A borborygmus signal feature recognition model construction method and device and a storage medium

Abstract

The present application relates to a kind of borborygmi signal feature recognition model construction method, device and storage medium, applied to borborygmi signal technical field, comprising: by Chebyshev polynomial, the acoustic characteristics of borborygmi signal is approximated fitting, effectively capture the detail features of signal, reduce the influence of noise on spectrum fitting, on this basis, by combining trained physiological sound recognition model and trained sound recognition model, use the similarity of physiological sound, i.e. physiological sound pre-training model and general sound pre-training model, can capture and understand borborygmi signal from different angles, improve the robustness of recognition, the borborygmi recognition model finally obtained has good generalization performance and strong robust performance when processing borborygmi signal with different characteristics and diversity, while it can effectively reduce the interference of noise, improve the accuracy of recognition.

Description

Technical Field

[0001] This invention relates to the field of bowel sound signal technology, specifically to a method, apparatus, and storage medium for constructing a bowel sound signal feature recognition model. Background Technology

[0002] Bowel sounds are physiological sounds that reflect the motility of the small intestine and have important clinical value. They can be used to monitor intestinal diseases. In order to continuously and for a long time monitor bowel sounds, it is necessary to extract and identify the bowel sound signals and calculate their related features.

[0003] However, bowel sound signals have complex characteristics, including frequency diversity and large amplitude variations. During the acquisition process, they are usually subject to various noise interferences from the surrounding environment, such as abdominal muscle activity and breathing. These noises cause problems for bowel sound recognition and pose challenges to existing recognition schemes.

[0004] Existing methods for recognizing bowel sounds include nonlinear dynamics analysis, time-frequency analysis, and machine learning. However, these methods have certain limitations in terms of the accuracy of bowel sound signal recognition, adaptability to environmental noise, and diversity. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, device and storage medium for constructing a bowel sound signal feature recognition model, so as to solve the problem that the recognition results are not accurate enough in the prior art because the bowel sound recognition method does not take into account various noise interferences from the surrounding environment, such as abdominal muscle activity and breathing.

[0006] According to a first aspect of the present invention, a method for constructing a bowel sound signal feature recognition model is provided, the method comprising:

[0007] The patient's original bowel sound signal is acquired, and the original bowel sound signal is preprocessed to obtain multiple bowel sound signals;

[0008] The multiple bowel sound signals are subjected to frame-segmentation and windowing operations, and each bowel sound signal is divided into bowel sound signals with the same number of frames.

[0009] The spectral representation of each frame of bowel sound signal in each segment of bowel sound signal is obtained by short-time Fourier transform;

[0010] The logarithmic spectrum of each frame of bowel sound signal is obtained based on the spectral representation of each frame of bowel sound signal;

[0011] Set the degree of the Chebyshev polynomial, and obtain the polynomial expression and weighting coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal.

[0012] The fitted spectrum of each frame of bowel sound signal is obtained based on the polynomial expression of the Chebyshev polynomial and the weighting coefficients.

[0013] The fitted spectrum of each frame of bowel sound signal in each segment of bowel sound signal is arranged in chronological order to obtain the logarithmic spectrum of each bowel sound signal segment.

[0014] A feature recognition model architecture for bowel sound signals is constructed. The logarithmic spectrum of each bowel sound signal segment is used as the input of the feature recognition model architecture, and whether the input signal is a bowel sound signal is used as the output of the feature recognition model architecture. The feature recognition model architecture is trained until it converges, thus obtaining the feature recognition model for bowel sound signals.

[0015] Preferably,

[0016] The preprocessing operation on the original bowel sound signal to obtain multiple bowel sound signals includes:

[0017] The original bowel sound signal is segmented into T-second segments, and each segment of the original bowel sound signal after T-second segmentation is filtered to obtain multiple bowel sound signals.

[0018] Preferably,

[0019] The step of obtaining the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal includes:

[0020] The power spectrum of each bowel sound signal is obtained by calculating the square of the modulus of the spectrum of each bowel sound signal based on the spectral representation of each frame of bowel sound signal.

[0021] The natural logarithm of the power spectrum of each frame of bowel sound signal is taken to obtain the logarithmic spectrum of each frame of bowel sound signal.

[0022] Preferably,

[0023] The bowel sound signal feature recognition model architecture includes a feature extraction module and a feature classification module;

[0024] The feature extraction module adopts a dual-branch structure. One branch uses the physiological sound feature extraction module of the pre-trained physiological sound recognition model, and the other branch uses the sound feature extraction module of the pre-trained sound recognition model.

[0025] The feature classification module uses a fully connected layer.

[0026] Preferably,

[0027] The physiological sound feature extraction module is used to extract the physiological sound feature vector from the logarithmic spectrum of each segment of the input bowel sound signal;

[0028] The sound feature extraction module is used to extract the sound feature vector from the logarithmic spectrum of each input bowel sound signal;

[0029] The physiological sound feature vector is converted into a one-dimensional physiological sound feature vector, and the sound feature vector is converted into a one-dimensional sound feature vector.

[0030] The one-dimensional physiological sound feature vector and the one-dimensional sound feature vector are concatenated to obtain a one-dimensional target feature vector, which is then input into the fully connected layer.

[0031] Preferably,

[0032] The training of the bowel sound signal feature recognition model architecture includes:

[0033] The parameters of the physiological sound feature extraction module of the pre-trained physiological sound recognition model and the sound feature extraction module of the pre-trained sound recognition model are used as the initial weights of the feature extraction module of the bowel sound signal feature recognition model architecture and loaded. All parameters of the physiological sound feature extraction module and the sound feature extraction module remain unchanged during the training process. The fully connected layer of the bowel sound signal feature recognition model architecture is trained by the logarithmic spectrogram of each bowel sound signal.

[0034] According to a second aspect of the present invention, an apparatus for constructing a bowel sound signal feature recognition model is provided, the apparatus comprising:

[0035] Data acquisition module: used to acquire the patient's original bowel sound signal, and to preprocess the original bowel sound signal to obtain multiple bowel sound signals;

[0036] Framing module: used to perform framing and windowing operations on the multiple segments of bowel sound signals, dividing each segment of bowel sound signal into bowel sound signals with the same number of frames;

[0037] Spectrum representation acquisition module: used to obtain the spectrum representation of each frame of bowel sound signal from each segment of bowel sound signal through short-time Fourier transform;

[0038] Logarithmic spectrum acquisition module: used to acquire the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal;

[0039] Chebyshev polynomial acquisition module: used to set the degree of the Chebyshev polynomial, and to obtain the polynomial expression and weight coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal;

[0040] Fitted spectrum acquisition module: used to obtain the fitted spectrum of each frame of bowel sound signal according to the polynomial expression of the Chebyshev polynomial and the weighting coefficients;

[0041] Logarithmic spectrum acquisition module: used to arrange the fitted spectrum of each frame of bowel sound signal in each bowel sound signal in chronological order to obtain the logarithmic spectrum of each bowel sound signal;

[0042] Model acquisition module: used to build a bowel sound signal feature recognition model architecture. The logarithmic spectrum of each bowel sound signal segment is used as the input of the bowel sound signal feature recognition model architecture, and whether the input signal is a bowel sound signal is used as the output of the bowel sound signal feature recognition model architecture. The bowel sound signal feature recognition model architecture is trained until it converges, thus obtaining the bowel sound signal feature recognition model.

[0043] According to a third aspect of the present invention, a storage medium is provided, the storage medium storing a computer program, which, when executed by a host controller, implements the steps of the above-described method.

[0044] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0045] This application approximates the acoustic features of bowel sound signals using Chebyshev polynomials, effectively capturing the detailed features of the signal and reducing the impact of noise on the spectral fitting. Based on this, by combining a pre-trained physiological sound recognition model and a pre-trained sound recognition model, the similarity of physiological sounds is utilized. That is, the physiological sound pre-training model and the general sound pre-training model can capture and understand bowel sound signals from different perspectives, improving the robustness of recognition. The final bowel sound recognition model has good generalization performance and strong robustness when processing bowel sound signals with different features and diversity, while effectively reducing noise interference and improving recognition accuracy.

[0046] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0048] Figure 1 This is a flowchart illustrating a method for constructing a bowel sound signal feature recognition model according to an exemplary embodiment;

[0049] Figure 2 This is a system schematic diagram of an apparatus for constructing a feature recognition model of bowel sounds according to another exemplary embodiment;

[0050] In the attached diagram: 1-Data acquisition module, 2-Frame segmentation module, 3-Spectrum representation acquisition module, 4-Logarithmic spectrum acquisition module, 5-Chebyshev polynomial acquisition module, 6-Fitted spectrum acquisition module, 7-Logarithmic spectrum acquisition module, 8-Model acquisition module. Detailed Implementation

[0051] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0052] Example 1

[0053] Figure 1 This is a flowchart illustrating a method for constructing a bowel sound signal feature recognition model according to an exemplary embodiment, the method comprising:

[0054] S1, acquire the patient's original bowel sound signal, and perform preprocessing operation on the original bowel sound signal to obtain multiple bowel sound signals;

[0055] S2, perform frame-segmentation and windowing operations on the multiple bowel sound signals respectively, dividing each bowel sound signal into bowel sound signals with the same number of frames;

[0056] S3, the spectral representation of each frame of bowel sound signal in each segment of bowel sound signal is obtained by short-time Fourier transform;

[0057] S4, obtain the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal;

[0058] S5, set the degree of the Chebyshev polynomial, and obtain the polynomial expression and weight coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal;

[0059] S6. The fitted spectrum of each frame of bowel sound signal is obtained according to the polynomial expression of the Chebyshev polynomial and the weighting coefficients.

[0060] S7. Arrange the fitted spectrum of each frame of bowel sound signal in each segment of bowel sound signal in chronological order to obtain the logarithmic spectrum of each segment of bowel sound signal.

[0061] S8. Construct a bowel sound signal feature recognition model architecture. Use the logarithmic spectrum of each bowel sound signal as the input of the bowel sound signal feature recognition model architecture, and use whether the input signal is a bowel sound signal as the output of the bowel sound signal feature recognition model architecture. Train the bowel sound signal feature recognition model architecture until the bowel sound signal feature recognition model architecture converges to obtain the bowel sound signal feature recognition model.

[0062] Understandably, a digital stethoscope is used to collect bowel sound signals from a patient, resulting in a raw bowel sound signal x(n). This raw bowel sound signal x(n) is then subjected to a T-second segmented filtering operation, and each segment of the preprocessed bowel sound signal X is processed. i (n) Perform frame-by-frame windowing operation to obtain the bowel sound signal X for each frame. ij (n), where j represents the j-th frame of the i-th bowel sound signal. A Hamming window can be used as the window function. It should be noted that each bowel sound signal X... i (n) The number of frames j obtained after framing and windowing is the same for each frame of bowel sound signal X. ij (n) Apply the short-time Fourier transform to calculate and obtain the spectral representation S of each frame of bowel sound signal. ij (n), obtain the logarithmic spectrum M(f) of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal, where f represents the frequency. Set the degree N of the Chebyshev polynomial, and calculate the weighting coefficient a of the Chebyshev polynomial for the logarithmic spectrum M(f) of each frame of bowel sound signal. k For k∈(0, N-1), we obtain the polynomial expression of the Chebyshev polynomial, as shown below:

[0063] T k (x)=cos(k cos -1 (x))

[0064] In the formula, x represents the normalized frequency, which is usually 1. Where f min 50Hz, f max For 3000Hz, we get T k (x) represents the k-th order polynomial of the Chebyshev polynomial. When k∈(0,N-1), a weight coefficient a is calculated for each k. k As shown below:

[0065]

[0066] Use the calculated weighting coefficient a k and Chebyshev polynomial T k (x), the approximate fitting spectrum curve C(f) is calculated, as shown below:

[0067]

[0068] It represents the spectral curve of the Chebyshev polynomial fitting, which represents the approximation of the logarithmic spectrum of each frame of bowel sound signal. Arranging the fitted spectrum of each frame of bowel sound signal in chronological order yields the logarithmic spectrum Z of the Chebyshev polynomial approximation fit for each segment of the bowel sound signal. i , used for training the following bowel sound recognition model;

[0069] This application approximates the acoustic features of bowel sound signals using Chebyshev polynomials, effectively capturing the detailed features of the signal and reducing the impact of noise on the spectral fitting. Based on this, by combining a pre-trained physiological sound recognition model and a pre-trained sound recognition model, it utilizes the similarity of physiological sounds and multimodal information, namely, a physiological sound pre-training model and a general sound pre-training model, to capture and understand bowel sound signals from different perspectives, improving the robustness of recognition. The resulting bowel sound recognition model has good generalization performance and strong robustness when processing bowel sound signals with different characteristics and diversity, while effectively reducing noise interference and improving recognition accuracy.

[0070] Preferably,

[0071] The preprocessing operation on the original bowel sound signal to obtain multiple bowel sound signals includes:

[0072] The original bowel sound signal is segmented into T-second segments, and each segment of the original bowel sound signal after T-second segmentation is filtered to obtain multiple bowel sound signals.

[0073] Understandably, the collected raw bowel sound signal x(n) is segmented into T-second segments, and then each segment of the raw bowel sound signal x is processed. i (n) Perform filtering operations, where the preprocessing process includes segmentation and filtering operations, with a filtering range of 50Hz-3000Hz. After the above preprocessing, the preprocessed bowel sound signal X for each segment is obtained. i (n), where i is the number of segments in the original bowel sound signal.

[0074] Preferably,

[0075] The step of obtaining the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal includes:

[0076] The power spectrum of each bowel sound signal is obtained by calculating the square of the modulus of the spectrum of each bowel sound signal based on the spectral representation of each frame of bowel sound signal.

[0077] The natural logarithm of the power spectrum of each frame of bowel sound signal is taken to obtain the logarithmic spectrum of each frame of bowel sound signal;

[0078] The spectrum representation of each frame of bowel sound signal obtained from the above calculation is S. ij (n), calculate the square of the modulus of the spectrum of each frame of bowel sound signal to obtain the power spectrum of each frame of bowel sound signal, and take the natural logarithm of the power spectrum of each frame of bowel sound signal to finally obtain the logarithmic spectrum M(f) of each frame of bowel sound signal, where f represents the frequency. The above-obtained logarithmic spectrum M(f) is the acoustic feature of the extracted bowel sound signal.

[0079] Preferably,

[0080] The bowel sound signal feature recognition model architecture includes a feature extraction module and a feature classification module;

[0081] The feature extraction module adopts a dual-branch structure. One branch uses the physiological sound feature extraction module of the pre-trained physiological sound recognition model, and the other branch uses the sound feature extraction module of the pre-trained sound recognition model.

[0082] The feature classification module uses a fully connected layer;

[0083] Understandably, the constructed bowel sound recognition model includes a feature extraction block and a feature classification block. The feature extraction block of the bowel sound recognition model adopts a dual-branch structure, with one branch being the feature extraction block of the trained physiological sound recognition model and the other branch being the feature extraction block of the trained sound recognition model. The feature classification block is used to classify the acoustic features of the input bowel sound signal. The feature classification block includes a fully connected layer for the binary classification task of bowel sound recognition. The binary classification task is to identify whether the input signal is a bowel sound signal based on the collected signal.

[0084] Preferably,

[0085] The physiological sound feature extraction module is used to extract the physiological sound feature vector from the logarithmic spectrum of each segment of the input bowel sound signal;

[0086] The sound feature extraction module is used to extract the sound feature vector from the logarithmic spectrum of each input bowel sound signal;

[0087] The physiological sound feature vector is converted into a one-dimensional physiological sound feature vector, and the sound feature vector is converted into a one-dimensional sound feature vector.

[0088] The one-dimensional physiological sound feature vector and the one-dimensional sound feature vector are concatenated to obtain a one-dimensional target feature vector, and the one-dimensional target feature vector is input into the fully connected layer.

[0089] Understandably, converting the feature vectors output from the two branches into one-dimensional feature vectors and concatenating them to form a one-dimensional feature vector, which is then input into the reconstructed feature classification block, achieves the fusion of different feature information extracted from the two branches.

[0090] Preferably,

[0091] The training of the bowel sound signal feature recognition model architecture includes:

[0092] The parameters of the physiological sound feature extraction module of the pre-trained physiological sound recognition model and the sound feature extraction module of the pre-trained sound recognition model are used as the initial weights of the feature extraction module of the bowel sound signal feature recognition model architecture and loaded. All parameters of the physiological sound feature extraction module and the sound feature extraction module remain unchanged during the training process. The fully connected layer of the bowel sound signal feature recognition model architecture is trained by the logarithmic spectrogram of each bowel sound signal.

[0093] Understandably, the parameters of the trained physiological sound recognition model and the feature extraction blocks in the trained sound recognition model are used as the initial weights of the bowel sound recognition model and loaded. This preserves the learned feature representations (i.e., extracted feature information) in the feature extraction blocks of the physiological sound recognition model and the sound recognition model. All parameters in the two feature extraction blocks are frozen, meaning the parameters of the feature extraction blocks are set to not be updated. The parameters of the feature classification blocks in the bowel sound recognition model are randomly initialized. The logarithmic spectrum Z of each bowel sound signal is approximated by a Chebyshev polynomial. i As training data for the bowel sound recognition model, the pre-constructed bowel sound recognition model is trained on the training data to optimize its parameters. The training process includes forward propagation, backpropagation, and parameter update. Forward propagation involves passing the training data to the pre-constructed bowel sound recognition model, which then outputs the probability of "a bowel sound signal." Backpropagation calculates the gradient of the bowel sound recognition model during training (indicating how to adjust the parameters), and uses this gradient to update the parameters in the model to reduce prediction errors. Parameter update refers to adjusting and optimizing the training process of the bowel sound recognition model based on the gradient information obtained from backpropagation. The above steps are repeated until the bowel sound recognition model converges, resulting in a well-trained bowel sound recognition model.

[0094] Example 2:

[0095] Figure 2 This is a system schematic diagram of a bowel sound signal feature recognition model construction device according to another exemplary embodiment, the device comprising:

[0096] Data acquisition module 1: used to acquire the patient's original bowel sound signal, and to preprocess the original bowel sound signal to obtain multiple bowel sound signals;

[0097] Frame segmentation module 2: used to perform frame segmentation and windowing operations on the multiple segments of bowel sound signals, dividing each segment of bowel sound signal into bowel sound signals with the same number of frames;

[0098] Spectrum representation acquisition module 3: used to obtain the spectrum representation of each frame of bowel sound signal by short-time Fourier transform of each frame of bowel sound signal in each segment of bowel sound signal;

[0099] Logarithmic spectrum acquisition module 4: used to acquire the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal;

[0100] Chebyshev polynomial acquisition module 5: used to set the degree of the Chebyshev polynomial, and to obtain the polynomial expression and weight coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal;

[0101] Fitted spectrum acquisition module 6: used to obtain the fitted spectrum of each frame of bowel sound signal according to the polynomial expression of the Chebyshev polynomial and the weighting coefficients;

[0102] Logarithmic spectrum acquisition module 7: used to arrange the fitted spectrum of each frame of bowel sound signal in each bowel sound signal in chronological order to obtain the logarithmic spectrum of each bowel sound signal;

[0103] Model acquisition module 8: used to build a bowel sound signal feature recognition model architecture, taking the logarithmic spectrum of each bowel sound signal as the input of the bowel sound signal feature recognition model architecture, and taking whether the input signal is a bowel sound signal as the output of the bowel sound signal feature recognition model architecture, training the bowel sound signal feature recognition model architecture until the bowel sound signal feature recognition model architecture converges, and obtaining the bowel sound signal feature recognition model;

[0104] Understandably, the data acquisition module 1 acquires the patient's original bowel sound signal, performs preprocessing on the original bowel sound signal to obtain multiple bowel sound signal segments; the framing module 2 performs framing and windowing operations on each of the multiple bowel sound signal segments, dividing each bowel sound signal segment into bowel sound signals of the same number of frames; the spectrum representation acquisition module 3 performs short-time Fourier transform on each frame of bowel sound signal in each bowel sound signal segment to obtain the spectrum representation of each frame of bowel sound signal; and the logarithmic spectrum acquisition module 4 obtains the bowel sound signal of each frame based on the spectrum representation of each frame of bowel sound signal. The logarithmic spectrum of the signal; the Chebyshev polynomial acquisition module 5 is used to set the degree of the Chebyshev polynomial, and obtain the polynomial expression and weighting coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal; the fitted spectrum acquisition module 6 is used to obtain the fitted spectrum of each frame of bowel sound signal based on the polynomial expression and weighting coefficients of the Chebyshev polynomial; the logarithmic spectrum acquisition module 7 is used to arrange the fitted spectrum of each frame of bowel sound signal in each segment of bowel sound signal in chronological order to obtain the logarithmic spectrum of each segment of bowel sound signal; through The model acquisition module 8 is used to build a feature recognition model architecture for bowel sound signals. It uses the logarithmic spectrum of each bowel sound signal as input to the model architecture and whether the input signal is a bowel sound signal as output. The model architecture is trained until it converges, thus obtaining the bowel sound signal feature recognition model. Compared to other deep learning methods, this invention first uses Chebyshev polynomials to approximate and fit the acoustic features of the bowel sound signal, which can effectively... By capturing the detailed features of the signal and reducing the impact of noise on the spectrum fitting, the Chebyshev polynomial property can highlight the frequency features. However, the frequency features of the bowel sound signal contain information about the health status of the intestines. By combining the trained physiological sound recognition model and the trained sound recognition model, the similarity of physiological sounds and multimodal information are utilized, namely the physiological sound pre-training model and the general sound pre-training model. This allows for the capture and understanding of bowel sound signals from different perspectives, improving the robustness of recognition. The two models are trained and fine-tuned to adapt to the specific recognition task of bowel sound signals.

[0105] Example 3:

[0106] This embodiment provides a storage medium storing a computer program, which, when executed by a host controller, implements the various steps in the above method.

[0107] It is understood that the storage medium mentioned above can be a read-only memory, a hard disk, or an optical disk, etc.

[0108] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0109] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0110] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0111] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0112] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0113] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0114] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0115] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0116] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for constructing a feature recognition model for bowel sound signals, characterized in that, The method includes: The patient's original bowel sound signal is acquired, and the original bowel sound signal is preprocessed to obtain multiple bowel sound signals; The multiple bowel sound signals are subjected to frame-segmentation and windowing operations, and each bowel sound signal is divided into bowel sound signals with the same number of frames. The spectral representation of each frame of bowel sound signal in each segment of bowel sound signal is obtained by short-time Fourier transform; The logarithmic spectrum of each frame of bowel sound signal is obtained based on the spectral representation of each frame of bowel sound signal; Set the degree of the Chebyshev polynomial, and obtain the polynomial expression and weighting coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal. The fitted spectrum of each frame of bowel sound signal is obtained based on the polynomial expression of the Chebyshev polynomial and the weighting coefficients. The fitted spectrum of each frame of bowel sound signal in each segment of bowel sound signal is arranged in chronological order to obtain the logarithmic spectrum of each bowel sound signal segment. A feature recognition model architecture for bowel sound signals is constructed. The logarithmic spectrum of each bowel sound signal segment is used as the input of the feature recognition model architecture, and whether the input signal is a bowel sound signal is used as the output of the feature recognition model architecture. The feature recognition model architecture is trained until it converges, thus obtaining the feature recognition model for bowel sound signals.

2. The method according to claim 1, characterized in that, The preprocessing operation on the original bowel sound signal to obtain multiple bowel sound signals includes: The original bowel sound signal is segmented into T-second segments, and each segment of the original bowel sound signal after T-second segmentation is filtered to obtain multiple bowel sound signals.

3. The method according to claim 2, characterized in that, The step of obtaining the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal includes: The power spectrum of each bowel sound signal is obtained by calculating the square of the modulus of the spectrum of each bowel sound signal based on the spectral representation of each frame of bowel sound signal. The natural logarithm of the power spectrum of each frame of bowel sound signal is taken to obtain the logarithmic spectrum of each frame of bowel sound signal.

4. The method according to claim 1, characterized in that, The bowel sound signal feature recognition model architecture includes a feature extraction module and a feature classification module; The feature extraction module adopts a dual-branch structure. One branch uses the physiological sound feature extraction module of the pre-trained physiological sound recognition model, and the other branch uses the sound feature extraction module of the pre-trained sound recognition model. The feature classification module uses a fully connected layer.

5. The method according to claim 4, characterized in that, The physiological sound feature extraction module is used to extract the physiological sound feature vector from the logarithmic spectrum of each segment of the input bowel sound signal; The sound feature extraction module is used to extract the sound feature vector from the logarithmic spectrum of each input bowel sound signal; The physiological sound feature vector is converted into a one-dimensional physiological sound feature vector, and the sound feature vector is converted into a one-dimensional sound feature vector. The one-dimensional physiological sound feature vector and the one-dimensional sound feature vector are concatenated to obtain a one-dimensional target feature vector, which is then input into the fully connected layer.

6. The method according to claim 5, characterized in that, The training of the bowel sound signal feature recognition model architecture includes: The parameters of the physiological sound feature extraction module of the pre-trained physiological sound recognition model and the sound feature extraction module of the pre-trained sound recognition model are used as the initial weights of the feature extraction module of the bowel sound signal feature recognition model architecture and loaded. All parameters of the physiological sound feature extraction module and the sound feature extraction module remain unchanged during the training process. The fully connected layer of the bowel sound signal feature recognition model architecture is trained by the logarithmic spectrogram of each bowel sound signal.

7. A device for constructing a feature recognition model for bowel sound signals, characterized in that, The device includes: Data acquisition module: used to acquire the patient's original bowel sound signal, and to preprocess the original bowel sound signal to obtain multiple bowel sound signals; Framing module: used to perform framing and windowing operations on the multiple segments of bowel sound signals, dividing each segment of bowel sound signal into bowel sound signals with the same number of frames; Spectrum representation acquisition module: used to obtain the spectrum representation of each frame of bowel sound signal from each segment of bowel sound signal through short-time Fourier transform; Logarithmic spectrum acquisition module: used to acquire the logarithmic spectrum of each frame of bowel sound signal based on the spectral representation of each frame of bowel sound signal; Chebyshev polynomial acquisition module: used to set the degree of the Chebyshev polynomial, and to obtain the polynomial expression and weight coefficients of the Chebyshev polynomial based on the degree of the Chebyshev polynomial and the logarithmic spectrum of each frame of bowel sound signal; Fitted spectrum acquisition module: used to obtain the fitted spectrum of each frame of bowel sound signal according to the polynomial expression of the Chebyshev polynomial and the weighting coefficients; Logarithmic spectrum acquisition module: used to arrange the fitted spectrum of each frame of bowel sound signal in each bowel sound signal in chronological order to obtain the logarithmic spectrum of each bowel sound signal; Model acquisition module: used to build a bowel sound signal feature recognition model architecture. The logarithmic spectrum of each bowel sound signal segment is used as the input of the bowel sound signal feature recognition model architecture, and whether the input signal is a bowel sound signal is used as the output of the bowel sound signal feature recognition model architecture. The bowel sound signal feature recognition model architecture is trained until it converges, thus obtaining the bowel sound signal feature recognition model.

8. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by the main controller, implements each step of the method for constructing a bowel sound signal feature recognition model as described in any one of claims 1-6.

Images