set of energy peak time interval sequence corresponding to each frequency sub-band.

[0045] Specifically, the local maximum value detection operation is performed on each frequency sub-band energy sequence in the frequency-time energy matrix in turn, the local peak point is identified by comparing the frequency sub-band energy value of the current time frame with the frequency sub-band energy value of the adjacent time frames before and after it, for example: if the energy value of the 10th frame in a certain frequency sub-band energy sequence is greater than that of the 9th frame and the 11th frame, the time position corresponding to the 10th frame is identified as the time position of the local peak value. Then, the frame interval number between the adjacent two local peak points is calculated in time sequence, for example: the interval between the 10th frame and the 25th frame is 15 frames, and the frame interval number is converted into a specific time interval value combined with the sampling time step, to form a set of energy peak time interval sequence corresponding to each frequency sub-band.

[0046] It needs to be explained that the local maximum value detection operation refers to that in the frequency-time energy matrix, the frequency sub-band energy values of adjacent frames are compared point by point along the time axis direction for the normalized energy distribution vector, and the position with larger frequency sub-band energy value between the adjacent frames is identified as the local peak value.

[0047] S2.2: For each set of time point sequences in the energy peak time interval sequence set, the time intervals between adjacent peaks are counted to form a local resonance period distribution for each frequency subband. The local resonance period distribution is then normalized to generate a resonance period vector set corresponding to each frequency subband.

[0048] Specifically, for the local peak time index sequence corresponding to each group of frequency subbands in the energy peak time interval sequence set, the index difference between two adjacent local peak time indexes is calculated in time order, and the index difference is converted into actual time length in combination with the sampling time step, as the energy peak time interval sequence corresponding to the frequency subband. Subsequently, based on the energy peak time interval sequence, the frequency distribution of the energy peak time interval is statistically analyzed to construct the local resonance period distribution of the frequency subband, and the linear normalization method is used to convert the local resonance period distribution of the frequency subband into the normalized local resonance period distribution of the frequency subband, so that the numerical range of the normalized local resonance period distribution of the frequency subband is controlled within the interval [0,1], and finally the resonance period vector set is generated by combining them.

[0049] It should be explained that the sampling time step refers to the length of the interval on the time axis between two adjacent sampling points when sampling a continuous time-domain RF signal.

[0050] S2.3: Construct a frequency subband weight model according to the relative positions of the frequency subbands, and perform weighted fusion on the set of resonance period vectors according to the preset weights and the weight model to generate a fused overall resonance period vector.

[0051] Specifically, a frequency subband weight model is constructed according to the relative position of each frequency subband within the set frequency range, and the frequency subband weight model is set in a position-related manner. Subsequently, each resonance periodic vector in the resonance periodic vector set corresponding to the frequency subband is weighted and added together according to the corresponding frequency subband weight value in the frequency subband weight model to generate an overall resonance periodic vector.

[0052] It should be explained that the frequency subband weight model constructs an initial indicator representing the importance of the frequency subband based on the statistical characteristics such as periodic stability, energy significance or peak density of each frequency subband during the resonance period extraction process. The initial indicator is then used as an input feature, and the normalization method is used to map the initial indicator within a uniform numerical range to form an initial weight vector. Subsequently, the matching error between the frequency subband resonance period vector in the historical sample and the global target period is combined, and the frequency subband weight value is optimized and adjusted by minimizing the error, and finally the optimized frequency subband weight model is generated.

[0053] It needs to be explained that the historical sample is a set of radio frequency signal samples that have completed resonance period extraction and have annotation information under the same or similar radio frequency signal collection environment, wherein the historical sample is derived from existing experimental data, typical scene records or artificially annotated resonance period reference data, which includes frequency-time energy matrix, frequency sub-band energy distribution, energy peak time interval sequence and target period reference value associated with actual physical phenomenon.

[0054] It needs to be explained that the matching error between the global target period is that the integrated overall resonance period vector is compared with the target period reference value corresponding to the historical sample, and the difference between the integrated overall resonance period vector and the target period reference value in the numerical space is obtained.

[0055] S2.4: Adopting cycle modulation filtering mode, calculating the response intensity distribution of each frequency sub-band energy sequence in the frequency-time energy matrix, then uniformly resampling and spatially aligning the response intensity distribution, and combining the integrated overall resonance period vector with the response intensity distribution to represent, and constructing a complete lung texture dynamic period perception atlas.

[0056] Specifically, adopting cycle modulation filtering mode, each frequency sub-band energy sequence in the frequency-time energy matrix is subjected to sliding convolution operation, the filtering response value under different period lengths is obtained, and the maximum response value in the filtering response value under all period lengths corresponding to each frequency sub-band in the frequency sub-band energy sequence is extracted as the response intensity distribution of the frequency sub-band. Then, a uniform scale resampling method is used to standardize all response intensity distributions to a consistent time length, and according to the arrangement order of the frequency sub-band on the frequency axis, all normalized response intensity distributions corresponding to the frequency sub-band are spatially aligned and sequentially spliced to construct a complete lung texture dynamic period perception atlas.

[0057] It needs to be explained that the sliding convolution operation is to take a weight sequence with a preset period length as a convolution kernel to perform sliding window operation on the frequency sub-band energy sequence. At each sliding position, the energy values in the current window are multiplied by the corresponding weight values element by element and summed to obtain the filtering response value at the current time position.

[0058] Among them, the preset period length is a set of representative time period length sets defined in advance according to the prior knowledge of the lung breathing period characteristics or the possible periodic change range in the target scene before performing the cycle modulation filtering operation. These period lengths can be set within a reasonable value range according to actual application requirements, for example, set to a plurality of equally spaced or non-uniformly spaced period length values ranging from 0.5 seconds to 5 seconds.

[0059] It needs to be explained that the unified scale resampling method refers to constructing equal-interval resampling time coordinates within a set target length range (for example: the standardized target length range is 128 time sampling points) based on the time index of the original response intensity distribution, and simultaneously using linear difference, cubic spline interpolation and other methods to interpolate and estimate the original response intensity value, so that the response intensity distribution of each frequency sub-band is mapped to the same time scale.

[0060] S3: performing cepstrum transformation on the original lung ultrasound radio frequency signal to generate an original radio frequency cepstrum spectrum, and constructing a multi-path reflection structure model based on the original radio frequency cepstrum spectrum, then aligning the period structure candidate features in the lung texture dynamic period perception spectrum with the original radio frequency cepstrum spectrum, and obtaining a time domain alignment error.

[0061] Specifically, the following steps are included, S3.1: performing Fourier transform on the original lung ultrasound radio frequency signal to convert it into an original lung ultrasound radio frequency signal frequency domain amplitude spectrum, and performing inverse Fourier transform to generate a cepstrum sequence, then performing time window framing and frequency sub-band recombination to obtain an original radio frequency cepstrum spectrum.

[0062] Specifically, Fourier transform is used to convert the original lung ultrasound radio frequency signal from time domain to frequency domain to obtain an original lung ultrasound radio frequency signal frequency domain amplitude spectrum, and logarithmic transformation is performed on the original lung ultrasound radio frequency signal frequency domain amplitude spectrum to form an original lung ultrasound radio frequency signal logarithmic amplitude spectrum, then inverse Fourier transform is performed on the original lung ultrasound radio frequency signal logarithmic amplitude spectrum to obtain a cepstrum sequence reflecting the change in the logarithmic spectrum structure, and time window framing operation is performed on the cepstrum sequence based on a set time window parameter to divide the continuous cepstrum sequence into several adjacent time frames, and finally the cepstrum components in each time frame are arranged and grouped according to the frequency sub-band recombination mode to generate an original radio frequency cepstrum spectrum containing time dimension, frequency sub-band dimension and cepstrum component dimension.

[0063] The Fourier transform formula is as follows, ; Wherein, represents a continuous time point on the time axis, represents a specific frequency position in the frequency domain, represents the time domain signal function value at time point , represents the Fourier transform of the function value , represents the complex value function value at frequency position after Fourier transform, represents the complex exponential kernel function, represents the imaginary unit, represents the product of the angular frequency and time.

[0064] It needs to be explained that the inverse Fourier transform is to restore the amplitude and phase information of the logarithmic amplitude spectrum of each original lung ultrasound radio frequency signal to a time-domain continuous signal by reconstructing the complex spectrum in the frequency domain, and recover the structural characteristics of the time-domain continuous signal in the time dimension.

[0065] S3.2: Adopting cepstrum peak detection method to identify the delay structure of the original radio frequency cepstrum spectrum, obtaining the local response position of the original radio frequency cepstrum spectrum, and then performing time clustering and energy aggregation to form a multi-path reflection structure model.

[0066] Specifically, the peak detection operation is performed on the cepstrum component sequence in each frequency sub-band dimension of the original radio frequency cepstrum spectrum, and the position of the local maximum value is extracted as the candidate position of the possible multi-path reflection delay. Subsequently, based on the distribution characteristics of the candidate position in the time dimension, a sliding time window is used for local clustering, and the local maximum values with similar time positions are merged into the same delay structure unit. Based on the delay structure unit, the cepstrum amplitudes corresponding to all candidate positions in the delay structure unit are weighted and averaged to obtain the aggregated delay response energy. Finally, the center time position of each delay structure unit and the aggregated energy value corresponding to the center time position are combined to generate a set of delay time and reflection intensity parameters containing multiple reflection paths, and are combined to form a multi-path reflection structure model.

[0067] It needs to be explained that the peak detection operation means that a certain data point in the cepstrum component sequence is regarded as a local maximum value, and when the amplitude value of the local maximum value is greater than the amplitude values of the previous and next points, it meets the trend of rising and then falling in amplitude. In order to improve the stability of peak detection, a minimum peak amplitude threshold and a minimum peak distance threshold can be set to filter the detection results.

[0068] For example: In a cepstrum component sequence, if the amplitude value of a certain sampling point is 0.82, and the amplitude values of its adjacent points are 0.75 and 0.79 respectively, then the point is judged as a local maximum value; if the interval between the point and the previous identified peak is less than 5 sampling points, then the higher one can be retained according to the amplitude value to suppress the false peak.

[0069] The minimum peak amplitude threshold and the minimum peak distance threshold are set in combination with the amplitude distribution characteristics and the signal-to-noise ratio level of the cepstrum component sequence in the original radio frequency cepstrum spectrum. The minimum peak amplitude threshold is used to eliminate non-significant peaks caused by background noise or weak disturbances, and is generally set according to a multiple of the average amplitude or median amplitude of the entire cepstrum component sequence, for example, the minimum peak amplitude threshold is set to 1.5 times the average amplitude of the cepstrum. The minimum peak distance threshold is used to prevent multiple approximate peaks from appearing at adjacent positions, causing repeated counting, and is generally set according to the sampling frequency and the minimum resolvable delay interval of the signal, for example, when the sampling frequency is 50MHz, the minimum peak distance threshold can be set to 5 sampling points, corresponding to a minimum delay resolution of 100 nanoseconds, ensuring that the detection result has sufficient distinguishability in the time dimension.

[0070] It needs to be explained that the training process of the multi-path reflection structure model is to take the structure-dynamic coupled lung texture feature vector corresponding to each frequency sub-band position as the input feature, take the structure instability label as the supervision signal, construct a training sample set, then optimize the regression parameters in the structure instability probability mapping model, and introduce regular optimization to obtain a structure instability probability mapping model that can accurately predict the structure instability probability.

[0071] S3.3: Record the period response feature position and period value of the frequency sub-band in the lung texture dynamic period perception spectrum, and combine to form a period structure candidate feature set.

[0072] Specifically, the lung texture dynamic period perception spectrum is traversed, and the normalized response intensity distribution corresponding to each frequency sub-band in the lung texture dynamic period perception spectrum is extracted in turn. The position where the amplitude reaches the local maximum in the normalized response intensity distribution is identified as the period response feature position of the frequency sub-band. Then, the period value corresponding to each period response feature position is obtained by combining the period length index list used in the generation of the normalized response intensity distribution. Finally, the frequency sub-band index, period response feature position and period value are combined in the order of the frequency sub-band to form a structured feature record, and all structured feature records are finally summarized in turn to form a period structure candidate feature set.

[0073] It needs to be explained that the period length index list is selected according to the prior knowledge of the lung dynamic period change range, combined with the actual sampling time interval and the physiological rhythm characteristics, and a number of representative period lengths are selected as target periods, for example, taking 0.6 milliseconds to 2.0 milliseconds as the range, and taking one period length every 0.1 millisecond to form a period length set. Then, each period length value is numbered in the order in the period length set to form a period length index list.

[0074] S3.4: Based on the periodic structure candidate feature set, find the closest periodic value in the frequency sub-band in the original radio frequency cepstrum spectrum, and compare it with the main path delay in the multi-path reflection structure model to generate the time domain alignment error.

[0075] Specifically, based on the periodic structure candidate feature set, the periodic response feature position corresponding to each frequency sub-band in the lung texture dynamic periodic perception spectrum is obtained, and the periodic value is taken as the basic feature item in the periodic structure candidate feature set. Then, in the original radio frequency cepstrum spectrum, the frequency sub-band corresponding to each periodic value in the periodic structure candidate feature set is selected, and the closest cepstrum index position to the periodic value is selected. The main path delay position in the multi-path reflection structure model is taken as the reference value, and based on the frequency sub-band cepstrum response position corresponding to each periodic value in the periodic structure candidate feature set in the original radio frequency cepstrum spectrum, the offset degree of each feature in the periodic structure candidate feature set in the time scale is calculated as the time domain alignment error.

[0076] S4: Based on the periodic structure candidate feature in the lung texture dynamic periodic perception spectrum, the time domain alignment error, and the reflection path parameter in the multi-path reflection structure model, a structure-dynamic coupled lung texture feature vector is constructed.

[0077] Specifically, the steps are as follows, S4.1: Based on the periodic structure candidate feature in the lung texture dynamic periodic perception spectrum and the reflection path parameter in the multi-path reflection structure model, a matching mapping table is constructed.

[0078] Specifically, based on the periodic structure candidate feature in the lung texture dynamic periodic perception spectrum and the reflection path parameter in the multi-path reflection structure model, the periodic length index, response intensity value and periodic position parameter corresponding to each frequency sub-band in the periodic structure candidate feature are extracted, and then the delay time, path energy and reflection intensity parameters of each reflection path in the multi-path reflection structure model are extracted. According to the arrangement order of the frequency sub-band on the frequency axis, the extracted two types of feature parameters are correspondingly numbered and structured, and at the same time, the mapping relationship between the periodic structure candidate feature and the reflection path parameter is constructed according to the frequency sub-band index consistency and the periodic value proximity. Then, based on the frequency sub-band index consistency, the similarity of the periodic value and the delay time, the periodic structure candidate feature and the reflection path parameter that meet the pairing condition are paired by taking, for example, the periodic difference value not exceeding the periodic-delay matching tolerance threshold as the judgment criterion. On the basis of completing the pairing, the corresponding frequency sub-band position index, periodic value in the periodic structure candidate feature and delay time in the multi-path reflection structure model in each matching pair are recorded in turn, and finally the matching mapping table with consistent frequency sub-band index dimension and corresponding periodic structure feature and reflection path parameter is formed.

[0079] It needs to be explained that the cycle-delay matching tolerance threshold is set according to the sampling accuracy of the original lung ultrasound radio frequency signal, the cycle response change range and the response resolution requirement of the target structure.

[0080] S4.2: Based on the matching items in the matching mapping table, the dominant cycle value in the time domain alignment error data is extracted, and the frequency sub-band position in the original radio frequency inverse spectrum is compared. The time error of the dominant cycle value and the frequency sub-band position is calculated, and the time error is taken as the coupling response deviation, which is defined as the dynamic difference measurement factor between the candidate feature and the structure path.

[0081] Specifically, based on the matching items in the matching mapping table, the cycle value in the cycle structure candidate feature corresponding to the matching mapping table is extracted, and the cycle value with the highest statistical frequency among all matching items is taken as the dominant cycle value. Subsequently, the time delay position corresponding to the dominant cycle value is searched in the original radio frequency inverse spectrum, and the difference value between the dominant cycle value and the main path delay time in the multi-path reflection structure model recorded in the matching mapping table is calculated. The alignment time error between the dominant cycle value and the main path delay time is obtained. Finally, the alignment time error is defined as the coupling response deviation between the cycle structure candidate feature and the reflection path in the multi-path reflection structure model, as a dynamic difference measurement factor for measuring the stability of the corresponding relationship between the cycle structure candidate feature and the reflection path.

[0082] S4.3: Based on the cycle structure candidate feature, the reflection path parameter and the dynamic difference measurement factor, a three-dimensional joint vector is constructed, and is sequentially spliced in frequency order to form an original lung texture structure-dynamic coupling vector set.

[0083] Specifically, based on the cycle structure candidate feature, the reflection path parameter in the multi-path reflection structure model and the dynamic difference measurement factor defined by the time domain alignment error, a three-dimensional time joint vector containing the cycle value, the reflection path delay time and the dynamic difference measurement factor is constructed for each frequency sub-band position by the time domain feature fusion method. Subsequently, according to the arrangement order of the frequency sub-band on the frequency axis, the three-dimensional joint vectors generated at all frequency sub-band positions are sequentially spliced in order, and finally an original lung texture structure-dynamic coupling vector set is formed.

[0084] It needs to be explained that the time domain feature fusion method refers to the joint coding of time domain related parameters (for example: cycle structure candidate feature, reflection path parameter and dynamic difference measurement factor) from different sources to generate a joint vector form with a unified structure format.

[0085] S4.4: Normalizing the original lung texture structure-dynamic coupling vector set, and introducing a fixed embedding code table to splice the normalized original lung texture structure-dynamic coupling vector set, and finally forming a structure-dynamic coupling lung texture feature vector.

[0086] Specifically, the period value, reflection path delay time and dynamic difference measure factor in the original lung texture structure-dynamic coupling vector set are linearly normalized to map the feature values in each dimension to a standardized numerical interval. Then, a preset fixed embedding code table is called to generate a corresponding embedding code vector for each frequency sub-band position. The embedding vectors in the fixed embedding code table are sequentially spliced with the vectors in the corresponding positions of the normalized original lung texture structure-dynamic coupling vector set according to the frequency sub-band position order. Finally, all the spliced position vectors are combined to form a structure-dynamic coupling lung texture feature vector.

[0087] It should be explained that the standardized numerical interval is obtained by calculating the minimum and maximum values of the period value, reflection path delay time and dynamic difference measure factor in the original lung texture structure-dynamic coupling vector set. Then, each original feature value is subtracted by the minimum value of the corresponding dimension and divided by the difference between the maximum and minimum values, so that all the feature values are converted to a unified standardized numerical interval.

[0088] For example, taking a three-dimensional time joint vector at a certain frequency sub-band position in the original lung texture structure-dynamic coupling vector set as an example, if the period value of the frequency sub-band is 240 milliseconds, the reflection path delay time is 75 milliseconds, and the dynamic difference measure factor is 15 milliseconds; and assuming that in the entire original lung texture structure-dynamic coupling vector set, the minimum value of the period value is 180 milliseconds, the maximum value is 300 milliseconds, the minimum value of the reflection path delay time is 50 milliseconds, the maximum value is 100 milliseconds, and the minimum value of the dynamic difference measure factor is 0 milliseconds, the maximum value is 20 milliseconds, then through the standardization operation, the three-dimensional time joint vector can be normalized to: the period value is normalized to 0.5, the reflection path delay time is normalized to 0.5, and the dynamic difference measure factor is normalized to 0.75. Finally, the normalized vector in the unified standardized numerical interval is [0.5, 0.5, 0.75].

[0089] It should be explained that the preset fixed embedding code table is a set of embedding vectors constructed according to the order of frequency sub-bands on the frequency axis, which is used to generate a unique and consistent code representation for each frequency sub-band position. The fixed embedding code table can be predefined in the form of position coding or regular interval numerical mapping.

[0090] S5: Based on the structure-dynamic coupling lung texture feature vector and the reflection delay abnormal information in the multi-path reflection structure model, a structure instability probability function reflecting the lung lesion structure is constructed.

[0091] Specifically, the steps are as follows, S5.1: According to the frequency sub-band index, the sub-band features in the structure-dynamic coupling lung texture feature vector are matched and screened with the reflection delay abnormal information corresponding to the sub-band features, and a screened fusion feature subset is obtained.

[0092] Specifically, according to the frequency sub-band index, the sub-band features corresponding to each frequency sub-band position in the structure-dynamic coupling lung texture feature vector are extracted in turn, and the extracted sub-band features are compared with the reflection delay abnormal information recorded in the multi-path reflection structure model corresponding to the frequency sub-band, and the frequency sub-band subset satisfying the abnormal response feature is retained through the set delay abnormality judgment condition, and finally all the sub-band features satisfying the screening condition are combined in the original frequency sub-band order to form the screened fusion feature subset.

[0093] It needs to be explained that the set delay abnormality judgment condition is set by analyzing the reflection path delay time corresponding to each frequency sub-band in the structure-dynamic coupling lung texture feature vector and the delay time distribution of the same frequency sub-band in the normal sample, and setting a threshold interval deviating from the mean value by a certain range; For example: the threshold judgment condition is set to be 2 times the standard deviation deviating from the mean value, if the delay time of each frequency sub-band in the structure-dynamic coupling lung texture feature vector exceeds the threshold judgment condition, it is judged as delay abnormality.

[0094] S5.2: Based on the screened fusion feature subset, a structure instability probability mapping model is established by using generalized logistic regression method.

[0095] Specifically, using the generalized logistic regression method, the structure-dynamic coupling lung texture feature vector corresponding to each frequency sub-band position in the screened fusion feature subset is used as the input variable, and a binary state label is set according to whether the corresponding sample has a structure instability state as the target variable of logistic regression, to obtain the structure instability state label, and then the regression coefficient in the generalized logistic regression method is optimized by maximum likelihood estimation method, to construct the structure instability probability mapping model.

[0096] It needs to be explained that the maximum likelihood estimation method is to evaluate the consistency between the model output probability and the actual instability state label under different regression coefficient settings, and constantly adjust the regression coefficient to make the prediction result and the actual label as consistent as possible, and finally obtain the optimal set of regression coefficients.

[0097] S5.3: Parameter training and regularization optimization of the structural instability probability mapping model using labeled training samples, to generate the trained lung structural instability probability mapping model.

[0098] Specifically, for each frequency sub-band position, the feature vectors in the screened fusion feature subset are taken as input features, and the labeled results of whether there is a structural instability state are taken as supervision signals to construct training samples. In the training stage, the regression parameters in the structural instability probability mapping model are optimized to make the structural instability probability predicted by the model most consistent with the structural instability state distribution labeled in the training samples in a statistical sense. Meanwhile, to prevent the lung structural instability probability mapping model from overfitting in the case of high feature dimension or limited training samples, a regularization optimization mechanism is introduced to constrain the regression parameters, thereby improving the generalization performance of the structural instability probability mapping model. Finally, after training and regularization optimization are completed, the trained lung structural instability probability mapping model is obtained.

[0099] It should be explained that the labeled training samples refer to the sample data used in the process of training the structural instability probability mapping model, and each sample data is composed of a group of structural-dynamic coupled lung texture feature vectors and corresponding structural instability state labels.

[0100] S5.4: Input the collected structural-dynamic coupled lung texture feature vectors into the trained lung structural instability probability mapping model, construct a lung lesion structural instability probability function, and calculate the structural instability probability values of different lung regions.

[0101] Specifically, the collected structural-dynamic coupled lung texture feature vectors are input into the trained lung structural instability probability mapping model, the structural-dynamic coupled lung texture feature vectors are predicted for structural instability probability according to the regression parameters in the lung structural instability probability mapping model, the structural instability probability values are obtained, then the structural instability probability values are arranged in the order of frequency sub-bands, and a probability function reflecting the overall structural instability trend of the lung lesion is constructed through weighted averaging and convolution integration. Finally, the structural instability probability values of different lung regions are calculated according to the lung lesion structural instability probability function.

[0102] The formula for calculating the structural instability probability values of different lung regions is as follows, ; Wherein, represents the current lung region index, represents the index of the convolution kernel, represents the length of the convolution kernel, represents the structural instability probability value of the i-th lung region, represents the structural instability probability value of the i-th lung region, represents the structural instability probability value of the i-th lung region, a weight coefficient of an element of a convolution kernel, during the convolution process, the index is a lung region structure instability probability value.

[0103] The embodiment also provides a computer device suitable for the recognition method of the lung ultrasonic signal feature based on an RF signal, which comprises a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the recognition method of the lung ultrasonic signal feature based on the RF signal proposed in the above embodiment.

[0104] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used for providing computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used for wired or wireless communication with external terminals. The wireless communication can be realized through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0105] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to realize the recognition method of the lung ultrasonic signal feature based on the RF signal proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0106] To sum up, the application realizes the fusion modeling of the periodic change mode of the original lung ultrasonic radio frequency signal under the multi-frequency sub-band by constructing the lung texture dynamic periodic perception atlas, effectively describes the distribution of the periodic structure and its dynamic strength in different sub-bands, thereby providing stable and time-frequency linkage characteristic periodic structure candidate information, and providing clear organization periodic atlas support for subsequent structure level recognition.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for identifying lung ultrasound signal features based on RF signals, characterized by: include, Acquiring original lung ultrasound radio frequency signals and extracting local spectrum energy distribution of the original lung ultrasound radio frequency signals in frequency sub-bands to form a frequency-time energy matrix; The resonance period vector of the local energy peak is calculated based on the frequency-time energy matrix, and a dynamic period perception map of lung texture is constructed through multi-scale fusion mapping. The original lung ultrasound RF signal is cepstral transformed to generate the original RF cepstral map, and a multipath reflection structure model is constructed based on the original RF cepstral map. The periodic structure candidate features in the dynamic periodic perception map of the lung texture are aligned with the original RF cepstral map to obtain the time domain alignment error. Based on the candidate periodic structure features in the dynamic periodic perception map of lung texture, the time domain alignment error, and the reflection path parameters in the multipath reflection structure model, a structure-dynamic coupled lung texture feature vector is constructed. Based on the structural-dynamic coupled lung texture feature vector and the reflection delay anomaly information in the multipath reflection structure model, a probability function reflecting the structural instability of lung lesions was constructed, and the structural instability probability values ​​of different lung regions were calculated.

2. The method for identifying lung ultrasound signal features based on RF signals according to claim 1, wherein: The steps of extracting the local spectrum energy distribution of the original lung ultrasound radio frequency signal in the frequency sub-band to form a frequency-time energy matrix are as follows: A radio frequency receiving device is used to collect continuous time-domain radio frequency signals as original lung ultrasound radio frequency signal sequences; Perform windowed short-time Fourier transform on the original lung ultrasound radiofrequency signal sequence to construct a time-frequency complex matrix; Divide the RF band into multiple pre-set non-uniform frequency sub-bands, calculate the local energy distribution within each frequency sub-band for each time frame, and perform frame-level accumulation to form a frequency sub-band energy matrix; The frequency subband energy matrix is ​​normalized, and the normalized frequency subband energy matrix is ​​interpolated and expanded to generate a frequency-time energy matrix.

3. The method for identifying lung ultrasound signal features based on RF signals according to claim 1, wherein: The steps of calculating the resonance period vector of the local energy peak according to the frequency-time energy matrix and constructing the dynamic period perception map of lung texture through multi-scale fusion mapping are as follows: Extracting the time position of the local peak of each frequency subband energy sequence in the frequency-time energy matrix, and then combining and generating a set of energy peak time interval sequences based on the time intervals between adjacent local peaks; The time intervals between adjacent peaks in the energy peak time interval sequence set are counted to form a local resonance period distribution, and the local resonance period distribution is normalized to generate a resonance period vector set; A frequency subband weight model is constructed according to the relative positions of the frequency subbands, and the resonance period vector set is weightedly fused with the weight model according to preset weights to generate a fused overall resonance period vector; The response intensity distribution of each frequency subband energy sequence in the frequency-time energy matrix is ​​calculated, and the response intensity distribution is resampled and spatially aligned to a uniform scale. Then, the fused overall resonance period vector and the response intensity distribution are combined to construct a dynamic periodic perception map of lung texture.

4. The method for identifying lung ultrasound signal features based on RF signals according to claim 3, wherein: The dynamic periodic perception map of lung texture uses the frequency subband index as the vertical axis and the standard time sampling point sequence after uniform scale resampling as the horizontal axis. The numerical value of each position in the dynamic periodic perception map of lung texture represents the normalized periodic response intensity value of the corresponding frequency subband at the time position.

5. The method for identifying lung ultrasound signal features based on RF signals according to claim 1, wherein: The original lung ultrasound radio frequency signal is subjected to cepstrum transformation to generate an original radio frequency cepstrum spectrum, and a multipath reflection structure model is constructed based on the original radio frequency cepstrum spectrum. The periodic structure candidate features in the dynamic periodic perception map of the lung texture are aligned with the original radio frequency cepstrum spectrum to obtain the time domain alignment error. The steps are as follows: The original lung ultrasound radio frequency signal is converted into the original lung ultrasound radio frequency signal frequency domain amplitude spectrum, and then the inverse Fourier transform is performed to generate a cepstrum sequence, which is then recombined to obtain the original radio frequency cepstrum spectrum; Perform delay structure recognition on the original RF cepstrum, obtain the local response position of the original RF cepstrum, perform time clustering and energy aggregation, and form a multi-path reflection structure model; Record the periodic response feature positions and periodic values ​​of the frequency sub-bands in the dynamic periodic perception map of lung texture, and combine them to form a candidate feature set of periodic structure; Based on the set of candidate periodic structure features, the cepstral response position with the smallest periodic value event position deviation is located in the frequency sub-band corresponding to the original RF cepstral spectrum, and compared with the main path delay in the multipath reflection structure model to generate the time domain alignment error.

6. The method for identifying lung ultrasound signal features based on RF signals according to claim 1, wherein: The steps of constructing a structure-dynamic coupled lung texture feature vector based on the candidate periodic structure features in the dynamic periodic perception map of lung texture, the time domain alignment error and the reflection path parameters in the multipath reflection structure model are as follows: A matching mapping table is constructed based on the candidate features of periodic structures in the dynamic periodic perception map of lung texture and the reflection path parameters in the multi-path reflection structure model; The dominant period value is extracted based on the matching mapping table and compared with the corresponding frequency subband position in the original RF cepstrum spectrum, and the time error is calculated as the dynamic difference measurement factor; Based on the candidate features of periodic structure, reflection path parameters and dynamic difference measurement factors, a three-dimensional joint vector is constructed and sequentially spliced ​​in frequency order to form a set of original lung texture structure-dynamic coupling vectors; The original lung texture structure-dynamic coupling vector set is normalized and a fixed embedding coding table is introduced for splicing to form the structure-dynamic coupling lung texture feature vector.

7. The method for identifying lung ultrasound signal features based on RF signals according to claim 1, wherein: The method is to construct a probability function reflecting the structural instability of lung lesions based on the structure-dynamic coupled lung texture feature vector and the reflection delay abnormality information in the multipath reflection structure model, and calculate the structural instability probability values ​​of different lung regions. The steps are as follows: According to the frequency sub-band index, the sub-band features in the structure-dynamic coupled lung texture feature vector are matched and filtered with the reflection delay abnormality information corresponding to the sub-band features to obtain the filtered fusion feature subset; Based on the filtered fusion feature subset, a structural instability probability mapping model is established; Utilize labeled training samples to perform parameter training and regularization optimization on the structural instability probability mapping model to generate a trained lung structural instability probability mapping model. The collected structure-dynamic coupled lung texture feature vectors were substituted into the trained lung structural instability probability mapping model to construct a probability function reflecting the structural instability of lung lesions, and the structural instability probability values ​​of different lung regions were calculated.

8. The method for identifying lung ultrasound signal features based on RF signals according to claim 7, wherein: The labeled training samples refer to a set of sample data used in the process of training the structural instability probability mapping model, wherein each sample data consists of a set of structure-dynamic coupled lung texture feature vectors and a corresponding structural instability state label.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the processor implements the steps of the method for identifying lung ultrasound signal features based on RF signals according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying lung ultrasound signal features based on RF signals according to any one of claims 1 to 8 are implemented.

Images