Digital stethoscope system based on piezoelectric film

Abstract

This invention discloses a digital stethoscope system based on piezoelectric thin film, belonging to the fields of medical electronics and biomedical engineering technology. Existing digital stethoscopes based on piezoelectric thin film face challenges such as signal noise and interference, limitations of signal processing algorithms, nonlinear response and time offset issues, and insufficient diagnostic accuracy and automation. This invention integrates dual audio acquisition methods of piezoelectric thin film and microphone to capture various physiological signals such as low-frequency vibrations and high-quality heart and lung sounds. Background noise is removed through adaptive filters and spectral subtraction techniques. By utilizing MFCC feature extraction and dynamic time warping algorithms, fine feature analysis and time axis alignment are performed on the preprocessed audio signal, achieving high-precision matching with the stored template. Intelligent feature processing accurately distinguishes different physiological states, and real-time, intuitive diagnostic feedback is achieved through wireless transmission and status indication, improving the reliability and efficiency of medical testing.

Description

Technical Field

[0001] This invention relates to the field of medical electronics and biomedical engineering technology, specifically to a digital stethoscope system based on piezoelectric thin films. Background Technology

[0002] Digital stethoscopes based on piezoelectric films utilize the piezoelectric effect to convert weak physiological vibration signals from the human body into electrical signals. Through high-precision signal amplification, filtering, and digital signal processing technologies, they can achieve high-quality acquisition and analysis of physiological sounds such as heart sounds and lung sounds. This technology has the advantages of being lightweight, non-invasive, and highly sensitive, and can be widely used in smart healthcare, remote monitoring, and artificial intelligence-assisted diagnosis.

[0003] However, it still faces challenges such as signal noise and interference, limitations of signal processing algorithms, nonlinear response and time shift issues, and insufficient diagnostic accuracy and automation, which affect its stability and accuracy in clinical applications. Summary of the Invention

[0004] The purpose of this invention is to solve the above-mentioned problems and to propose a digital stethoscope system based on piezoelectric thin films.

[0005] The objective of this invention can be achieved through the following technical solution: a digital stethoscope system based on piezoelectric thin film, comprising: a smart wearable stethoscope, a data acquisition and analysis module, and a feature processing module;

[0006] The smart wearable stethoscope consists of a stethoscope body (3), a chest strap (1), and a buckle (2); wherein the stethoscope body (3) includes an operation button (11), an operation button (12), a chest strap interface (13), an integration hole (14), a stethoscope battery compartment (15), and a piezoelectric film (16); wherein the chest strap (1) is connected to the chest strap interface (13), and the buckle (2) is installed on the chest strap (1) and is used to adjust the length of the chest strap; the chest strap interface (13), the integration hole (14), the operation button (11), the stethoscope battery compartment (15), and the piezoelectric film (16) are all installed on the stethoscope body (12);

[0007] The acquisition and analysis module acquires audio data from the smart wearable stethoscope via a wireless module and distinguishes it into microphone audio data and piezoelectric film audio data. The microphone audio data is used for heart and lung sound analysis to identify heart disease or breathing problems. The piezoelectric film audio data is used for low-frequency signal analysis to identify heart vibrations and pulse fluctuations.

[0008] Microphone audio data is denoised using an adaptive filter, which dynamically adjusts the filter weights using the LMS algorithm to optimize noise removal. Piezoelectric film audio data is denoised using spectral subtraction, which estimates the noise spectrum through short-time Fourier transform and subtracts it from the signal. The time-domain signal is recovered using inverse STFT, and smoothing and signal enhancement are performed to optimize the denoising effect.

[0009] The feature processing module performs a unified format conversion and feature analysis on the denoised microphone and piezoelectric film audio data to obtain the similarity XSD; if the XSD is lower than a preset threshold, it is determined to be different audio features, otherwise it is considered to be the same audio.

[0010] The feature processing module performs audio feature matching, compares the acquired data with the stored preset audio templates, and calculates the matching degree PPD. When the PPD reaches the threshold, it is judged as a successful match, and the corresponding physiological state or abnormal mode is output. The result is displayed through indicator lights.

[0011] In a preferred embodiment of the present invention, the specific process of distinguishing between microphone audio data and piezoelectric film audio data is as follows:

[0012] The audio data file format is distinguished and matched with the format corresponding to the microphone audio data. If the two match, the acquired audio data is microphone audio data. Similarly, the file format corresponding to the piezoelectric film audio data is matched with the data file format of the acquired audio data. If the two match, the acquired audio data is piezoelectric film audio data.

[0013] In a preferred embodiment of the present invention, the specific process of denoising the microphone audio data using an adaptive filter is as follows:

[0014] An adaptive filter is invoked for processing. This filter automatically adjusts its coefficients based on changes in the input signal. The adaptive filter is calculated using the LMS algorithm. The mixed audio signal x(n), containing the target signal and noise, along with the introduced noise reference signal d(n), is substituted into the adaptive filter to output the microphone filtered signal y(n). From the formula: Output microphone filter signal y(n), w k (n) represents the filter's weighting coefficients, M is the filter's order, and n is the current time. Next, the error between the filter output y(n) and the actual acquired signal x(n) is calculated using the formula: e(n) = x(n) - y(n). The error e(n) represents the noise that the filter failed to effectively remove. The filter's weighting coefficients w are then adjusted based on the error e(n). k (n); The LMS algorithm uses gradient descent to update the weights: w k (n+1)=w k(n)+μ·e(n)·d(nk), where μ is the step size factor that controls the magnitude of each update; (nk) is the delayed version of the reference signal.

[0015] In a preferred embodiment of the present invention, the specific process of denoising the piezoelectric thin film audio data using spectral subtraction is as follows:

[0016] Denoising is performed by subtracting the spectrum:

[0017] Step 1: Since there is no target information in the signal, these time periods are considered pure noise periods; during the time periods without signal, calculate the spectral characteristics of the noise; use the short-time Fourier transform to extract the spectral information of the signal: using the formula: t∈No signal segment, X(f,t) is the noise spectrum estimate at time t and frequency f; X(f,t) is the Fourier transform result of the signal; N is the number of signal-free segments used for noise estimation.

[0018] Step 2: After noise spectrum estimation, subtract the noise spectrum from the original signal spectrum; use short-time Fourier transform to decompose the noisy signal into a spectrum; the signal spectrum is represented as X(f,t), which includes the mixture of the target signal and noise; for each time frequency point, the estimated noise spectrum... Subtract from the signal's spectrum |X(f,t)|; introducing a threshold, we obtain the formula:

[0019] This is the spectrum of the denoised signal; the max function ensures that the spectrum value will not be negative.

[0020] Step 3: Convert the processed spectrum back to the time domain signal using the inverse short-time Fourier transform to obtain the denoised audio signal, as shown by the formula: x c (t) is the denoised time-domain signal, and ISTFT is the inverse short-time Fourier transform operation;

[0021] Step 4: After completing the spectral subtraction, reduce high-frequency residual noise through smoothing to maintain the smoothness of the signal; for target signals that are still weak after denoising, enhance the signal strength through enhancement algorithms.

[0022] In a preferred embodiment of the present invention, the specific process of obtaining the similarity XSD is as follows:

[0023] Audio data is framed according to a fixed duration and smoothed using a window function; MFCC features are extracted, including pre-emphasis, FFT transformation, power spectrum calculation, Mel filter bank processing, logarithmic transformation, and DCT transformation, to finally obtain MFCC feature vectors; to solve the nonlinear offset on the time axis, the DTW algorithm is used to perform optimal alignment of the MFCC sequence, calculate the cumulative matching error, and normalize it to obtain the similarity XSD.

[0024] In a preferred embodiment of the present invention, the audio data is divided into frames of fixed duration and smoothed using a window function as follows:

[0025] The synchronized audio data is divided into frames of fixed duration, with each frame containing N1 = sampling rate × 0.025 samples; frame overlap is set to smooth signal transition; then a window function is calculated for each frame's data using the formula: Output windowed frame data x i,ω (n1).

[0026] In a preferred embodiment of the present invention, the specific process of feature extraction by the MFCC is as follows:

[0027] For each frame of audio data, MFCC is used for feature extraction: the signal is pre-emphasized to enhance high-frequency components: x p (n1)=x i1,ω (n1)-α·x i1,ω (n1-1), α=0.95; Perform a Fast Fourier Transform on the pre-emphasized signal to convert the time-domain signal to the frequency domain, obtaining the spectrum X. i1 (k); Based on this spectrum, calculate the power spectrum of each frame using the formula: P(k)=|X i1 (k)| 2 Based on the nonlinear characteristics of human ear sensitivity to different frequencies, a Mel filter bank is constructed, and the power spectrum is decomposed across several Mel frequency bands. The energy output E of each filter is then calculated. m Specifically, for the m-th filter in the Mel filter bank, let its response function be H. m (k), usually, H m (k) is triangular in shape and defined within a certain frequency range, with values ​​of 0 outside this range; for the m-th filter, the energy is calculated by weighted summation of the power spectrum within the frequency range covered by the filter's response: k s and k e These represent the start and end bins of the frequency range corresponding to filter m in the FFT; H m(k) represents the filter weights at frequency bink. To prevent the energy of some filters from approaching or equaling zero, a very small constant ∈ is added to ensure that the value is always greater than zero, i.e., E'. m =E m +ε; The output energy E' of each filter is expressed using the natural logarithm or a commonly used logarithmic form. m Perform a logarithmic transformation using the natural logarithm: logE m =ln(E' m Finally, the logarithmic energy sequence is input into the discrete cosine transform to obtain a set of discrete coefficients. The first 12 to 13 coefficients are selected as the MFCC feature vector of the frame.

[0028] In a preferred embodiment of the present invention, the specific process of obtaining the similarity XSD is as follows:

[0029] The dynamic time warping algorithm is used to optimally align two sequences. The specific process is as follows:

[0030] Let the microphone MFCC feature vector of the i2th frame be... The piezoelectric film feature vector of frame j2 is The distance between two vectors is defined as the Euclidean distance between them: D1 is the dimension of the MFCC feature; let D(i2,j2) represent the minimum cumulative distance between the first i2 frames and the first j2 frames; use the recursive update formula: D(i2,j2)=d(i2,j2)+min{D(i2-1,j2),D(i2,j2-1),D(i2-1,j2-1)}, boundary conditions: starting point: D(1,1); the first row and the first column can be initialized by accumulation: D(i2,1)=d(i2,1)+min{D(i2-1,1),D(i2-1,1)}, D(1,j2)=d(1,j2)+min{D(1,j2),D(1,j2-1)}; construct the entire DTW matrix, let the length of the optimal alignment path be L, then the normalization error is: N m and N p These represent the number of frames for the microphone and piezoelectric film sequences, respectively; D n This represents the overall matching error between the two sequences; the normalized error is converted into a similarity index between 0 and 1: Output similarity XSD, where λ is an adjustment parameter that controls the rate at which the similarity decays.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. This invention employs an integrated design, combining a piezoelectric film with a high-quality microphone to achieve multimodal audio acquisition. Specifically, the piezoelectric film focuses on capturing low-frequency signals such as heart vibrations and pulse fluctuations, while the microphone is used to capture high-fidelity heart and lung sounds, breathing sounds, and environmental friction and noise. During the acquisition process, the system's built-in adaptive filter uses the LMS algorithm to adjust the filter weights according to the real-time changes in the input signal, effectively suppressing noise components in the mixed signal. Simultaneously, for the audio data from the piezoelectric film, a spectral subtraction technique based on Short-Time Fourier Transform (STFT) is used to estimate and subtract the noise spectrum by extracting the spectral characteristics of the "pure noise" segment, thereby significantly improving signal clarity and signal-to-noise ratio. This multi-level and meticulous noise suppression strategy provides a reliable data foundation for subsequent feature extraction and diagnostic analysis, ensuring that the device can still obtain accurate and stable physiological signal data in complex environments.

[0033] 2. After signal preprocessing, this invention converts the signal into a unified file format, frames it with a fixed duration, and applies window function weighting to each frame to ensure temporal continuity and smooth transition. Subsequently, the system pre-emphasizes each frame, converts the time-domain signal to the frequency domain using Fast Fourier Transform (FFT), and calculates the power spectrum of each frequency component. Considering the human ear's sensitivity to different frequencies, the system uses a weighted summation of Mel filter banks to obtain the energy output of each filter. After logarithmic transformation and discrete cosine transform, representative MFCC feature vectors are extracted. Since the audio data acquired by the microphone and piezoelectric film may have non-linear temporal offsets, the system further employs a dynamic time warping algorithm for optimal sequence alignment. By calculating the normalized matching error and converting it into a similarity index (XSD), accurate comparison of the two audio features is achieved. Finally, the obtained matching degree is compared with preset template features. When the matching degree reaches a threshold, the system can display the status in real time via indicator lights and transmit the diagnostic results to a smart device using a wireless module, providing accurate and intelligent support for the rapid identification of abnormal conditions such as heart disease and respiratory problems. Attached Figure Description

[0034] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0035] Figure 1 This is a schematic diagram of the principle of the present invention;

[0036] Figure 2 This is a schematic diagram of the smart wearable stethoscope of the present invention. Figure 1 ;

[0037] Figure 3 This is a schematic diagram of the smart wearable stethoscope of the present invention. Figure 2 ;

[0038] Figure 4 This is an illustration of the stethoscope of the present invention. Figure 1 ;

[0039] Figure 5 This is an illustration of the stethoscope of the present invention. Figure 2 .

[0040] Figure descriptions: 1. Chest strap, 2. Buckle, 3. Stethoscope surface, A. Stethoscope surface one, B. Stethoscope surface two, 11. Operation button, 12. Operation button, 13. Chest strap interface, 14. Integrated hole, 15. Stethoscope battery compartment, 16. Piezoelectric film. Detailed Implementation

[0041] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0042] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0043] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0044] Please see Figure 1 As shown, the digital stethoscope system based on piezoelectric thin film includes: a smart wearable stethoscope, a data acquisition and analysis module, and a feature processing module;

[0045] The smart wearable stethoscope consists of a stethoscope body 3, a chest strap 1, and a buckle 2. The stethoscope body 3 includes operation buttons 11 and 12, a chest strap interface 13, an integrated hole 14, a stethoscope battery compartment 15, and a piezoelectric diaphragm 16. The chest strap 1 connects to the chest strap interface 13, and the buckle 3 is mounted on the chest strap 1 and used to adjust the length of the chest strap. The chest strap interface 13, integrated hole 14, operation buttons 11, stethoscope battery compartment 15, and piezoelectric diaphragm 16 are all mounted on the stethoscope body 12. It should be noted that the smart wearable stethoscope integrates a wireless module (such as Bluetooth, Wi-Fi, or other low-power wireless technologies), a microphone, indicator lights, and a storage device. The wireless module transmits the stethoscope's data to other devices. The microphone is responsible for high-quality capture of internal and external sounds, breathing sounds, friction sounds, and background noise. The indicator lights provide feedback on the device's status, typically displaying battery level, device connection status, and whether it is working properly. The storage device stores the audio data recorded by the stethoscope and other diagnostic information.

[0046] The acquisition and analysis module obtains and analyzes the audio data recorded by the stethoscope in the storage device inside the smart wearable stethoscope through the wireless module;

[0047] The audio data is differentiated into microphone audio data and piezoelectric film audio data. Specifically, the file format of the acquired audio data is distinguished and matched against the format corresponding to the microphone audio data. If they match, the acquired audio data is microphone audio data. Similarly, the file format corresponding to the piezoelectric film audio data is matched against the file format of the acquired audio data. If they match, the acquired audio data is piezoelectric film audio data. It should be noted that microphone audio data is used for further analysis of heart and lung sounds, and feature extraction is used to identify heart disease or respiratory problems; piezoelectric film audio data focuses on the analysis of low-frequency signals, such as heart vibrations and pulse fluctuations.

[0048] The microphone audio data is denoised using an adaptive filter. The key to an adaptive filter is its ability to automatically adjust its coefficients based on changes in the input signal. The adaptive filter is then calculated using the LMS algorithm, specifically:

[0049] Substituting the mixed audio signal x(n) containing the target signal and noise, along with the noise-introduced reference signal d(n) (a signal obtained from the ambient noise source or other sensors on the microphone receiver; the reference signal is used to help the filter distinguish between noise and the target signal), into the adaptive filter outputs the microphone-filtered signal y(n); From the formula: Output the microphone filter signal y(n), where w k(n) represents the filter's weighting coefficients, M is the filter's order, and n is the current time. The error (i.e., the signal residual) between the filter output y(n) and the actual acquired signal x(n) is calculated, reflecting the difference between the filter output and the actual signal. This error is expressed by the formula: e(n) = x(n) - y(n), where e(n) represents the noise portion that the filter failed to effectively remove. The filter's weighting coefficients w are then adjusted based on this error e(n). k (n), enabling the filter to remove noise more accurately in the next step; the LMS algorithm uses gradient descent to update the weights: w k (n+1)=w k (n)+μ·e(n)·d(nk), where μ is the step size factor, which controls the magnitude of each update; (nk) is the delayed version of the reference signal; by continuously updating the filter weights, the LMS algorithm can adapt to signal changes in real time, thereby effectively removing noise;

[0050] The audio data of the piezoelectric film is denoised by using spectral subtraction (by estimating the noise spectrum and subtracting it from the spectrum of the noisy signal, the influence of noise is reduced, resulting in a clearer signal).

[0051] Step 1: Treat the signal as having no target information (e.g., when the patient is not making any sound); consider these time periods as "pure noise" segments (the spectrum of these time periods mainly consists of environmental noise and does not contain any useful physiological signals); calculate the spectral characteristics of the noise during the signal-free time periods; use the Short Time Fourier Transform (STFT) to extract the spectral information of the signal: using the formula: t∈no signal segment, where, X(f,t) is the noise spectrum estimate at time t and frequency f; X(f,t) is the Fourier transform result of the signal; N is the number of signal-free segments used for noise estimation.

[0052] Step 2: After noise spectrum estimation, subtract the noise spectrum from the original signal spectrum; use Short Time Fourier Transform (STFT) to decompose the noisy signal into a spectrum; the signal spectrum is represented as X(f,t), which includes the mixture of the target signal and noise; for each time frequency point, the estimated noise spectrum... Subtract from the signal's spectrum |X(f,t)|; to avoid a negative spectral value, a threshold needs to be introduced to ensure that the subtracted noise does not cause the spectral value to become unrealistic, resulting in the formula: in, This is the spectrum of the denoised signal; the max function ensures that the spectrum value is not negative, avoiding spectrum errors caused by subtracting noise; Explanation: Through this process, the noise spectrum is removed, and only the spectrum of the target signal is retained;

[0053] Step 3: Convert the processed spectrum back to the time domain signal using the Inverse Short Time Fourier Transform (ISTFT) to obtain the denoised audio signal, as shown by the formula: Where, x c (t) is the denoised time-domain signal, and ISTFT is the inverse short-time Fourier transform operation; Explanation: Through this step, an audio signal with background noise removed is obtained, and the main components of the target signal are recovered;

[0054] Step 4: After completing the spectral subtraction, reduce high-frequency residual noise through smoothing (such as moving average or low-pass filtering) to maintain the smoothness of the signal; for target signals that are still weak after denoising, enhance the signal strength through enhancement algorithms (such as gain control).

[0055] The feature processing module is used to perform feature processing on the denoised data, specifically:

[0056] The processed microphone audio data and piezoelectric film audio data are converted into the same file format and subjected to the same feature analysis.

[0057] The synchronized audio data is divided into frames of fixed duration (e.g., 25 milliseconds), with each frame containing N1 = sampling rate × 0.025 samples; frame overlap is set (e.g., 50% overlap) to smooth signal transition; then a window function is calculated for each frame of data using the formula: Output windowed frame data x i,ω (n1), and n1 = 1, 2, 3, ..., N1-1; For each frame of audio data, MFCC is used for feature extraction: the signal is pre-emphasized to enhance high-frequency components: x p (n1)=x i1,ω (n1)-α·x i1,ω (n1-1), where α≈0.95 is usually chosen to enhance high-frequency components and compensate for the attenuation of high-frequency signals during the acquisition process; next, a Fast Fourier Transform (FFT) is performed on the pre-emphasized signal to transform the time-domain signal into the frequency domain, obtaining the spectrum X. i1 (k); Based on this spectrum, calculate the power spectrum of each frame using the formula: P(k)=|X i1 (k)| 2 This reflects the energy distribution of each frequency component; then, based on the nonlinear characteristics of human ear sensitivity to different frequencies, a Mel filter bank is constructed to decompose the power spectrum across several Mel frequency bands, and the energy output E of each filter is calculated. m Specifically, for the m-th filter in the Mel filter bank, let its response function be H. m (k), usually, H m(k) is triangular in shape and defined within a certain frequency range (e.g., the filter starts at f). m-1 , in f m Reached its peak, ended at f m+1 Values ​​outside this range are 0; for the m-th filter, its energy is calculated by weighted summation of the power spectrum within the frequency range covered by the filter's response: Where, k s and k e These represent the start and end bins of the frequency range corresponding to filter m in the FFT; H m (k) represents the filter weight at frequency bink, reflecting the contribution of that frequency component to the filter energy; to prevent the energy of some filters from approaching or equaling zero, a very small constant ∈ (e.g., 10) is usually added. -10 This ensures that the value is always greater than zero, i.e., E' m =E m +ε; Furthermore, using the natural logarithm or the commonly used logarithmic form, the output energy E' of each filter is... m Perform a logarithmic transformation using the natural logarithm (ln): logE m =ln(E' m Finally, the logarithmic energy sequence is input into the discrete cosine transform (DCT) to obtain a set of discrete coefficients, of which the first 12 to 13 coefficients are selected as the MFCC feature vector of the frame.

[0058] Since the two MFCC feature sequences may have non-linear offsets or length differences on the time axis, direct frame-by-frame comparison is difficult to reflect their true similarity; therefore, the Dynamic Time Warping (DTW) algorithm is used to perform optimal alignment of the two sequences. The specific process is as follows:

[0059] Let the microphone MFCC feature vector of the i2th frame be... The piezoelectric film feature vector of frame j2 is The distance between two vectors is defined as the Euclidean distance between them: Where D1 is the dimension of the MFCC feature (e.g., 12 or 13, possibly higher if differential features are added); let D(i2,j2) represent the minimum cumulative distance between the previous i2 frames (microphone) and the previous j2 frames (piezoelectric); a recursive update formula is used: D(i2,j2)=d(i2,j2)+min{D(i2-1,j2),D(i2,j2-1),D(i2-1,j2-1)}, boundary conditions: starting point: D(1,1); the first row and first column can be... Initialize using a cumulative method: D(i2,1)=d(i2,1)+min{D(i2-1,1),D(i2-1,1)}, D(1,j2)=d(1,j2)+min{D(1,j2),D(1,j2-1)}; through this process, construct the entire DTW matrix, where each element reflects the cumulative matching error of the two sequences under different frame alignments; for ease of comparison, normalize the cumulative distance; assuming the length of the optimal alignment path is L, the normalized error is: Where N m and N p These are the frame numbers for the microphone and piezoelectric film sequences, respectively; this value represents the overall matching error between the two sequences; to make the results more intuitive, the normalized error is converted into a similarity index between 0 and 1: Output similarity XSD, where λ is an adjustment parameter that controls the rate at which the similarity decays; the closer the value is to 1, the more similar the two sequences are, and vice versa.

[0060] The similarity XSD is compared with a preset similarity threshold. If the similarity XSD is less than the preset similarity threshold, they are determined to be two audio features; otherwise, if the similarity XSD is greater than or equal to the preset similarity threshold, they are determined to be the same audio feature.

[0061] The feature processing module is also used for audio data feature matching, including microphone audio data and piezoelectric film audio data; it matches the acquired data with preset audio data features in the data storage unit (the data storage unit pre-stores a large number of preset audio data feature templates, which correspond to different physiological states or diagnostic criteria, such as standard heart sounds, normal breath sounds, and possible abnormal states); similarly, it calculates the similarity XSD, which is recorded as the matching degree PPD. When the matching degree PPD reaches a preset threshold, it can be judged as "matching successfully", and the corresponding physiological state or abnormal pattern is confirmed; the device status (such as normal, abnormal, connection status, etc.) is displayed through indicator lights, and the matching results are transmitted to other devices (such as smartphones, monitoring platforms) through the wireless module for further diagnosis.

[0062] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A digital stethoscope system based on piezoelectric thin film, comprising: The intelligent wearable stethoscope comprises a data acquisition and analysis module and a feature processing module; its features are: The smart wearable stethoscope consists of a stethoscope body (3), a chest strap (1), and a buckle (2); wherein, the stethoscope body (3) includes an operation button (11), an operation button (12), a chest strap interface (13), an integration hole (14), a stethoscope battery compartment (15), and a piezoelectric film (16); wherein the chest strap (1) is connected to the chest strap interface (13), and the buckle (2) is installed on the chest strap (1) and is used to adjust the length of the chest strap; the chest strap interface (13), the integration hole (14), the operation button (11), the stethoscope battery compartment (15), and the piezoelectric film (16) are all installed on the stethoscope body (12); The acquisition and analysis module acquires audio data from the smart wearable stethoscope via a wireless module and distinguishes it into microphone audio data and piezoelectric film audio data. The acquired audio data file format is differentiated and matched with the format corresponding to the microphone audio data; if they match, the acquired audio data is microphone audio data. Similarly, the file format corresponding to the piezoelectric film audio data is matched with the data file format of the acquired audio data; if they match, the acquired audio data is piezoelectric film audio data. Microphone audio data is used for heart and lung sound analysis to identify heart disease or respiratory problems; piezoelectric film audio data is used for low-frequency signal analysis to identify heart vibrations and pulse fluctuations. The microphone audio data is denoised using an adaptive filter: the adaptive filter is invoked for processing, and its coefficients are automatically adjusted according to changes in the input signal; the adaptive filter is calculated using the LMS algorithm, substituting the mixed audio signal x(n) containing the target signal and noise, along with the noise reference signal d(n), into the adaptive filter to output the microphone filtered signal y(n); from the formula: Output microphone filter signal y(n), w k (n) represents the weighting coefficients of the filter, M is the order of the filter, and n is the current time; then, the error between the filter output y(n) and the actual acquired signal x(n) is calculated using the formula: The error e(n) represents the noise that the filter failed to remove effectively; the weighting coefficients w of the filter are adjusted based on the error e(n). k (n); The LMS algorithm uses gradient descent to update the weights: μ is the step size factor, which controls the amplitude of each update; (n−k) is the delayed version of the reference signal; the piezoelectric film audio data is denoised using spectral subtraction, which estimates the noise spectrum through short-time Fourier transform and subtracts it from the signal; the time-domain signal is recovered using inverse STFT, and smoothing and signal enhancement are performed to optimize the denoising effect; The feature processing module performs a unified format conversion and feature analysis on the denoised microphone and piezoelectric film audio data to obtain the similarity XSD; if the XSD is lower than a preset threshold, it is determined to be different audio features, otherwise it is considered to be the same audio. The feature processing module performs audio feature matching, compares the acquired data with the stored preset audio templates, and calculates the matching degree PPD. When the PPD reaches the threshold, it is judged as a successful match, and the corresponding physiological state or abnormal mode is output. The result is displayed through indicator lights.

2. The digital stethoscope system based on piezoelectric thin film according to claim 1, characterized in that, The specific process of denoising the piezoelectric thin film audio data using spectral subtraction is as follows: Denoising is performed by subtracting the spectrum: Step 1: Since there is no target information in the signal, these time periods are considered pure noise periods; during the time periods without signal, calculate the spectral characteristics of the noise; use the short-time Fourier transform to extract the spectral information of the signal: using the formula: , It is a noise spectrum estimate at time t and frequency f; It is the Fourier transform result of the signal; N is the number of signal-free segments used for noise estimation; Step 2: After noise spectrum estimation, subtract the noise spectrum from the original signal spectrum; use short-time Fourier transform to decompose the noisy signal into its spectrum; the signal spectrum is represented as follows. It contains a mixture of target signal and noise; For each time-frequency point, the estimated noise spectrum will be... From the signal spectrum Subtract from the middle; introduce a threshold, and we get the formula: , This is the spectrum of the denoised signal; the max function ensures that the spectrum value will not be negative. Step 3: Convert the processed spectrum back to the time domain signal using the inverse short-time Fourier transform to obtain the denoised audio signal, as shown by the formula: , It is the denoised time-domain signal; ISTFT is the inverse short-time Fourier transform operation. Step 4: After completing the spectral subtraction, reduce high-frequency residual noise through smoothing to maintain the smoothness of the signal; for target signals that are still weak after denoising, enhance the signal strength through enhancement algorithms.

3. The digital stethoscope system based on piezoelectric thin film according to claim 1, characterized in that, The specific process for obtaining the similarity XSD is as follows: Audio data is framed according to a fixed duration and smoothed using a window function; MFCC features are extracted, including pre-emphasis, FFT transformation, power spectrum calculation, Mel filter bank processing, logarithmic transformation, and DCT transformation, to finally obtain MFCC feature vectors; to solve the nonlinear offset on the time axis, the DTW algorithm is used to perform optimal alignment of the MFCC sequence, calculate the cumulative matching error, and normalize it to obtain the similarity XSD.

4. The digital stethoscope system based on piezoelectric thin film according to claim 3, characterized in that, The specific process of dividing the audio data into frames of fixed duration and smoothing it using a window function is as follows: The synchronized audio data is divided into frames of fixed duration, with each frame containing N1 = sampling rate × 0.025 samples; frame overlap is set to smooth signal transition; then a window function is calculated for each frame's data using the formula: Output windowed frame data .

5. The digital stethoscope system based on piezoelectric thin film according to claim 4, characterized in that, The specific process of feature extraction by the MFCC is as follows: For each frame of audio data, MFCC is used for feature extraction: the signal is pre-emphasized to enhance high-frequency components. α=0.95; Perform a Fast Fourier Transform on the pre-emphasized signal to convert the time-domain signal to the frequency domain, obtaining the spectrum. Based on this spectrum, the power spectrum of each frame is calculated using the formula: Based on the nonlinear characteristics of human ear sensitivity to different frequencies, a Mel filter bank is constructed, and the power spectrum is decomposed across several Mel frequency bands. The energy output E of each filter is then calculated. m Specifically, for the m-th filter in the Mel filter bank, let its response function be H. m (k), usually, H m (k) is triangular in shape and defined within a certain frequency range, with values ​​of 0 outside this range; for the m-th filter, the energy is calculated by weighted summation of the power spectrum within the frequency range covered by the filter's response: k s and k e These represent the start and end bins of the frequency range corresponding to filter m in the FFT; H m (k) represents the filter weights at frequency bink. To prevent the energy of some filters from approaching or equaling zero, a very small constant ϵ is added to ensure that the value is always greater than zero. ; The output energy of each filter is calculated using the natural logarithm or a commonly used logarithmic form. Perform a logarithmic transformation using the natural logarithm: Finally, the logarithmic energy sequence is input into the discrete cosine transform to obtain a set of discrete coefficients. The first 12 to 13 coefficients are selected as the MFCC feature vector of the frame.

6. The digital stethoscope system based on piezoelectric thin film according to claim 5, characterized in that, The specific process for obtaining the similarity XSD is as follows: The dynamic time warping algorithm is used to optimally align two sequences. The specific process is as follows: Let the microphone MFCC feature vector of the i2th frame be... The piezoelectric thin film feature vector of frame j2 is The distance between two vectors is defined as the Euclidean distance between them. D1 is the dimension of the MFCC feature; Let D(i2,j2) represent the minimum cumulative distance between the first i2 frames and the first j2 frames; a recursive update formula is used: Boundary conditions: Starting point: D(1,1); The first row and first column can be initialized using an accumulation method: , ; Construct the entire DTW matrix, and let the length of the optimal alignment path be L. Then the normalization error is: N m and N p These represent the number of frames for the microphone and piezoelectric film sequences, respectively; D n This represents the overall matching error between the two sequences; the normalized error is converted into a similarity index between 0 and 1: Output similarity XSD, where λ is an adjustment parameter that controls the rate at which the similarity decays.

Images