Dual channel adaptive interference rejection pet cardiopulmonary acoustic signal monitoring device

Abstract

The application discloses a pet heart-lung sound signal monitoring device with double-channel adaptive interference suppression, solves the problem that existing pet heart-lung sound monitoring is polluted by wearing interference, and belongs to the field of pet intelligent wearable devices.The device comprises a necklace body, a first sound collecting unit and a second sound collecting unit arranged on the inner side of the necklace body, and a main controller connected with the two sound collecting units.The main controller constructs wearing interference reference components according to a heart-lung sound main signal and a reference acoustic signal, estimates noise components in the heart-lung sound main signal through an adaptive filter, and obtains a denoising candidate signal by deduction, adjusts an effective update step of the adaptive filter according to the envelope cycle peak stability of the denoising candidate signal, iterates until convergence to obtain a denoising heart-lung sound signal, and extracts state characteristics according to the denoising heart-lung sound signal and outputs a pet state classification result.

Description

Technical Field

[0001] This application relates to a dual-channel adaptive interference suppression pet cardiopulmonary sound signal monitoring device, belonging to the field of pet smart wearable devices. Background Technology

[0002] With the development of smart wearable devices for pets and pet health monitoring and behavior analysis technologies, continuous monitoring of pets' conditions using wearable devices such as collars has become an important direction in the field of pet health management. Compared with monitoring methods based on cameras or environmental sound pickup, collar-based surface acoustic acquisition has advantages such as strong concealment, continuous acquisition, and low dependence on ambient light, and can obtain cardiopulmonary sound signals containing heart sounds, lung sounds, or respiratory sounds. Indicators such as heart rate, respiratory rate, rhythm stability, and changes in lung sound energy can reflect the differences in the physiological state of pets in states such as calm, after activity, stress, and panting.

[0003] However, the acquisition environment for heart and lung sounds differs significantly from that of traditional auscultation in pet collar scenarios. The collar's position is affected by fur, skin, body size, movement posture, and collar tightness, making it difficult to maintain stable contact between the sound acquisition unit and the pet's body. When the pet runs, shakes its head, lowers its head to eat, scratches, or experiences slight collar slippage, the acquired audio will be mixed with fur friction noise, collar structural vibration noise, impact noise from sudden changes in body surface contact, and external environmental noise. These interferences may overlap with the target heart and lung sounds in both the time and frequency domains, making it difficult to reliably isolate them using simple bandpass filtering, static noise reduction, or overall energy thresholding methods.

[0004] Existing methods for processing heart and lung sounds are mostly designed for relatively quiet scenarios such as manual auscultation or fixed sensor acquisition. They typically assume stable contact between the acquisition device and the body surface, and that noise components can be processed through fixed-band filtering or spectral suppression. However, interference from pet collar acquisition exhibits significant non-stationarity and wear-related characteristics: on the one hand, fur friction and collar movement usually appear as short-duration bursts of energy, broadband noise, or local spikes; on the other hand, ambient sound has low correlation between the main and reference paths, while collar structural conduction sound has high correlation between the two paths. If these are processed uniformly without differentiation, effective noise reduction is difficult to achieve. Without constructing a reference denoising mechanism for this type of wear-related interference, subsequent heart rate, respiratory rate, and rhythm fluctuations are easily contaminated by spurious peaks, leading to misjudgments of the pet's condition.

[0005] Furthermore, resting heart rate and resting respiratory rate vary among different pets. Even if heart rate and respiratory rate are calculated, they are difficult to use as reliable indicators of a pet's condition if the audio clips on which they are based are not adequately denoised or are heavily polluted with noise. For example, collar friction peaks may be mistaken for heart rate peaks, and local noise fluctuations in the respiratory tract may be mistaken for respiratory peaks, leading to abnormal estimations of heart rate and respiratory rate. Summary of the Invention

[0006] To address the problem of interference contamination of heart and lung sound signals during existing pet heart and lung sound monitoring, this application provides a dual-channel adaptive interference suppression pet heart and lung sound signal monitoring device. This dual-channel adaptive interference suppression pet heart and lung sound signal monitoring device includes:

[0007] The collar body has a first sound acquisition unit and a second sound acquisition unit disposed inside the collar body. The first sound acquisition unit is used to acquire the main signal of heart and lung sounds, and the second sound acquisition unit is used to acquire the reference acoustic signal.

[0008] The main controller is connected to both the first sound acquisition unit and the second sound acquisition unit.

[0009] The main controller is configured as follows:

[0010] Based on the main cardiopulmonary sound signal and the reference acoustic signal, a wear interference reference component is constructed;

[0011] Based on the wearing interference reference component, the noise component in the main cardiopulmonary sound signal is estimated by an adaptive filter, and the noise component is subtracted from the main cardiopulmonary sound signal to obtain a denoised candidate signal;

[0012] The effective update step size of the adaptive filter is adjusted according to the envelope periodic peak stability of the denoised candidate signal. The estimation and subtraction of the noise components are re-executed with the updated effective update step size until the envelope periodic peak stability of the denoised candidate signal meets the preset convergence condition, thereby obtaining the denoised cardiopulmonary sound signal.

[0013] Based on the noise-reduced cardiopulmonary sound signal output status characteristics and pet status classification results.

[0014] Preferably, the first sound acquisition unit is located on the inner side of the collar body at a position corresponding to the transmission area of ​​the pet's neck or chest cavity, and the second sound acquisition unit is located at another position on the inner side of the collar body, spaced apart from the first sound acquisition unit.

[0015] Preferably, the method for adjusting the effective update step size of the adaptive filter based on the envelope periodic peak stability of the denoised candidate signal includes:

[0016] Extract the heart sound envelope and lung sound envelope of the denoised candidate signal;

[0017] Calculate the periodic peak confidence of the heart sound envelope and the periodic peak confidence of the lung sound envelope respectively, and generate the periodicity preservation coefficient;

[0018] The effective update step size of the adaptive filter is adjusted according to the period preservation coefficient, wherein the period preservation coefficient is negatively correlated with the effective update step size.

[0019] As a preferred method, the effective update step size of the adaptive filter is adjusted according to the period preservation coefficient as follows:

[0020] ;

[0021] in, Let n be the effective update step size of the adaptive filter at the nth sampling time. Let be the initial step size of the adaptive filter. To maintain weights for the period, and , The periodicity retention coefficient is given, and .

[0022] The beneficial effects of this application are as follows: This application forms a dual-channel structure of main path and reference path through a first sound acquisition unit and a second sound acquisition unit. It utilizes the reference path signal to construct a reference component for wearing interference, and uses an adaptive filter to estimate and cancel related interference in the main path, thereby reducing the contamination of cardiopulmonary sound signals by non-target acoustic components. Specifically, by adjusting the adaptive update step size through a periodicity-maintaining coefficient, the time windows for clearer heartbeat and respiratory peaks are better protected, thus improving the stability of state feature extraction while reducing noise. This application not only extracts cardiopulmonary state features but also combines quality indicators such as reference path correlation, spectral flatness, silence ratio, and energy mutation to determine the validity of cardiopulmonary sound segments, thereby avoiding misjudging invalid audio as states of tension, excitement, or wheezing. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the pet heart and lung sound signal monitoring device of this application;

[0024] Figure 2 This is a flowchart illustrating the overall process of dual-channel adaptive interference suppression and state feature extraction for the cardiopulmonary sound signals of pet collars in this application.

[0025] Figure 3 The flowchart of adaptive denoising between the main path cardiopulmonary sound signal and the reference acoustic signal;

[0026] Figure 4 Flowchart for separating and extracting state features of the heart sound channel and lung sound channel in denoising heart and lung sound signals;

[0027] Figure 5 A schematic diagram showing the regions for determining the characteristics of heart and lung sounds corresponding to different pet states;

[0028] Figure 6The diagram shows a comparison of the original cardiopulmonary sound spectrum signal and the cardiopulmonary sound signal after denoising processing according to this application in the time and frequency domains. (a) is the time domain waveform of the original cardiopulmonary sound signal, (b) is the time domain waveform of the denoised cardiopulmonary sound signal, (c) is the power spectral density of the original cardiopulmonary sound signal, (d) is the power spectral density of the denoised cardiopulmonary sound signal, and (e) is the detection result of the cardiopulmonary sound envelope and periodic peak after denoising.

[0029] Figure 7 The diagram shows the comparison of cardiopulmonary sound signals and feature extraction effects before and after processing under the interference of hair friction. (a) is the original cardiopulmonary sound spectrogram, (b) is the cardiopulmonary sound spectrogram after denoising, and (c) is the envelope and periodic peak detection results after denoising.

[0030] Figure 8 The diagram shows a comparison of the cardiopulmonary sound signals and feature extraction effects before and after processing under the interference of vocalizations / harmonics. (a) is the original cardiopulmonary sound spectrogram, (b) is the cardiopulmonary sound spectrogram after denoising, and (c) is the envelope and periodic peak detection results after denoising.

[0031] Figure 9 The diagram shows a comparison of the cardiopulmonary sound signals and feature extraction effects before and after processing under mixed environmental / friction interference. (a) is the original cardiopulmonary sound spectrogram, (b) is the cardiopulmonary sound spectrogram after denoising, and (c) is the envelope and periodic peak detection results after denoising.

[0032] Figure 10 The diagram shows a comparison of the cardiopulmonary sound signals and feature extraction effects before and after processing by the present application under far-field / tissue filtering attenuation. (a) is the original cardiopulmonary sound spectrogram, (b) is the cardiopulmonary sound spectrogram after denoising, and (c) is the envelope and periodic peak detection results after denoising. Detailed Implementation

[0033] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0035] The present application will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the application.

[0036] This embodiment provides a dual-channel adaptive interference suppression pet cardiopulmonary sound signal monitoring device. The device uses two sound acquisition units on the pet's collar to acquire a main signal containing heart and lung sounds, and a reference signal containing wearing interference. The reference signal is used to adaptively estimate and subtract the wearing interference noise in the main signal. Simultaneously, during denoising, the update step size of the adaptive filter is dynamically adjusted based on the stability of the envelope periodic peak of the denoised signal to prevent weakening of the periodic structure of heart and lung sounds. After iterative convergence, the denoised cardiopulmonary sound signal is obtained, and state features such as heart rate, respiratory rate, rhythm fluctuations, and energy percentage are extracted from it. Finally, the pet's state classification result is output.

[0037] In this embodiment, the main controller receives the main signals of heart and lung sounds collected by the first sound acquisition unit. Reference acoustic signal acquired by the second sound acquisition unit The main controller constructs a wear-related interference reference component based on the two signals. An adaptive filter is then used to estimate the noise component in the main path based on this reference component and subtract it from the main path to obtain a denoised candidate signal. The stability of the envelope periodic peak of this denoised candidate signal is detected, and the effective update step size of the adaptive filter is adjusted accordingly. Noise estimation and subtraction are then re-executed with the updated step size, repeating this process until the stability of the envelope periodic peak meets a preset convergence condition, resulting in the final denoised cardiopulmonary sound signal. The main controller extracts state features from the denoised cardiopulmonary sound signal and outputs the pet state classification result.

[0038] The dual-channel adaptive interference suppression pet cardiopulmonary sound signal monitoring device of this embodiment, such as Figure 1 As shown, it includes a collar body 1, a first sound acquisition unit 2, a second sound acquisition unit 3, and a main controller 4.

[0039] The collar body 1 has a ring-shaped or strap-like structure and is made of a flexible material, such as silicone, fabric, or elastic band, to accommodate different pet neck circumference sizes. A first sound acquisition unit 2 and a second sound acquisition unit 3 are located on the inner side of the collar body 1.

[0040] The first sound acquisition unit 2 is used to collect the main signals of heart and lung sounds from the pet's body surface. The first sound acquisition unit 2 uses a miniature electret microphone, a microelectromechanical system (MEMS) microphone, a contact sound sensor, or a piezoelectric acoustic sensor. To ensure effective acoustic coupling on the body surface, a flexible bonding pad or a perforated acoustic coupling layer is provided on the outer side of the first sound acquisition unit 2. The bonding pad is made of medical-grade silicone or polyurethane foam.

[0041] The second sound acquisition unit 3 is used to acquire reference acoustic signals. The second sound acquisition unit 3 uses the same type of sensor as the first sound acquisition unit 2 to ensure that the frequency response characteristics of the two signals are consistent. The second sound acquisition unit 3 is located on the inner side of the collar body 1 at a distance from the first sound acquisition unit 2, and is used to pick up acoustic signals related to wearing interference. These acoustic signals include hair friction sound, collar structure vibration sound, impact sound caused by changes in body surface contact, and ambient sound.

[0042] A main controller 4 is located on the outer side or top of the collar body 1. The main controller 4 is connected to the first sound acquisition unit 2 and the second sound acquisition unit 3 respectively. The main controller 4 adopts an embedded microcontroller or digital signal processor, and integrates an analog-to-digital conversion interface, a data buffer, and embedded program logic.

[0043] In a preferred embodiment, the first sound acquisition unit 2 is positioned inside the collar body 1 at a location corresponding to the pet's neck or chest conduction area. When the pet wears the collar, the inner surface of the collar body 1 contacts the pet's neck skin and chest area. The first sound acquisition unit 2 is located in the area on the inner side of the collar body 1, in front of the pet's neck or above the sternum. This area is close to the surface projection of the pet's trachea and heart, enabling better conduction of heart and respiratory sound signals. A flexible coupling pad is provided between the outer surface of the first sound acquisition unit 2 and the pet's body surface to ensure the stability of acoustic coupling. The second sound acquisition unit 3 is positioned at a different location inside the collar body 1, spaced apart from the first sound acquisition unit 2. The second sound acquisition unit 3 is positioned on the inner side of the collar body 1 opposite to the first sound acquisition unit 2, or on the inner side of the collar body 1 near the back of the neck. The distance between the second sound acquisition unit 3 and the first sound acquisition unit 2 is determined according to the pet's neck circumference. By placing the second sound acquisition unit 3 at a position far from the chest cavity conduction area, the second sound acquisition unit 3 can more easily pick up the acoustic signals of wearing interference caused by collar movement, hair friction and changes in body surface contact, while picking up relatively fewer effective components of heart sounds and lung sounds.

[0044] Reference Figure 2 The main controller 4 is configured to perform the following signal processing steps:

[0045] Step 1, refer to Figure 3 Construct reference components for wearing interference;

[0046] The main controller 4 acquires the main signal of heart and lung sounds collected by the first sound acquisition unit 2. Reference acoustic signal acquired by the second sound acquisition unit 3 The main controller 4 operates based on the main signal of heart and lung sounds. and reference acoustic signal Construct a reference component for wearing interference.

[0047] In one implementation, the main controller 4 first performs synchronous preprocessing on the two signals. This synchronous preprocessing includes at least one or more of the following: mono conversion, sampling rate unification, amplitude normalization, DC offset removal, and short-time framing, to obtain the main preprocessed signal. and reference path preprocessed signal This ensures that the two signals are comparable in both time and amplitude scales.

[0048] Subsequently, the main controller 4 calculates the main path preprocessing signal. Preprocessed signal with reference path The short-time cross-correlation coefficient between them. In a preferred embodiment, the short-time cross-correlation coefficient is:

[0049]

[0050] in, This represents the short-time cross-correlation coefficient between the main path preprocessed signal and the reference path preprocessed signal within the current short-time analysis window. This indicates the sampling time corresponding to the starting sampling point of the current short-time analysis window. This indicates the sampling point number within the current short-term analysis window. Indicates the number of times the short-term analysis window is currently open. The sampling time corresponding to each sampling point; To prevent extremely small constants with a denominator of zero; This indicates the main path preprocessed signal at the sampling time. The signal value at that location; This indicates the reference path preprocessed signal at the sampling time. The signal value at that location; This indicates the number of sampling points contained in the short-time analysis window.

[0051] Simultaneously, the main controller 4 calculates the short-time energy change of the reference path preprocessed signal within the current short-time analysis window. In a preferred embodiment, the calculation method for the short-time energy change is as follows:

[0052]

[0053] in, This represents the short-time energy of the reference path preprocessed signal within the current short-time analysis window, starting from sampling time t. , This represents the short-time energy of the reference path preprocessed signal within the previous short-time analysis window. To prevent extremely small constants with a denominator of zero.

[0054] The main controller 4 calculates the short-time energy value of the reference path preprocessed signal within the current window. In a preferred embodiment, the short-time energy value is:

[0055]

[0056] in, To perform a short-time Fourier transform on the preprocessed signal of the reference path, the current short-time analysis window is used to obtain the first... The power spectrum corresponding to each frequency point The number of frequency points involved in the calculation.

[0057] The main controller 4 determines whether the short-time cross-correlation coefficient, short-time energy change, and spectral flatness all meet the preset interference judgment conditions. The preset interference judgment conditions include the following:

[0058] A short-time cross-correlation coefficient below the first threshold indicates low correlation between the main preprocessed signal and the reference preprocessed signal, suggesting that the reference preprocessed signal primarily contains independent noise unrelated to the effective components of the main cardiopulmonary sound signal. A short-time energy change above the second threshold indicates sudden energy changes in the reference preprocessed signal, consistent with the time-domain characteristics of wear-related interference. A spectral flatness above the third threshold indicates a flat spectral distribution in the reference preprocessed signal, exhibiting broadband noise characteristics, consistent with the frequency-domain characteristics of wear-related interference.

[0059] When all three conditions are met, the main controller 4 determines that there is a valid wearing interference component in the current window and outputs the preprocessed reference signal or its weighted signal as the wearing interference reference component. When any one of the conditions is not met, the main controller 4 determines that there is no significant wearing interference component in the current window and does not output the wearing interference reference component.

[0060] Step Two, Refer to Figure 3 The adaptive filter estimates and subtracts noise;

[0061] Based on the wearing interference reference component constructed in step one, the main controller 4 estimates the noise component in the main cardiopulmonary sound signal using an adaptive filter. Specifically, the main controller 4 uses the wearing interference reference component as the reference input signal of the adaptive filter and the main cardiopulmonary sound signal as the expected signal of the adaptive filter. The adaptive filter employs a normalized least mean square algorithm, iteratively adjusting the adaptive filter weight coefficients to gradually approximate the true value of the noise component in the main cardiopulmonary sound signal, thereby achieving the estimation of the noise component.

[0062]

[0063] in, , This is the effective step size.

[0064] noise components It is estimated that the following relationship is satisfied:

[0065]

[0066] in, For the order of the adaptive filter, For the first The first-order adaptive filter coefficients are used to characterize the propagation relationship of the reference interference in the main signal.

[0067] Order of adaptive filter The choice of order should be based on the actual application scenario. For scenarios where wear-related interference is mainly low-frequency, a smaller order should be chosen; for scenarios where wear-related interference is complex and includes a wide frequency band, a larger order should be chosen. The larger the order of the adaptive filter, the stronger its ability to model complex interference, but the computational load also increases accordingly. In actual deployment, a trade-off should be made based on the computing power and power consumption requirements of the main controller 4.

[0068] Main controller 4 preprocesses signals from the main path The noise component obtained by subtracting the above estimation Thus, the candidate signals for denoising are obtained.

[0069] .

[0070] Step 3: Adjust and iterate the cycle step size;

[0071] To avoid excessive attenuation of periodic heart and lung sounds during adaptive denoising, a period preservation coefficient can be introduced into the adaptive filtering process. The periodicity retention coefficient is obtained based on the periodic peak stability of the center sound envelope and lung sound envelope of the denoised candidate signal.

[0072] Unlike conventional adaptive filtering algorithms where the effective update step size is adjusted solely based on the input signal power or error signal magnitude, this implementation adjusts the effective update step size based on the stability of the envelope periodic peaks of the denoised candidate signal; that is, the feedback quantity is the physiological periodic structure itself. In conventional normalized least mean square algorithms, the step size normalization coefficient only reflects the amplitude of the reference input signal. Its optimization objective is to ensure the algorithm's convergence stability, completely disregarding whether the processed signal contains physiologically significant cardiopulmonary periodic structures. When the reference signal amplitude is large, the effective update step size automatically decreases to maintain convergence. However, if the main cardiopulmonary sound signal contains clear heartbeat and respiratory peaks, decreasing the effective update step size actually helps protect these periodic structures from being weakened. Conversely, when the reference signal amplitude is small, the effective update step size automatically increases. However, if the central lung periodic structure of the main cardiopulmonary sound signal is incomplete or contaminated by noise, increasing the effective update step size may further damage the remaining periodic structure. Therefore, there is a mismatch between the traditional effective update step size adjustment method based on signal power and the objective of protecting physiological periodic structures.

[0073] In this embodiment, the main controller 4 extracts the heart sound envelope and lung sound envelope from the denoising candidate signal. By analyzing the regularity of the periodic peaks in the envelope, it directly assesses the integrity of the physiological periodic structure in the current signal and adjusts the effective update step size of the adaptive filter accordingly. The higher the stability of the envelope periodic peaks, the smaller the effective update step size. The lower the stability of the envelope periodic peaks, the larger the effective update step size. This mechanism ensures that the convergence behavior of the adaptive filter is no longer driven solely by mathematical indicators, i.e., signal power, but by biological indicators, i.e., the stability of the envelope periodic peaks. Thus, protecting the cardiopulmonary periodic structure is always the core constraint during the denoising process.

[0074] The method for extracting the heart sound envelope is as follows: The denoised candidate signal is subjected to cardiac frequency band filtering, with a passband range of 20Hz to 200Hz. Then, the filtered signal is subjected to Hilbert transform or rectification smoothing to obtain the heart sound envelope curve. The method for extracting the lung sound envelope is as follows: The denoised candidate signal is subjected to lung frequency band filtering, with a passband range of 100Hz to 2000Hz. Then, the filtered signal is subjected to Hilbert transform or rectification smoothing to obtain the lung sound envelope curve.

[0075] For the heart sound envelope, the main controller 4 performs peak detection on the envelope curve, marking the position of each periodic peak and its corresponding time index to obtain the heartbeat peak time series. The time interval between adjacent heartbeat peaks is calculated to form a heartbeat peak interval series. The mean and standard deviation of this heartbeat peak interval series are calculated, and the ratio of the standard deviation to the mean is used as the interval variation coefficient. The smaller the interval variation coefficient, the more uniform the interval between adjacent periodic peaks, and the higher the consistency of the periodic peak intervals. Simultaneously, the main controller 4 extracts the amplitude of each periodic peak in the heart sound envelope, and calculates the ratio of each periodic peak amplitude to the envelope mean as the peak relative amplitude. The larger the peak relative amplitude, the more prominent the periodic peak is from the background envelope, and the higher the peak significance. The main controller 4 performs weighted fusion based on the interval consistency and the peak significance to obtain the confidence score of the heart sound periodic peaks. Similarly, the same processing is performed on the lung sound envelope curve to obtain the lung sound periodic peak confidence score.

[0076] The main controller 4 generates a periodicity preservation coefficient by performing a weighted fusion or minimum value fusion based on the confidence levels of the heart sound periodicity peaks and lung sound periodicity peaks. In one embodiment, the periodicity preservation coefficient is the weighted average of the confidence levels of the heart sound envelope periodicity peaks and lung sound envelope periodicity peaks, with the heart sound confidence level having a weight of 0.6 and the lung sound confidence level having a weight of 0.4. In another embodiment, the periodicity preservation coefficient is the smaller value between the heart sound envelope periodicity peak confidence level and the lung sound envelope periodicity peak confidence level, to employ a conservative strategy to protect the periodic structure.

[0077] The period preservation coefficient is calculated based on the envelope of the denoised candidate signal, not the envelope of the reference acoustic signal. The denoised candidate signal is an intermediate result after preliminary noise subtraction, in which residual noise components have been partially suppressed, and the cardiopulmonary cycle structure is clearer than the original signal. However, the denoised candidate signal may also suffer from weakened heart or lung sound cycle peaks due to over-adjustment of the adaptive filter. Therefore, analyzing the stability of the envelope cycle peaks of the denoised candidate signal can directly reflect the degree of disruption to the physiological cycle structure caused by the denoising operation at the current effective update step size setting. This allows the effective update step size of the adaptive filter to be adjusted in real time according to the actual impact of the current denoising effect on the physiological structure, rather than relying on a fixed value set in the open loop or mathematical indicators unrelated to the physiological cycle.

[0078] The periodicity preservation coefficient is negatively correlated with the effective update step size; that is, the higher the periodicity preservation coefficient, the smaller the effective update step size, and vice versa. The periodicity preservation coefficient ranges from 0 to 1. When the periodicity preservation coefficient approaches 1, it indicates the presence of stable, equally spaced periodic peaks in the heart sound and lung sound envelopes, indicating a clear and complete cardiopulmonary cycle structure. In this case, the effective update step size is compressed to a minimum to protect this physiological structure from being weakened by subsequent iterations. When the periodicity preservation coefficient approaches 0, it indicates that the envelope is disordered or the periodic peak amplitude is extremely weak, indicating that the effective cardiopulmonary cycle structure is insignificant or severely contaminated. In this case, the effective update step size is amplified to a maximum to enhance the denoising and signal recovery quality. Through this negative correlation, the effective update step size adjustment of the adaptive filter achieves a shift from being driven by mathematical convergence to being driven by the integrity of the biological cycle.

[0079] The main controller 4 calculates the adaptive filter in the following formula at the 1st... Effective update step size for each sampling time:

[0080]

[0081] in, For the first The effective update step size of the adaptive filter at each sampling time. This is the initial step size of the adaptive filter, which is preset based on signal characteristics and empirical values. To maintain weights for the period, and It is used to control the adjustment strength of the periodicity retention coefficient on the effective update step size. The periodicity retention coefficient, and .

[0082] The main controller 4 re-executes the noise component estimation and subtraction operations in step two with the updated effective update step size to obtain a new denoised candidate signal. It then recalculates the envelope periodic peak stability of this new denoised candidate signal, generates a new period preservation coefficient, and adjusts the effective update step size accordingly. This process is iteratively repeated until the envelope periodic peak stability of the denoised candidate signal meets a preset convergence condition. The preset convergence condition is at least one of the following: The period preservation coefficient is greater than a preset threshold, which is 0.85. Or the change in the effective update step size between two adjacent iterations is less than a preset threshold. Or the current iteration count reaches a preset maximum iteration count. When any of the above conditions are met, the main controller 4 determines that the iteration has converged and uses the current denoised candidate signal as the final denoised cardiopulmonary sound signal.

[0083] Step 4: Output state features and pet state classification results;

[0084] The main controller 4 extracts state features based on the denoised cardiopulmonary sound signal obtained in step three. (Refer to...) Figure 4 The main controller 4 separates the heart sound channel and lung sound channel in the denoised heart and lung sound signals. In a preferred embodiment, the main controller 4 inputs the denoised heart and lung sound signals into the heart sound bandpass filter and the lung sound bandpass filter, respectively.

[0085]

[0086] in, For heart sound bandpass filter, It serves as a bandpass filter for lung sounds or breath sounds. For the denoised heart sound channel signal, This is for the denoised lung sound channel signal. The passband frequency range of the heart sound bandpass filter is 20Hz to 200Hz, used to extract heart sound components to obtain the denoised heart sound channel signal. The passband frequency range of the lung sound bandpass filter is 100Hz to 2000Hz, used to extract lung sound or breath sound components to obtain the denoised lung sound channel signal. Both bandpass filters are implemented using either finite impulse response (FIR) filters or infinite impulse response (IR) filters.

[0087] The main controller 4 performs envelope extraction and peak detection on the denoised heart sound channel signal and the denoised lung sound channel signal, respectively. The envelope extraction uses the Hilbert transform method or the rectification smoothing method to obtain the heart sound envelope curve and the lung sound envelope curve.

[0088] The main controller 4 performs peak detection on the heart sound envelope curve, marks the position of each periodic peak in the envelope curve, and obtains the heartbeat peak sequence. Peak detection was performed on the lung sound envelope curve, and the positions of each periodic peak in the envelope curve were marked to obtain the respiratory peak sequence. Peak detection employs a local maximum detection method, identifying the sampling point with the largest amplitude within a continuous region as the peak point, and setting a minimum peak interval to prevent misidentification of multiple fluctuations within the same period as multiple independent peaks. (Minimum interval for heartbeat peaks) Determine using the following formula:

[0089] ;

[0090] in, This is the preset maximum effective heart rate, measured in beats per minute. When the time interval between two adjacent heart rate peaks is less than... At that time, the main controller 4 determines the latter to be an invalid peak and removes it. Minimum interval of respiratory peaks. Determine using the following formula:

[0091] ;

[0092] in, To preset the maximum effective respiratory rate, the unit is breaths per minute. When the time interval between two adjacent respiratory peaks is less than... At that time, the main controller 4 determines that the latter is an invalid peak and removes it.

[0093] The main controller 4 calculates the heart rate based on the heart rate peak sequence. and heart rhythm fluctuations :

[0094]

[0095]

[0096] in, This is a sequence of time intervals between adjacent heartbeat peaks;

[0097] The main controller 4 calculates the respiratory rate based on the respiratory peak sequence. and respiratory rhythm fluctuations :

[0098] ,

[0099] in, This is a sequence of time intervals between adjacent respiratory peaks. The main controller 4 calculates the energy percentage of heart sounds. and the proportion of lung sound energy :

[0100]

[0101] The main controller 4 also calculates the spectral flatness. Silence ratio Heart rate relative to resting baseline ratio and the ratio of respiratory rate to resting baseline .

[0102] Spectral flatness The calculation method is to perform a Fourier transform on the denoised cardiopulmonary sound signal to obtain the power spectrum, and then calculate the ratio of the geometric mean to the arithmetic mean of the power spectrum.

[0103] Silent ratio The calculation method is to count the proportion of sampling points whose signal amplitude is lower than the preset silence threshold within the set time window to the total number of sampling points.

[0104] Heart rate ratio relative to resting baseline The calculation method is as follows:

[0105]

[0106] respiratory rate relative to resting baseline The calculation method is as follows:

[0107]

[0108] , It can be the system default value, or it can be obtained by statistical analysis of multiple heart and lung sounds from the same pet in a resting state.

[0109] The main controller 4 is built to include heart rate monitoring. respiratory rate Heart rhythm fluctuations Circadian rhythm fluctuations Heart sound energy ratio Lung sound energy ratio Spectral flatness Silence ratio Heart rate relative to resting baseline ratio and the ratio of respiratory rate to resting baseline Cardiac and lung sound state feature vector .

[0110] The main controller 4 will generate the cardiopulmonary sound state feature vector. Matching with preset state characteristic standards. These preset state characteristic standards are stored in the main controller 4's built-in memory or external storage medium, organized in the form of lookup tables or databases. Each preset state characteristic standard contains multiple state labels, each corresponding to a set of characteristic threshold intervals and a set of characteristic weights. The state labels include five categories: calm or resting, excited or post-activity, tense or anxious, panting or rapid breathing, and noisy or ineffective. Figure 5 As shown, in the cardiopulmonary sound characteristic determination area corresponding to different pet states, the ratio of heart rate to resting baseline and the ratio of respiratory rate to resting baseline can be used to reflect the degree of increase in heart rate and respiratory rate relative to resting baseline. The proportion of lung sound energy and respiratory rhythm fluctuations can be used to help distinguish rhythm fluctuations and changes in lung sound energy. Through this determination area, the system can distinguish between calm, excited, tense, panting, and invalid noise states based on noise reduction features.

[0111] For each state label, the main controller 4 determines the matching score of that state label based on the weighted scoring results of each feature in the heart and lung sound state feature vector falling into the corresponding threshold interval:

[0112]

[0113] in, State feature vector The first in One characteristic, status label Corresponding to the The weights of each feature, and The feature is represented by the state label. The lower and upper limits, For indicator functions. When the feature Fall into the corresponding status label When the threshold range is reached, Select 1; otherwise select 0.

[0114] The main controller 4 determines the pet's state classification result based on the matching scores of each state label and preset priority rules. The preset priority rules are any one of the following: The state label with the highest matching score is selected as the final pet state classification result. When multiple state labels have the same matching score, the final pet state classification result is determined according to a preset priority order: calm takes precedence over excited, and inactive takes precedence over all other states. When the matching score of a noise or invalid label reaches a preset threshold, the noise or invalid classification result is output first to avoid outputting misleading valid state results when signal quality is poor.

[0115] The feature threshold ranges corresponding to each status label are pre-set based on pet species, age, size, and clinical experience. The main controller 4 sends the pet status classification results to a mobile terminal or display device via a wireless communication module. The wireless communication module uses Bluetooth Low Energy or Wi-Fi communication. The health management application running on the mobile terminal receives and displays the pet status classification results, status characteristics, and judgment basis text. The judgment basis text is automatically generated based on the threshold hit rate in the preset status feature standards, explaining the correspondence between the current pet status classification result and features such as increased heart rate, increased respiratory rate, increased rhythm fluctuations, enhanced lung sound energy, or abnormal signal quality.

[0116] Experimental verification: such as Figure 6 As shown in (a), in the actual scenario of collecting data while the pet collar is being worn, the original cardiopulmonary sound signal contains non-target interference such as collar movement, fur friction, changes in body surface contact, and ambient noise. This manifests as sudden spikes and local amplitude anomalies in the time-domain waveform, as well as mid-to-high frequency spurious noise in the frequency domain. After the dual-channel adaptive denoising processing described in this application, as... Figure 6 As shown in (b), the sudden interference components in the denoised cardiopulmonary sound signal are suppressed, the time domain waveform tends to be stable overall, and the periodic structure of cardiopulmonary sound is relatively clearer.

[0117] Further integration Figure 6The power spectral density comparison between (c) and (d) shows that the original cardiopulmonary sound signal contains significant spurious noise components in the mid-to-high frequency range. After processing by this application, the effective acoustic structure within the main frequency band of the heart sound is preserved, and the energy of mid-to-high frequency spurious noise is reduced. This demonstrates that this application does not simply smooth the signal through fixed-band filtering, but rather uses a reference acoustic channel to adaptively estimate and cancel wearing interference, thereby reducing non-target noise while preserving the periodic acoustic components related to the pet's cardiopulmonary activity.

[0118] When further performing envelope extraction and periodic peak detection on the denoised cardiopulmonary sound signal, such as Figure 6 As shown in (e), the periodic peaks in the envelope curve are easier to identify, providing a more stable input for subsequent construction of heart rate, respiratory rate, heart rhythm fluctuations, respiratory rhythm fluctuations, heart sound energy proportion, lung sound energy proportion, and cardiopulmonary sound state feature vectors. Through the above processing, this application can improve the usability and reliability of state feature extraction of pet collar cardiopulmonary sound signals under the condition of wearing interference, and is suitable for daily pet state observation, non-diagnostic trend prompts, and state result display on smart collars or mobile terminals.

[0119] Furthermore, such as Figures 7 to 10 As shown in (a) and (b), the original cardiopulmonary sound spectrograms exhibit different forms of non-target interference or changes in effective structure in the spectrograms, considering factors such as hair friction interference, vocal / harmonic interference, mixed environmental and friction interference, and attenuation from far-field acquisition or tissue filtering in the real-world samples. Hair friction interference mainly manifests as local broadband spurious textures and short-term energy disturbances; vocal / harmonic interference mainly manifests as local strong energy components and interference with periodic structures; mixed environmental and friction interference manifests as increased background noise and more mid-to-high frequency spurious textures; and attenuation from far-field acquisition or tissue filtering manifests as weakened effective cardiopulmonary sound structures, discontinuous energy distribution, or blurred periodic structures. After the dual-channel adaptive denoising processing described in this application, some non-target spurious components and background noise are suppressed, and the low-frequency periodic structures related to cardiopulmonary sounds are relatively clearer in several segments, indicating that the reference acoustic channel can provide effective information for wear interference estimation and signal cleansing.

[0120] At the same time, such as Figures 7 to 10 As shown in (c), when performing envelope extraction and periodic peak detection on the denoised cardiopulmonary sound signal, the main periodic peaks in the envelope curve can be identified, and the intervals between periodic peaks in some segments are clearer. This indicates that, while suppressing interference from hair friction, vocalization / harmonic interference, and mixed environmental and frictional interference, this application can retain effective periodic components related to the pet's cardiopulmonary activity, providing a usable and relatively stable signal basis for subsequent construction of heart rate, respiratory rate, heart rhythm fluctuations, respiratory rhythm fluctuations, and cardiopulmonary sound state feature vectors.

[0121] While this application has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of this application. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of this application as defined by the appended claims. It should be understood that different dependent claims and features herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other embodiments.

Claims

1. A dual-channel adaptive interference suppression pet heart and lung sound signal monitoring device, characterized in that, include: The collar body has a first sound acquisition unit and a second sound acquisition unit disposed inside the collar body. The first sound acquisition unit is used to acquire the main signal of heart and lung sounds, and the second sound acquisition unit is used to acquire the reference acoustic signal. The main controller is connected to both the first sound acquisition unit and the second sound acquisition unit. The main controller is configured as follows: Based on the main cardiopulmonary sound signal and the reference acoustic signal, a wear interference reference component is constructed; Based on the wearing interference reference component, the noise component in the main cardiopulmonary sound signal is estimated by an adaptive filter, and the noise component is subtracted from the main cardiopulmonary sound signal to obtain a denoised candidate signal; The effective update step size of the adaptive filter is adjusted according to the envelope periodic peak stability of the denoised candidate signal. The estimation and subtraction of the noise components are re-executed with the updated effective update step size until the envelope periodic peak stability of the denoised candidate signal meets the preset convergence condition, thereby obtaining the denoised cardiopulmonary sound signal. Based on the noise-reduced cardiopulmonary sound signal output status characteristics and pet status classification results.

2. The pet heart and lung sound signal monitoring device according to claim 1, characterized in that, The first sound acquisition unit is located on the inner side of the collar body at a position corresponding to the pet's neck or chest conduction area, and the second sound acquisition unit is located at another position on the inner side of the collar body, spaced apart from the first sound acquisition unit.

3. The pet heart and lung sound signal monitoring device according to claim 1, characterized in that, A method for adjusting the effective update step size of the adaptive filter based on the envelope periodic peak stability of the denoised candidate signal includes: Extract the heart sound envelope and lung sound envelope of the denoised candidate signal; Calculate the periodic peak confidence of the heart sound envelope and the periodic peak confidence of the lung sound envelope respectively, and generate the periodicity preservation coefficient; The effective update step size of the adaptive filter is adjusted according to the period preservation coefficient, wherein the period preservation coefficient is negatively correlated with the effective update step size.

4. The pet heart and lung sound signal monitoring device according to claim 3, characterized in that, The method for adjusting the effective update step size of the adaptive filter based on the period preservation coefficient is as follows: ； in, Let n be the effective update step size of the adaptive filter at the nth sampling time. Let be the initial step size of the adaptive filter. To maintain weights for the period, and , The periodicity retention coefficient is given, and .

5. The pet heart and lung sound signal monitoring device according to claim 1, characterized in that, The method for constructing the wearing interference reference component includes: The main signal of cardiopulmonary sounds and the reference acoustic signal are synchronously preprocessed to obtain the main path preprocessed signal and the reference path preprocessed signal; Calculate the short-time cross-correlation coefficient between the main path preprocessed signal and the reference path preprocessed signal; when the short-time cross-correlation coefficient, the short-time energy change of the reference path preprocessed signal, and the spectral flatness of the reference path preprocessed signal all meet the preset interference judgment conditions, output the reference path preprocessed signal as the wear interference reference component.

6. The pet heart and lung sound signal monitoring device according to claim 5, characterized in that, The short-time cross-correlation coefficient is: ； in, This indicates the sampling time corresponding to the starting sampling point of the current short-time analysis window. This indicates the sampling point number within the current short-term analysis window. Indicates the number of times the short-term analysis window is currently active. The sampling time corresponding to each sampling point; To prevent extremely small constants with a denominator of zero; This indicates the main path preprocessed signal at the sampling time. The signal value at that location; This indicates the reference path preprocessed signal at the sampling time. The signal value at that location; Indicates the number of sampling points contained in the short-time analysis window; This represents the short-time cross-correlation coefficient between the main path preprocessed signal and the reference path preprocessed signal within the current short-time analysis window.

7. The pet heart and lung sound signal monitoring device according to claim 1, characterized in that, The method based on the output state characteristics of the denoised cardiopulmonary sound signal includes: The denoised heart and lung sound signals are separated into heart sound channels and lung sound channels to obtain denoised heart sound channel signals and denoised lung sound channel signals. Envelope extraction and peak detection were performed on the denoised heart sound channel signal and the denoised lung sound channel signal, respectively, to obtain the heartbeat peak sequence and the respiratory peak sequence. Heart rate and heart rhythm fluctuations are calculated based on the aforementioned heart rate peak sequence; Calculate the respiratory rate and respiratory rhythm fluctuations based on the respiratory peak sequence; Calculate the energy percentage of heart sounds and the energy percentage of lung sounds; Construct and output a state feature vector that includes heart rate, respiratory rate, heart rhythm fluctuation, respiratory rhythm fluctuation, heart sound energy ratio, lung sound energy ratio, spectral flatness, silence ratio, heart rate ratio relative to resting baseline, and respiratory rate ratio relative to resting baseline.

8. The pet heart and lung sound signal monitoring device according to claim 7, characterized in that, The peak detection includes setting a minimum interval for heartbeat peaks and a minimum interval for respiratory peaks. The minimum interval for heartbeat peaks is determined based on a preset maximum effective heart rate, and the minimum interval for respiratory peaks is determined based on a preset maximum effective respiratory rate.

9. The pet heart and lung sound signal monitoring device according to claim 1, characterized in that, The pet status classification results are output based on the denoised cardiopulmonary sound signals, specifically including: The cardiopulmonary sound state feature vector is matched with a preset state feature standard, which includes the feature threshold range and feature weight corresponding to each state label. Based on the weighted scoring results of each feature in the cardiopulmonary sound state feature vector falling into the corresponding threshold interval, the matching score of each state label is determined, and the pet state classification result is determined according to the preset priority rules.

10. The pet heart and lung sound signal monitoring device according to claim 9, characterized in that, The status labels include calm or resting, excited or after activity, tense or anxious, panting or shortness of breath, and noisy or ineffective.

Images