A method and device for separating and detecting a swallowing signal based on wavelet transform, and a storage medium

Abstract

This invention discloses a method, apparatus, and storage medium for separating and detecting swallowing signals based on wavelet transform. It relates to the field of data processing technology and facilitates the accurate detection and separation of swallowing signals. The method includes: configuring a db4 wavelet basis as the basis function for multi-scale signal decomposition; configuring each decomposition scale as a dynamic adjustment parameter of the db4 wavelet basis; adjusting each decomposition scale based on the basic information of the target population; performing multi-scale decomposition on the preprocessed mixed vibration signal based on the db4 wavelet basis and the adjusted corresponding decomposition scales to obtain multiple wavelet components at different scales; calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and filtering on the wavelet components at multiple different scales, separating the effective swallowing component and the component corresponding to the interference signal; the swallowing signal feature template library is configured with thresholds for filtering feature parameters; and reconstructing the effective swallowing component in reverse to obtain the effective swallowing signal.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus and storage medium for separating and detecting swallowing signals based on wavelet transform. Background Technology

[0002] Dysphagia is a common complication in people with stroke, Parkinson's disease, or post-operative head and neck tumors. Clinical nutritional intervention for these individuals requires a tailored food texture plan based on an objective assessment of their swallowing ability to reduce the risk of aspiration and ensure adequate nutrient intake. Swallowing signals, as core data reflecting swallowing ability, are crucial, and the accuracy of their detection directly impacts the rationality of the food texture plan, thus affecting aspiration risk control and nutrient intake assurance.

[0003] In existing technologies, swallowing signals are primarily acquired by sensor devices positioned in the neck, which collect mixed vibration signals containing both valid swallowing signals and interference signals. After preprocessing, the mixed signals are filtered to separate the valid swallowing signals. However, in existing technologies, the sensor signal filtering and feature extraction often rely on fixed parameter settings or single-dimensional feature judgments. These methods fail to adapt to the inherent low-frequency and smooth characteristics of swallowing signals and the individual differences in swallowing signals among people with different etiologies. Consequently, it is difficult to accurately separate valid swallowing signals from interference signals, affecting the accuracy of the final swallowing signal. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus and storage medium for swallowing signal separation and detection based on wavelet transform, which facilitates more accurate detection and separation of swallowing signals.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: In a first aspect, embodiments of the present invention provide a method for separating and detecting swallowing signals based on wavelet transform, comprising: acquiring a mixed vibration signal containing effective swallowing signals and interference signals through a flexible sensing patch, and preprocessing the mixed vibration signal; the flexible sensing patch being attached to the skin surface area between the hyoid bone and the larynx of a target population; configuring a db4 wavelet basis as the basis function for multi-scale decomposition of the signal, configuring each decomposition scale as a dynamic adjustment parameter of the db4 wavelet basis, and adjusting each decomposition scale based on the basic information of the target population; the basic information of the target population including age and etiological types of swallowing-related complications; performing multi-scale decomposition of the preprocessed mixed vibration signal based on the db4 wavelet basis and the adjusted corresponding decomposition scale to obtain multiple wavelet components of different scales; calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening on the multiple wavelet components of different scales to separate the effective swallowing component and the component corresponding to the interference signal; the swallowing signal feature template library being configured with thresholds for screening feature parameters; and reconstructing the effective swallowing component in reverse to obtain the effective swallowing signal.

[0006] According to one embodiment of the present invention, the preprocessing of the hybrid vibration signal includes: performing analog-to-digital conversion on the acquired analog hybrid vibration signal to convert the analog signal into a digital hybrid vibration signal; performing baseline drift correction on the digital hybrid vibration signal using a linear correction algorithm to eliminate drift interference in the low-frequency band; and performing amplitude normalization on the corrected digital hybrid vibration signal according to a normalization algorithm to uniformly map the amplitude of the normalized digital hybrid vibration signal to a preset standard range.

[0007] According to one embodiment of the present invention, adjusting each decomposition scale based on the basic information of the target population includes: setting an initial decomposition scale of the db4 wavelet basis, the initial decomposition scale corresponding to a split interval covering the conventional frequency range of the swallowing signal; when the target population is a stroke recovery group, adjusting the initial decomposition scale in a hierarchical increasing direction; when the target population is a Parkinson's disease group, keeping the initial decomposition scale unchanged, and covering the core frequency range of the swallowing signal of the Parkinson's disease group through the frequency split interval corresponding to the initial decomposition scale; and determining, according to a signal decomposition accuracy verification algorithm, whether the adjusted or unchanged db4 wavelet basis decomposition scale meets the preset swallowing signal feature parameter extraction completeness requirement, wherein the signal decomposition accuracy verification algorithm is an algorithm used to evaluate the wavelet component feature parameter extraction completeness.

[0008] According to one embodiment of the present invention, the step of determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction accuracy requirement based on the signal decomposition accuracy verification algorithm includes: extracting feature parameters of the multiple wavelet components of different scales obtained by decomposition at the adjusted or unchanged db4 wavelet basis decomposition scale; calculating the extraction completeness of the feature parameters of the multiple wavelet components of different scales; comparing the extraction completeness of the feature parameters of the multiple wavelet components of different scales with a preset swallowing signal feature parameter completeness threshold; if the feature parameter extraction completeness is higher than the preset swallowing signal feature parameter completeness threshold, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction completeness requirement; if the feature parameter extraction completeness is lower than the preset swallowing signal feature parameter completeness threshold, then the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the feature parameter extraction completeness reaches the preset swallowing signal feature parameter completeness threshold.

[0009] According to one embodiment of the present invention, the step of calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening on the multiple wavelet components of different scales to separate the effective swallowing components and the components corresponding to the interference signals includes: calling the preset swallowing signal feature template library to extract multi-dimensional screening feature parameters of the multiple wavelet components of different scales, wherein the multi-dimensional screening feature parameters include amplitude peak value, frequency center, and signal duration; comparing the extracted amplitude peak value, frequency center, and signal duration with the feature parameter thresholds of the corresponding target population in the swallowing signal feature template library; screening out a first wavelet component whose amplitude peak value, frequency center, and signal duration all fall within the corresponding feature parameter threshold range, and determining the first wavelet component as the effective swallowing component; identifying a second wavelet component whose amplitude peak value, frequency center, and signal duration exceed the corresponding feature parameter threshold range, determining the second wavelet component as the component corresponding to the interference signal, and performing a rejection operation.

[0010] According to one embodiment of the present invention, the step of reconstructing the effective swallowing components in reverse to obtain an effective swallowing signal includes: performing scale-level matching on the effective swallowing components to determine the db4 wavelet basis decomposition scale-level information corresponding to each effective swallowing component, wherein the decomposition scale-level information includes a decomposition level number and a frequency splitting interval corresponding to the level number; performing db4 wavelet inverse transform operation to determine the superposition order of the effective swallowing components based on the decomposition level number; and performing frequency calibration on the corresponding inverse transform signal components based on the corresponding frequency splitting interval to make the frequency range of each signal component consistent with the frequency splitting interval during decomposition, thereby obtaining a preliminary time-domain swallowing signal; and using a moving average algorithm to smooth and correct the signal amplitude of the preliminary time-domain swallowing signal to obtain the final effective swallowing signal.

[0011] Secondly, embodiments of the present invention also provide a swallowing signal detection device based on wavelet transform, comprising: a flexible sensing patch, attached to the skin surface area between the hyoid bone and the larynx in front of the neck of a target population, for collecting a mixed vibration signal containing a valid swallowing signal and an interference signal; a server configured to: preprocess the mixed vibration signal, and configure a db4 wavelet basis as the basis function for multi-scale decomposition of the signal, configure each decomposition scale as a dynamic adjustment parameter of the db4 wavelet basis, and adjust each decomposition scale based on the basic information of the target population; the basic information of the target population includes age and the etiological type of swallowing-related complications; perform multi-scale decomposition of the preprocessed mixed vibration signal based on the db4 wavelet basis and the adjusted corresponding decomposition scale to obtain multiple wavelet components of different scales; call a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening on the multiple wavelet components of different scales, and separate the components corresponding to the valid swallowing component and the interference signal; the swallowing signal feature template library is configured with thresholds for screening feature parameters; and perform reverse reconstruction on the valid swallowing component to obtain the valid swallowing signal.

[0012] According to one embodiment of the present invention, adjusting each decomposition scale based on the basic information of the target population includes: setting an initial decomposition scale of the db4 wavelet basis, the initial decomposition scale corresponding to a split interval covering the conventional frequency range of the swallowing signal; when the target population is a stroke recovery group, adjusting the initial decomposition scale in a hierarchical increasing direction; when the target population is a Parkinson's disease group, keeping the initial decomposition scale unchanged, and covering the core frequency range of the swallowing signal of the Parkinson's disease group through the frequency split interval corresponding to the initial decomposition scale; and determining, according to a signal decomposition accuracy verification algorithm, whether the adjusted or unchanged db4 wavelet basis decomposition scale meets the preset swallowing signal feature parameter extraction completeness requirement, wherein the signal decomposition accuracy verification algorithm is an algorithm used to evaluate the wavelet component feature parameter extraction completeness.

[0013] According to one embodiment of the present invention, the step of determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction accuracy requirement based on the signal decomposition accuracy verification algorithm includes: extracting feature parameters of the multiple wavelet components of different scales obtained by decomposition at the adjusted or unchanged db4 wavelet basis decomposition scale; calculating the extraction completeness of the feature parameters of the multiple wavelet components of different scales; comparing the extraction completeness of the feature parameters of the multiple wavelet components of different scales with a preset swallowing signal feature parameter completeness threshold; if the feature parameter extraction completeness is higher than the preset swallowing signal feature parameter completeness threshold, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction completeness requirement; if the feature parameter extraction completeness is lower than the preset swallowing signal feature parameter completeness threshold, then the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the feature parameter extraction completeness reaches the preset swallowing signal feature parameter completeness threshold.

[0014] The present invention also provides a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the wavelet transform-based swallowing signal separation and detection method described in any of the first aspects.

[0015] The swallowing signal separation and detection method, device, and storage medium based on wavelet transform provided in this invention collects and preprocesses mixed vibration signals through a flexible sensing patch, configures the db4 wavelet basis as the signal multi-scale decomposition basis function, dynamically adjusts each decomposition scale in combination with the age of the target population and the etiology of swallowing-related complications, performs multi-scale decomposition on the preprocessed signal based on the adjusted db4 wavelet basis and decomposition scale, calls a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening of wavelet components and separate effective swallowing components, and finally reconstructs the effective swallowing components in reverse. By improving the db4 wavelet basis and configuring it as a multi-scale decomposition basis function for the signal, and dynamically adjusting each decomposition scale of the db4 wavelet basis in combination with the age of the target population and the etiology of swallowing-related complications, the traditional fixed parameter setting mode is replaced. Based on the adjusted db4 wavelet basis and decomposition scale, the preprocessed signal is decomposed into multiple wavelet components of different scales. Then, through a preset swallowing signal feature template library, multi-dimensional feature parameter matching and screening are performed on the wavelet components. Multi-dimensional judgment replaces single-dimensional feature judgment, which can better adapt to the individual differences in swallowing signals of different etiological populations, thereby improving the separation accuracy of effective swallowing components and interference signals. Therefore, it is easier to detect and separate swallowing signals more accurately. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a wavelet transform-based swallowing signal separation and detection method according to an embodiment of this application; Figure 2 for Figure 1 A schematic flowchart of an embodiment of step S120; Figure 3 for Figure 1 A schematic flowchart of an embodiment of step S150; Figure 4 This is a schematic diagram of the structure of one embodiment of the server in this application. Detailed Implementation

[0018] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0019] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application. Example 1

[0020] This invention provides a wavelet transform-based method for separating and detecting swallowing signals. It involves acquiring and preprocessing mixed vibration signals using a flexible sensing patch, configuring the db4 wavelet basis as the signal's multi-scale decomposition basis function, and dynamically adjusting each decomposition scale based on the target population's age and the etiology of swallowing-related complications. The preprocessed signal is then decomposed into multiple scales based on the adjusted db4 wavelet basis and decomposition scales. A pre-defined swallowing signal feature template library is then used to perform multi-dimensional feature parameter matching and screening of the wavelet components to separate the effective swallowing components. Finally, the effective swallowing components are reconstructed in reverse. This method replaces the existing technology's reliance on fixed parameters and single-dimensional feature judgment for signal screening. It accurately adapts to the inherent low-frequency, smooth characteristics of swallowing signals and individual differences in swallowing signals among different etiologies, solving the technical problem of inaccurate separation of effective swallowing signals and interference signals, which affects the accuracy of signal acquisition. This improves the separation accuracy and detection reliability of effective swallowing signals.

[0021] Figure 1 This is a schematic flowchart of a wavelet transform-based swallowing signal separation and detection method according to an embodiment of this application. (See attached diagram) Figure 1 The method includes: S110, acquiring a mixed vibration signal containing effective swallowing signals and interference signals through a flexible sensing patch, and preprocessing the mixed vibration signal; the flexible sensing patch is attached to the skin surface area in front of the neck of the target population, between the hyoid bone and the larynx.

[0022] Specifically, in this embodiment, the flexible sensing patch can be made of a flexible piezoelectric film material with a thickness of 0.1~0.3mm, and can be closely attached to the skin surface area in front of the neck of the target person, between the hyoid bone and the larynx. Since this area is the core area of ​​laryngeal cartilage vibration during swallowing, it can capture the low-frequency vibration signals generated by swallowing muscle contraction and larynx elevation to the greatest extent. Of course, it will also inevitably collect interference signals such as respiratory fluctuations, involuntary neck muscle spasms, and slight vibrations of the external environment. Therefore, the original signal collected is a mixed vibration signal containing effective swallowing signals and interference signals.

[0023] Unprocessed hybrid vibration signals suffer from issues such as baseline drift, inconsistent amplitude ranges, and incompatible formats, making them unsuitable for direct use in subsequent multi-scale decomposition. In this embodiment, preprocessing primarily aims to eliminate redundant interference and standardize the signal format, providing high-quality input signals for subsequent multi-scale decomposition.

[0024] S120. Configure the db4 wavelet basis as the basis function for multi-scale decomposition of the signal, configure each decomposition scale as a dynamic adjustment parameter of the db4 wavelet basis, and adjust each decomposition scale based on the basic information of the target population; the basic information of the target population includes age and etiological type of swallowing-related complications.

[0025] There are many types of wavelet bases, such as the Haar wavelet base and the db4 wavelet base. In this embodiment, the db4 wavelet base is selected as the decomposition basis function. This wavelet base belongs to the compactly supported orthogonal wavelet base, with a support length of 8 and a vanishing moment of 4. It has good low-frequency characteristics and smoothness. Since its waveform is highly matched with the inherent low-frequency and slowly varying characteristics of the swallowing signal, it can retain the feature information of the effective swallowing signal to the greatest extent during the decomposition process, while suppressing high-frequency interference signals. Compared with other types of wavelet bases such as the Haar wavelet base, it can effectively avoid the problem of loss of effective features.

[0026] In existing technologies, the decomposition scale is generally a fixed parameter. However, in this application, the decomposition scale is configured as a dynamically adjustable parameter of the db4 wavelet basis, meaning the decomposition scale can be dynamically adjusted to adapt to different target populations. The number of decomposition scale levels relates to the precision of frequency interval division; one decomposition scale level corresponds to one frequency interval, and the more levels there are, the more refined the frequency interval division.

[0027] The target population is divided into three age ranges: youth (18-40 years old), middle-aged (41-60 years old), and elderly (61 years old and above). The etiologies of swallowing-related complications include stroke recovery period and Parkinson's disease. People in the stroke recovery period generally have low swallowing vibration signal amplitude and concentrated frequency distribution, while people with Parkinson's disease show stable swallowing vibration signal frequency characteristics.

[0028] S130. Based on the db4 wavelet basis and the adjusted corresponding decomposition scale, the preprocessed hybrid vibration signal is decomposed into multiple scales to obtain multiple wavelet components of different scales.

[0029] Specifically, in this embodiment, the Mallat fast wavelet decomposition algorithm is invoked by the backend server. Based on the configured db4 wavelet basis and the adjusted decomposition scale, a layer-by-layer decomposition operation is performed on the preprocessed digital hybrid vibration signal. The Mallat fast wavelet decomposition algorithm is a mature existing algorithm; to highlight the innovative aspect of this application, its technical details will not be elaborated further. Since the Mallat fast wavelet decomposition algorithm does not require direct convolution of the wavelet function, signal decomposition can be achieved iteratively using low-pass and high-pass filter banks, thereby significantly improving computational efficiency.

[0030] In the Mallat fast wavelet decomposition algorithm, each level of decomposition uses low-pass and high-pass filters to separate the input signal into approximate and detail components. The approximate component corresponds to the low-frequency part of the signal, typically 5-20 Hz, and mainly contains the effective swallowed signal; the detail component corresponds to the high-frequency part of the signal, typically above 20 Hz, and mainly contains interference signals. After N levels of decomposition, N+1 wavelet components of different scales are finally obtained, including one Nth-level approximate component and N Nth-level detail components. Each wavelet component corresponds to an independent frequency range, and there is no frequency overlap between the two components.

[0031] S140. Call the preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and filtering on the multiple wavelet components of different scales, and separate the effective swallowing components and the components corresponding to the interference signals; the swallowing signal feature template library is configured with thresholds for filtering feature parameters.

[0032] Specifically, the swallowing signal feature template library is a database pre-constructed using sample statistical analysis. The backend server can directly call this template library to complete feature matching. For example, before step S110, the method further includes: constructing the swallowing signal feature template library, specifically by: collecting over a thousand valid swallowing signal samples from target populations of different ages and etiologies; extracting the amplitude peak, frequency center, and signal duration of each sample; determining the threshold range of characteristic parameters corresponding to different populations by calculating statistical quantities such as parameter mean, standard deviation, and extreme values; and finally associating and storing information such as population classification labels, characteristic parameter thresholds, and signal frequency intervals to form the swallowing signal feature template library.

[0033] In this embodiment, a multi-dimensional feature parameter matching and screening method is adopted. A threshold comparison algorithm is used to separate effective components from interference components. Compared with the single-dimensional judgment method of existing technologies, this effectively avoids missed detections and false detections. Specifically, multi-dimensional feature parameters of each wavelet component are extracted and compared one by one with the threshold range of the corresponding population in the template library. Components that meet the threshold range are considered effective swallowing components, while those that do not are considered interference signals, thus achieving accurate separation of effective swallowing components and interference signal components.

[0034] S150. The effective swallowing component is reconstructed in reverse to obtain the effective swallowing signal.

[0035] Specifically, reverse reconstruction is the inverse process of multi-scale decomposition. In this embodiment, the backend server calls the Mallat wavelet inverse transform algorithm to reconstruct and superimpose the selected effective swallowing components layer by layer according to the hierarchical order of decomposition, restoring them into a time-domain signal that can directly characterize the swallowing action. This time-domain signal is the effective swallowing signal. The reconstructed effective swallowing signal has removed interfering components such as breathing and muscle spasms, and can be used for subsequent swallowing function assessments, such as swallowing frequency statistics and swallowing amplitude analysis, providing data reference for the nutrition department to formulate nutritional food texture plans.

[0036] In some embodiments, the preprocessing of the hybrid vibration signal includes: performing analog-to-digital conversion on the acquired analog hybrid vibration signal to convert the analog signal into a digital hybrid vibration signal; and applying a linear correction algorithm to perform baseline drift correction on the digital hybrid vibration signal to eliminate drift interference in the low-frequency band. Exemplarily, the linear correction algorithm employs a least-squares linear correction algorithm, specifically: selecting the signal from the non-swallowing period in the digital hybrid vibration signal as a baseline fitting sample, constructing a baseline linear fitting equation y=ax+b using the least-squares method, where a is the slope, b is the intercept, x is the abscissa of the sampling point (time point), and y is the fitted baseline amplitude; substituting the abscissa of each sampling point to calculate the corresponding baseline amplitude, i.e., the baseline drift; and subtracting the corresponding baseline drift from the actual amplitude of the digital signal to complete the baseline correction.

[0037] The amplitude of the corrected digital hybrid vibration signal is normalized using a normalization algorithm, mapping the normalized amplitude to a preset standard range. For example, the normalization algorithm used is the min-max normalization algorithm, with the preset standard range being [0,1]. This normalization operation enables standardized processing of signal amplitudes from different groups and in different acquisition scenarios, facilitating subsequent unified comparison of multi-dimensional feature parameters.

[0038] See Figure 2 As shown, in some embodiments, step S120, adjusting each decomposition scale based on the basic information of the target population, is executed by a backend server. Specifically, this includes: S121, setting an initial decomposition scale for a preset db4 wavelet basis. This initial decomposition scale corresponds to a segmentation interval covering the conventional frequency range of the swallowing signal. Specifically, the conventional frequency range of the swallowing signal is generally 5~20Hz, and the corresponding number of initial decomposition scale levels can be set to 3. The frequency segmentation intervals corresponding to the 3 decomposition scale levels are: Level 1 10~20Hz, Level 2 5~10Hz, and Level 3 0~5Hz. These 3 intervals can completely cover the frequency range of the conventional swallowing signal, meeting the swallowing signal detection needs of the general population.

[0039] S122A. When the target population is individuals recovering from stroke, the initial decomposition scale is adjusted in a hierarchical manner. Because the swallowing function of stroke survivors is weak, the vibration signal amplitude generated by swallowing is low, and the signal frequency distribution is more concentrated, requiring finer frequency range division to extract effective signals. Therefore, the initial decomposition scale can be adjusted from level 3 to level 5, with the newly added level 4 and level 5 frequency division ranges being 2.5~5Hz and 0~2.5Hz respectively. Through this hierarchical adjustment, the division of the low-frequency range is further refined, achieving accurate capture of low-amplitude swallowing signals.

[0040] S122B. When the target population is Parkinson's disease patients, the initial decomposition scale remains unchanged, and the core frequency range of the swallowing signal of Parkinson's disease patients is covered by the frequency segmentation interval corresponding to the initial decomposition scale. The frequency characteristics of the swallowing signal of Parkinson's disease patients are generally relatively stable, with a core frequency range of 8~15Hz, which can fall within the frequency range of the 3-level initial decomposition scale. Therefore, there is no need to adjust the number of decomposition scale levels, and it can also adapt to the stable characteristics of the swallowing signal of Parkinson's disease patients.

[0041] S123. Based on the signal decomposition accuracy verification algorithm, determine whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of swallowing signal feature parameter extraction; the signal decomposition accuracy verification algorithm is an algorithm used to evaluate the completeness of wavelet component feature parameter extraction.

[0042] For example, the signal decomposition accuracy verification algorithm can employ the F1 score algorithm. The input consists of wavelet components at the adjusted decomposition scale and their corresponding labels, where the labels are either effective swallowed components or interference components. Feature parameters such as the amplitude peak and frequency center of each wavelet component are extracted, and the effective feature recall and precision are calculated. When the weighting coefficients for both recall and precision are 0.5, the calculation is performed according to the weighting formula... Calculate the F1 score, where R is the effective feature recall and P is the effective feature precision. Compare the F1 score with a preset threshold; in one example, the preset threshold is 0.9. If the F1 score ≥ 0.9, the decomposition scale is deemed to meet the requirements; if the F1 score < 0.9, the number of decomposition scale levels is readjusted until the F1 score reaches the threshold. In this embodiment, by verifying the decomposition scale, the problem of effective signal loss or interference signal residue caused by improper decomposition scale can be avoided.

[0043] In some embodiments, determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of swallowing signal feature parameter extraction based on the signal decomposition accuracy verification algorithm specifically includes: extracting feature parameters of the multiple wavelet components at different scales obtained by decomposition at the adjusted or unchanged db4 wavelet basis, and calculating the completeness of feature parameter extraction of the multiple wavelet components at different scales; wherein, the F1 score is the completeness of feature parameter extraction; comparing the completeness of feature parameter extraction of the multiple wavelet components at different scales with a preset threshold for the completeness of swallowing signal feature parameters; if the completeness of feature parameter extraction is higher than the preset threshold for the completeness of swallowing signal feature parameters, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of swallowing signal feature parameters; if the completeness of feature parameter extraction is lower than the preset threshold for the completeness of swallowing signal feature parameters, then the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the completeness of feature parameter extraction reaches the preset threshold for the completeness of swallowing signal feature parameters.

[0044] In some embodiments, if the decomposition scale of the adjusted or unchanged db4 wavelet basis does not meet the preset requirement for the completeness of swallowing signal feature parameter extraction, the method further includes: parsing the P / R imbalance type corresponding to the F1 score; if it is an imbalance type with low recall R, performing a step size compression operation on the low-frequency range of 5Hz~10Hz corresponding to the low-frequency band of the swallowing signal wavelet component in the decomposition scale; for an imbalance type with low precision P, splitting the feature frequency band of the main frequency band of the swallowing signal wavelet component in the decomposition scale from 10Hz~20Hz into two secondary sub-frequency bands; for an imbalance type with both low P / R, performing the low-frequency band step size compression and main frequency band hierarchical splitting operations in conjunction.

[0045] In some embodiments, the step of calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and filtering on the multiple wavelet components at different scales, separating the effective swallowing components and the components corresponding to the interference signals, includes: calling the preset swallowing signal feature template library to extract multi-dimensional filtering feature parameters of the multiple wavelet components at different scales, wherein the multi-dimensional filtering feature parameters include amplitude peak value, frequency center, and signal duration; specifically, the backend server calls a pre-built swallowing signal feature template library, which stores the feature parameter threshold ranges corresponding to target populations of different ages and different etiologies; iterates through all wavelet components, extracts the amplitude peak value of each component, calculates the frequency center of each component through Fourier transform, and determines the signal duration of each component based on the signal start and end points.

[0046] The extracted peak amplitude, frequency center, and signal duration are compared with the corresponding feature parameter thresholds of the target population in the swallowing signal feature template library.

[0047] Specifically, differentiated thresholds are set for different population groups in the template library to achieve personalized adaptation of screening criteria. For example, for the general population, the template library presets the peak amplitude threshold as 0.3~0.8, the center frequency threshold as 8~15Hz, and the signal duration threshold as 0.5~1.2s; for people in the recovery period after stroke, the peak amplitude threshold is 0.1~0.4, the center frequency threshold is 5~10Hz, and the signal duration threshold is 0.8~1.5s. Before comparing the peak amplitude thresholds, the extracted peak amplitude values ​​need to be normalized to fall within the 0~1 range.

[0048] The first wavelet component, whose amplitude peak value, frequency center, and signal duration all fall within the corresponding characteristic parameter threshold range, is selected and determined as the effective swallowing component.

[0049] Specifically, when the peak amplitude, frequency center, and signal duration of a wavelet component all fall within the threshold range of the corresponding population in the template library, this wavelet component is identified as the first wavelet component, i.e., the effective swallowing component. By comprehensively judging through the collaborative screening of multi-dimensional parameters, compared with the single-dimensional matching judgment in the existing technology, the recognition accuracy of the effective swallowing component can be greatly improved, so that the selected effective component has complete swallowing signal characteristics.

[0050] Identify a second wavelet component whose amplitude peak value, frequency center value, and signal duration exceed the threshold range of the corresponding characteristic parameter, determine the second wavelet component as the component corresponding to the interference signal, and perform a rejection operation.

[0051] Specifically, when at least one of the three parameters of a wavelet component—peak amplitude, center frequency, and signal duration—exceeds the threshold range of the corresponding population in the template library, that wavelet component is identified as the second wavelet component, i.e., the component corresponding to the interference signal. The backend server performs a rejection operation on this type of component to ensure that the signal used for subsequent reconstruction has high purity, thus enabling high-quality restoration of the final effective swallowed signal.

[0052] See Figure 3 As shown, in some embodiments, the reverse reconstruction of the effective swallowing components to obtain the effective swallowing signal (S150) includes: S151, performing scale-level matching on the effective swallowing components to determine the db4 wavelet basis decomposition scale-level information corresponding to each effective swallowing component, wherein the decomposition scale-level information includes the decomposition level number and the frequency splitting interval of the corresponding level number.

[0053] Specifically, the backend server reads the source information for each valid swallowed component. This source information records the decomposition level number corresponding to the component during the multi-scale decomposition process, as well as the frequency splitting interval corresponding to that level. For example, the decomposition level number corresponding to a certain valid swallowed component is level 2, and its frequency splitting interval is 5~10Hz.

[0054] S152. Perform db4 wavelet inverse transform operation, determine the superposition order of effective swallowing components based on the decomposition level number, and perform frequency calibration on the corresponding level inverse transform signal components based on the corresponding frequency split interval, so that the frequency range of each signal component is consistent with the frequency split interval during decomposition, and obtain the preliminary time-domain swallowing signal.

[0055] Specifically, the inverse transform operation can be performed based on the Mallat wavelet inverse transform algorithm of the db4 wavelet basis. The stacking order of the effective swallowing components is determined according to the decomposition level number, and the components are arranged in order from the highest level to the lowest level, which is the reverse of the order in the multi-scale decomposition. For each level component, frequency calibration is performed according to the corresponding frequency splitting interval. A bandpass filter is used to filter out the frequency offset components generated during calibration, ensuring that the frequency range of each component is consistent with the frequency splitting interval during decomposition. Finally, the calibrated components of each level are stacked and fused layer by layer to obtain the preliminary time-domain swallowing signal. Because the Mallat wavelet inverse transform algorithm corresponding to the decomposition algorithm is used, a high degree of coordination between the decomposition and inverse transform processes can be achieved, thereby improving the quality of the preliminary reconstructed signal.

[0056] S153. The amplitude of the preliminary time-domain swallowing signal is smoothed and corrected using a moving average algorithm to obtain the final effective swallowing signal.

[0057] In this embodiment, a moving average algorithm with a predetermined window length can be selected. The backend server traverses each sampling point of the initial time-domain swallowing signal, calculates the average amplitude of the sampling point and the two sampling points before and after it, and uses the average value to replace the amplitude of the original sampling point to complete the smoothing correction, thereby effectively eliminating the small amplitude fluctuations generated during the inverse transformation process.

[0058] The swallowing signal separation and detection method based on wavelet transform provided in this invention involves acquiring and preprocessing mixed vibration signals using a flexible sensing patch, configuring the db4 wavelet basis as the signal multi-scale decomposition basis function, dynamically adjusting each decomposition scale based on the age of the target population and the etiology of swallowing-related complications, performing multi-scale decomposition on the preprocessed signal based on the adjusted db4 wavelet basis and decomposition scale, then calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening of wavelet components and separate effective swallowing components, and finally reconstructing the effective swallowing components in reverse. By improving the db4 wavelet basis and configuring it as a multi-scale decomposition basis function for the signal, and dynamically adjusting each decomposition scale of the db4 wavelet basis in combination with the age of the target population and the etiology of swallowing-related complications, the traditional fixed parameter setting mode is replaced. Based on the adjusted db4 wavelet basis and decomposition scale, the preprocessed signal is decomposed into multiple wavelet components of different scales. Then, through a preset swallowing signal feature template library, multi-dimensional feature parameter matching and screening are performed on the wavelet components. Multi-dimensional judgment replaces single-dimensional feature judgment, which can better adapt to the individual differences in swallowing signals of different etiological populations, thereby improving the separation accuracy of effective swallowing components and interference signals. Therefore, it is easier to detect and separate swallowing signals more accurately.

[0059] Example 2 This application also provides a swallowing signal detection device based on wavelet transform, comprising: a flexible sensing patch, attached to the skin surface area between the hyoid bone and the larynx in front of the neck of a target population, for collecting a mixed vibration signal containing effective swallowing signals and interference signals; a server configured to: preprocess the mixed vibration signal, and configure a db4 wavelet basis as the basis function for multi-scale decomposition of the signal, configure each decomposition scale as a dynamic adjustment parameter of the db4 wavelet basis, and adjust each decomposition scale based on the basic information of the target population; the basic information of the target population includes age and etiological type of swallowing-related complications; perform multi-scale decomposition of the preprocessed mixed vibration signal based on the db4 wavelet basis and the adjusted corresponding decomposition scale to obtain multiple wavelet components of different scales; call a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and screening on the multiple wavelet components of different scales, and separate the effective swallowing component and the component corresponding to the interference signal; the swallowing signal feature template library is configured with thresholds for screening feature parameters; and perform reverse reconstruction of the effective swallowing component to obtain the effective swallowing signal.

[0060] According to one embodiment of the present invention, adjusting each decomposition scale based on the basic information of the target population includes: setting an initial decomposition scale of the db4 wavelet basis, the initial decomposition scale corresponding to a split interval covering the conventional frequency range of the swallowing signal; when the target population is a stroke recovery group, adjusting the initial decomposition scale in a hierarchical increasing direction; when the target population is a Parkinson's disease group, keeping the initial decomposition scale unchanged, and covering the core frequency range of the swallowing signal of the Parkinson's disease group through the frequency split interval corresponding to the initial decomposition scale; and determining, according to a signal decomposition accuracy verification algorithm, whether the adjusted or unchanged db4 wavelet basis decomposition scale meets the preset swallowing signal feature parameter extraction completeness requirement, wherein the signal decomposition accuracy verification algorithm is an algorithm used to evaluate the wavelet component feature parameter extraction completeness.

[0061] According to one embodiment of the present invention, the step of determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction accuracy requirement based on the signal decomposition accuracy verification algorithm includes: extracting feature parameters of the multiple wavelet components of different scales obtained by decomposition at the adjusted or unchanged db4 wavelet basis decomposition scale; calculating the extraction completeness of the feature parameters of the multiple wavelet components of different scales; comparing the extraction completeness of the feature parameters of the multiple wavelet components of different scales with a preset swallowing signal feature parameter completeness threshold; if the feature parameter extraction completeness is higher than the preset swallowing signal feature parameter completeness threshold, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction completeness requirement; if the feature parameter extraction completeness is lower than the preset swallowing signal feature parameter completeness threshold, then the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the feature parameter extraction completeness reaches the preset swallowing signal feature parameter completeness threshold.

[0062] The swallowing signal detection device based on wavelet transform provided in this application embodiment is similar in implementation scheme and technical effect to the swallowing signal detection method based on wavelet transform described in the foregoing embodiment. It can also be used to execute other technical solutions in embodiment one, and can be referred to each other. It will not be described again here.

[0063] This application also provides a server. Figure 4 This is a schematic diagram of the structure of a server according to one embodiment of this application, as shown below. Figure 4As shown, the server may include: a housing 71, a processor 72, a memory 73, a circuit board 74, and a power supply circuit 75. The circuit board 74 is housed within the space enclosed by the housing 71, and the processor 72 and memory 73 are mounted on the circuit board 74. The power supply circuit 75 supplies power to the various circuits or devices of the server. The memory 73 stores executable program code. The processor 72 runs a program corresponding to the executable program code by reading the executable program code stored in the memory 73, performing the following steps: acquiring a mixed vibration signal containing valid swallowing signals and interference signals; and configuring the server to preprocess the mixed vibration signal and configure the db4 wavelet basis as the basis function for multi-scale decomposition of the signal. Each decomposition scale is configured as a dynamic adjustment parameter of the db4 wavelet basis, and each decomposition scale is adjusted based on the basic information of the target population; the basic information of the target population includes age and the etiological type of swallowing-related complications; based on the db4 wavelet basis and the adjusted corresponding decomposition scale, the preprocessed mixed vibration signal is decomposed into multiple scales to obtain multiple wavelet components of different scales; a preset swallowing signal feature template library is called to perform multi-dimensional feature parameter matching and screening on the multiple wavelet components of different scales to separate the effective swallowing components and the components corresponding to the interference signals; the swallowing signal feature template library is configured with thresholds for screening feature parameters; the effective swallowing components are reverse reconstructed to obtain the effective swallowing signal.

[0064] The specific execution process of the above steps by the processor 72, as well as the steps further executed by the processor 72 by running executable program code, can be found in the description of Embodiment 1 of this application, and will not be repeated here.

[0065] This server exists in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include: smartphones (such as iPhones), multimedia phones, feature phones, and low-end phones, etc.

[0066] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features.

[0067] (3) Other servers with data interaction functions.

[0068] This application also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the wavelet transform-based swallowing signal detection method described in any of the embodiments.

[0069] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for separating and detecting swallowing signals based on wavelet transform, characterized in that, include: A mixed vibration signal containing effective swallowing signals and interference signals is acquired by a flexible sensing patch, and the mixed vibration signal is preprocessed. The flexible sensing patch is applied to the skin surface area in front of the neck of the target population, between the hyoid bone and the throat. The db4 wavelet basis is configured as the basis function for multi-scale decomposition of the signal, and each decomposition scale is configured as a dynamic adjustment parameter of the db4 wavelet basis. Each decomposition scale is adjusted based on the basic information of the target population. The basic information of the target population includes age and etiological type of swallowing-related complications. Based on the db4 wavelet basis and the adjusted corresponding decomposition scale, the preprocessed hybrid vibration signal is decomposed into multiple scales to obtain multiple wavelet components of different scales. A preset swallowing signal feature template library is invoked to perform multi-dimensional feature parameter matching and filtering on the multiple wavelet components of different scales, separating the effective swallowing components and the components corresponding to the interference signals; the swallowing signal feature template library is configured with thresholds for filtering feature parameters; The effective swallowing component is reconstructed in reverse to obtain the effective swallowing signal.

2. The swallowing signal separation and detection method according to claim 1, characterized in that, The preprocessing of the hybrid vibration signal includes: performing analog-to-digital conversion on the acquired analog hybrid vibration signal to convert the analog signal into a digital hybrid vibration signal; A linear correction algorithm is used to correct the baseline drift of the digital hybrid vibration signal in order to eliminate drift interference in the low-frequency band. The amplitude of the corrected digital hybrid vibration signal is normalized according to the normalization algorithm, and the amplitude of the normalized digital hybrid vibration signal is uniformly mapped to a preset standard range.

3. The swallowing signal separation and detection method according to claim 1, characterized in that, The adjustment of each decomposition scale based on the basic information of the target population includes: The initial decomposition scale of the preset db4 wavelet basis is defined as a decomposition interval that covers the conventional frequency range of the swallowing signal. When the target population is people in the recovery period after stroke, the initial decomposition scale is adjusted in a hierarchical increasing direction; When the target population is Parkinson's disease patients, the initial decomposition scale is kept unchanged, and the core frequency range of the swallowing signal of Parkinson's disease patients is covered by the frequency splitting interval corresponding to the initial decomposition scale. According to the signal decomposition accuracy verification algorithm, it is determined whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of swallowing signal feature parameter extraction. The signal decomposition accuracy verification algorithm is an algorithm used to evaluate the completeness of wavelet component feature parameter extraction.

4. The swallowing signal separation and detection method according to claim 3, characterized in that, The step of determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction accuracy requirements based on the signal decomposition accuracy verification algorithm includes: Based on the decomposition scale of the adjusted or unchanged db4 wavelet basis, multiple wavelet components of different scales are obtained, feature parameters of the multiple wavelet components of different scales are extracted, and the completeness of the extraction of feature parameters of the multiple wavelet components of different scales is calculated. The completeness of feature parameter extraction of the multiple wavelet components at different scales is compared with a preset threshold for the completeness of feature parameters of the swallowing signal. If the completeness of the extracted feature parameters is higher than the preset threshold for the completeness of the swallowing signal feature parameters, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of the extracted swallowing signal feature parameters. If the completeness of the extracted feature parameters is lower than the preset threshold for the completeness of swallowing signal feature parameters, the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the completeness of the extracted feature parameters reaches the preset threshold for the completeness of swallowing signal feature parameters.

5. The swallowing signal separation and detection method according to claim 1, characterized in that, The process involves calling a preset swallowing signal feature template library to perform multi-dimensional feature parameter matching and filtering on the multiple wavelet components at different scales, separating the effective swallowing components and the components corresponding to the interference signals, including: The preset swallowing signal feature template library is invoked to extract multi-dimensional filtering feature parameters of the multiple wavelet components at different scales. The multi-dimensional filtering feature parameters include amplitude peak value, frequency center and signal duration. The extracted peak amplitude, frequency center, and signal duration are compared with the corresponding feature parameter thresholds of the target population in the swallowing signal feature template library. The first wavelet component, whose peak amplitude, frequency center, and signal duration all fall within the threshold range of the corresponding characteristic parameters, is selected and determined as the effective swallowing component. Identify a second wavelet component whose amplitude peak value, frequency center value, and signal duration exceed the threshold range of the corresponding characteristic parameter, determine the second wavelet component as the component corresponding to the interference signal, and perform a rejection operation.

6. The swallowing signal separation and detection method according to claim 5, characterized in that, The reverse reconstruction of the effective swallowing component to obtain the effective swallowing signal includes: Scale-level matching is performed on the effective swallowing components to determine the db4 wavelet basis decomposition scale-level information corresponding to each effective swallowing component. The decomposition scale-level information includes the decomposition level number and the frequency splitting interval of the corresponding level number. Perform db4 wavelet inverse transform operation, determine the superposition order of effective swallowing components based on the decomposition level number, and perform frequency calibration on the corresponding level inverse transform signal components based on the corresponding frequency split interval, so that the frequency range of each signal component is consistent with the frequency split interval during decomposition, and obtain the preliminary time-domain swallowing signal. The amplitude of the initial time-domain swallowing signal is smoothed and corrected using a moving average algorithm to obtain the final effective swallowing signal.

7. A swallowing signal detection device based on wavelet transform, characterized in that, include: A flexible sensing patch is attached to the skin surface area in front of the neck of the target population, between the hyoid bone and the larynx, to collect a mixed vibration signal containing effective swallowing signals and interference signals. The server is configured as follows: The mixed vibration signal is preprocessed, and the db4 wavelet basis is configured as the basis function for multi-scale decomposition of the signal. Each decomposition scale is configured as a dynamic adjustment parameter of the db4 wavelet basis, and each decomposition scale is adjusted based on the basic information of the target population. The basic information of the target population includes age and etiological type of swallowing-related complications. Based on the db4 wavelet basis and the adjusted corresponding decomposition scale, the preprocessed hybrid vibration signal is decomposed into multiple scales to obtain multiple wavelet components of different scales. A preset swallowing signal feature template library is invoked to perform multi-dimensional feature parameter matching and filtering on the multiple wavelet components of different scales, separating the effective swallowing components and the components corresponding to the interference signals; the swallowing signal feature template library is configured with thresholds for filtering feature parameters; The effective swallowing component is reconstructed in reverse to obtain the effective swallowing signal.

8. The swallowing signal detection device according to claim 7, characterized in that, The adjustment of each decomposition scale based on the basic information of the target population includes: The initial decomposition scale of the preset db4 wavelet basis is defined as a decomposition interval that covers the conventional frequency range of the swallowing signal. When the target population is people in the recovery period after stroke, the initial decomposition scale is adjusted in a hierarchical increasing direction; When the target population is Parkinson's disease patients, the initial decomposition scale is kept unchanged, and the core frequency range of the swallowing signal of Parkinson's disease patients is covered by the frequency splitting interval corresponding to the initial decomposition scale. According to the signal decomposition accuracy verification algorithm, it is determined whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of swallowing signal feature parameter extraction. The signal decomposition accuracy verification algorithm is an algorithm used to evaluate the completeness of wavelet component feature parameter extraction.

9. The swallowing signal detection device according to claim 8, characterized in that, The step of determining whether the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset swallowing signal feature parameter extraction accuracy requirements based on the signal decomposition accuracy verification algorithm includes: Based on the decomposition scale of the adjusted or unchanged db4 wavelet basis, multiple wavelet components of different scales are obtained, feature parameters of the multiple wavelet components of different scales are extracted, and the completeness of the extraction of feature parameters of the multiple wavelet components of different scales is calculated. The completeness of feature parameter extraction of the multiple wavelet components at different scales is compared with a preset threshold for the completeness of feature parameters of the swallowing signal. If the completeness of the extracted feature parameters is higher than the preset threshold for the completeness of the swallowing signal feature parameters, then it is determined that the decomposition scale of the adjusted or unchanged db4 wavelet basis meets the preset requirement for the completeness of the extracted swallowing signal feature parameters. If the completeness of the extracted feature parameters is lower than the preset threshold for the completeness of swallowing signal feature parameters, the number of levels of the decomposition scale of the db4 wavelet basis is readjusted until the completeness of the extracted feature parameters reaches the preset threshold for the completeness of swallowing signal feature parameters.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the wavelet transform-based swallowing signal separation and detection method according to any one of claims 1 to 6.

Images