Method and system for extracting snore characteristic parameters

Abstract

The invention provides a method and system for extracting snore characteristic parameters. The method comprises the following steps that a snore signal of a patient is obtained through a sound collecting device; a denoising module is adopted to carry out preprocessing on the collected signals, wherein the preprocessing comprises denoising processing and signal normalization; framing the preprocessed signal by using a feature extraction module, and extracting time domain, frequency domain, cepstrum feature and nonlinear feature parameters of the signal; and a feature selection module is adopted to screen out representative and mutually independent features from the extracted features through correlation analysis and principal component analysis, and a machine learning module is adopted to train the selected features so as to realize snore recognition and classification. According to the method, the frequency domain features are extracted, and the time domain and cepstrum features are integrated, so that the physiological and pathological characteristics of the snore signals can be reflected more comprehensively, and the integrity and accuracy of feature extraction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of snoring processing, and in particular to a method and system for extracting snoring characteristic parameters. Background Art

[0002] Snoring is one of the common symptoms of sleep-disordered breathing and is often closely related to diseases such as sleep apnea syndrome (SAS). Snoring is caused by partial obstruction of the airway during sleep, which causes vibrations when the airflow passes through. Accurately identifying and analyzing snoring characteristic parameters has important clinical value for the diagnosis and treatment of sleep-disordered breathing.

[0003] With the development of signal processing and pattern recognition technology, the extraction of snoring feature parameters has become a research hotspot. However, the existing snoring feature extraction methods have certain limitations, mainly including:

[0004] The feature extraction method is single and cannot fully reflect the complex characteristics of snoring. It is sensitive to noise interference and has low recognition accuracy. It has poor real-time performance and cannot meet the needs of clinical monitoring.

[0005] In the prior art, a method for extracting snoring features using spectrum analysis combined with a machine learning algorithm is disclosed. The following is a detailed introduction to the solution:

[0006] Spectrum analysis: This solution first collects the patient's snoring signal through a microphone, and then uses Fourier transform to convert the time domain signal into a frequency domain signal to obtain a spectrogram. The spectrogram can display the frequency distribution of the signal and provide basic data for subsequent feature extraction.

[0007] Feature extraction: Based on the spectrogram, a series of characteristic parameters are extracted, such as spectrum entropy, spectrum center, spectrum bandwidth, etc. These characteristic parameters can reflect the energy distribution, frequency characteristics and other information of the snoring signal.

[0008] Machine learning algorithm: The extracted feature parameters are input into machine learning models, such as support vector machines (SVM), neural networks (NN), etc., for training and classification. The trained model can identify the type and severity of snoring based on the feature parameters.

[0009] Based on an in-depth analysis of existing snoring feature parameter extraction technologies, the following main shortcomings are found:

[0010] Incomplete feature extraction: Current feature extraction methods mostly focus on frequency domain analysis, ignoring other key features such as time domain and cepstrum, which limits the mining of deep information of snoring signals.

[0011] Susceptible to noise interference: In actual situations, the interference of environmental noise on snoring signals is a common problem. The existing technology is not capable of processing this problem, resulting in a decrease in recognition accuracy.

[0012] Limited applicability: The applicability of existing technologies is limited to specific populations and sleep states, and is not adaptable enough to different individuals.

[0013] Equipment dependence: Some technologies require professional recording and signal processing equipment, which increases the cost of use and limits their popularization and promotion. Summary of the invention

[0014] In view of the deficiencies in the prior art, the present invention provides a method and system for extracting characteristic parameters of snoring, which more comprehensively reflects the physiological and pathological characteristics of snoring signals by extracting frequency domain features and integrating time domain and cepstrum features, thereby improving the integrity and accuracy of feature extraction.

[0015] In order to achieve the above object, the present invention adopts the following technical solutions:

[0016] A method for extracting characteristic parameters of snoring includes the following steps:

[0017] Collecting snoring signals, and obtaining snoring signals of patients through a sound collection device;

[0018] Preprocess the collected signals, including denoising and signal normalization;

[0019] Divide the preprocessed signal into frames, and extract the signal's time domain, frequency domain, cepstrum features and nonlinear characteristic parameters;

[0020] Representative and independent features are selected from the extracted features through correlation analysis and principal component analysis;

[0021] The selected features are input into the machine learning model for training to realize the recognition and classification of snoring.

[0022] Preferably, the denoising process uses wavelet transform or short-time Fourier transform.

[0023] Preferably, the characteristic parameters include time domain characteristics: energy, mean, variance, kurtosis, skewness; frequency domain characteristics: spectral entropy, spectral center, spectral bandwidth; cepstrum characteristics: Mel frequency cepstrum coefficients; nonlinear characteristics: Lyapunov exponent, correlation dimension.

[0024] Preferably, the machine learning model is a support vector machine, a random forest or a neural network.

[0025] Preferably, the feature screening method comprises the following steps:

[0026] Correlation analysis: used to evaluate the correlation between feature parameters and reduce redundant features;

[0027] Principal Component Analysis: used to reduce dimensionality and select the features with the most information.

[0028] Preferably, the machine learning model includes the identification and classification of characteristics of different populations and sleep states when being trained.

[0029] Preferably, the identification and classification of characteristics of different groups of people and sleep states includes:

[0030] The machine learning model, support vector machine (SVM) classification, was adopted.

[0031] A system for extracting characteristic parameters of snoring, comprising:

[0032] A sound collection device for collecting snoring signals;

[0033] A denoising module, used for denoising the collected signal;

[0034] Feature extraction module, used to extract time domain, frequency domain and cepstrum feature parameters of the signal;

[0035] Feature selection module, used to select the most representative and independent features from the extracted features;

[0036] The machine learning module is used to train the model and realize the recognition and classification of snoring.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention not only extracts frequency domain features, but also integrates time domain and cepstrum features, thereby being able to more comprehensively reflect the physiological and pathological characteristics of snoring signals and improving the integrity and accuracy of feature extraction.

[0039] The present invention adopts a more advanced denoising algorithm, which effectively reduces the influence of environmental noise on the recognition of snoring signals and can maintain a high recognition accuracy even in a noisy environment.

[0040] The present invention significantly improves the speed of feature extraction and recognition by optimizing algorithms and data processing procedures, realizes real-time monitoring, and meets the high clinical requirements for real-time performance.

[0041] The present invention takes the characteristics of different groups of people and sleeping states into consideration during design, so that the technology has better universality and adaptability and a wider range of applications.

[0042] The design of the present invention reduces the need for professional recording equipment, and feature extraction and recognition can be performed on ordinary consumer-grade recording equipment, thereby reducing costs and facilitating the popularization and promotion of the technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 It is a schematic diagram of the overall system block diagram of the present invention. DETAILED DESCRIPTION

[0045] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be used for other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0046] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0047] See also Figure 1 The present invention provides a method for extracting characteristic parameters of snoring, comprising the following steps:

[0048] Collecting snoring signals, and obtaining snoring signals of patients through a sound collection device;

[0049] Preprocess the collected signals, including denoising and signal normalization;

[0050] Divide the preprocessed signal into frames, and extract the signal's time domain, frequency domain, cepstrum features and nonlinear characteristic parameters;

[0051] Representative and independent features are selected from the extracted features through correlation analysis and principal component analysis;

[0052] The selected features are input into the machine learning model for training to realize the recognition and classification of snoring.

[0053] In the present invention, a sound collection device combining a high-fidelity microphone (including Studio-Project B3) and a data acquisition card (including M-AudioFastTrackUltra8R) is used to ensure high-quality collection of snoring audio data.

[0054] When collecting, the recording device needs to be a certain distance away from the patient's mouth, and the environment should be kept as quiet as possible to reduce the interference of background noise in order to obtain a clearer snoring signal. Therefore, the steps for collection are:

[0055] 1. Place the recording device 10-20 cm away from the patient's mouth;

[0056] 2. Check and adjust the settings of the recording device, including volume, sampling rate and other parameters;

[0057] 3. Choose a quiet collection environment, close doors and windows, and reduce interference from external noise;

[0058] 4. During the collection process, pay attention to the status of the recording equipment to ensure that there is no malfunction or distortion, so as not to affect the recording quality.

[0059] In the present invention, the collected snoring signal is preprocessed, and the preprocessing includes:

[0060] 1. Denoising:

[0061] - Wavelet transform: The multi-resolution characteristics of wavelet transform can effectively separate the snoring signal from the ambient noise. By selecting the wavelet basis and the number of decomposition layers, the signal can be finely denoised, retaining the main features of the snoring while removing the background noise.

[0062] -Short-Fourier transform (STFT): STFT is a time-frequency analysis method that divides the signal into a series of overlapping short-time frames and performs Fourier transform on each frame to obtain the time-frequency spectrum of the signal. In the denoising process, the frequency components below the threshold can be suppressed by setting a threshold, thereby achieving the purpose of denoising.

[0063] 2. Normalization processing:

[0064] Normalization is the process of adjusting the signal amplitude to a uniform range. In snoring signal processing, normalization can eliminate the amplitude differences between different patients and different recording devices, making subsequent feature extraction more accurate and reliable.

[0065] 3. Frame processing:

[0066] Framing is to divide the preprocessed snoring signal into a series of short frames, each frame contains a signal of a certain time length. The characteristics of the snoring signal may change over time, and framing can capture these local characteristics.

[0067] In the present invention, the pre-processed snoring signal is further extracted to obtain time domain, frequency domain, cepstrum features and nonlinear feature parameters:

[0068] Time domain features: extract time domain features such as signal energy, mean, variance, kurtosis, and skewness.

[0069] Frequency domain features: The spectral features of the signal are extracted through Fourier transform, including spectral entropy, spectral center, spectral bandwidth, etc.

[0070] Cepstrum features: Mel-frequency cepstrum coefficients (MFCC) are used to extract the cepstrum features of the signal, reflecting the resonance characteristics of the signal.

[0071] Nonlinear features: Extract nonlinear features such as Lyapunov exponent and correlation dimension to capture the complexity of the signal.

[0072] In the present invention, correlation analysis, principal component analysis (PCA) and other methods are used to screen out the most representative and independent features from the extracted features, and machine learning algorithms, including support vector machine (SVM), random forest (RF), neural network (NN), etc., are used to input the screened features into the model for training.

[0073] Among them, correlation analysis is used to evaluate the correlation between feature parameters and reduce redundant features. The steps are as follows:

[0074] Step 1: Construct a snoring feature data matrix X = [X1, X2, ..., Xn], where Xi is the i-th feature;

[0075] Step 2: Calculate the correlation coefficient matrix R between features, where Ri represents the Pearson correlation coefficient between features Xi and Xj:

[0076]

[0077] Step 3: Analyze the correlation coefficient matrix. If the correlation between two features is high (such as R ij >0.8), then keep one and delete the other.

[0078] Principal component analysis is used to reduce the dimension and select the features with the most information. The steps are as follows:

[0079] Step 1: Centralize the feature data to obtain a zero-mean data matrix Z = X-μ;

[0080] Step 2: Calculate the covariance matrix C:

[0081]

[0082] Step 3: Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ,λ2,...,λn and corresponding eigenvectors V1,V2,...,Vn;

[0083] Step 4: Select the eigenvectors corresponding to the first k largest eigenvalues ​​as the new feature space after dimensionality reduction:

[0084] X PCA= Z · [V1, V2, V k ]

[0085] Step 5: Keep these principal components as the most representative snoring features.

[0086] In the present invention, the machine learning model is trained to identify and classify the characteristics of different groups of people and sleep states, including:

[0087] Real-time monitoring: Real-time monitoring and feature extraction of patients’ snoring.

[0088] Classification and recognition: Based on the feature vector output by the model, the snoring sound is classified and recognized to determine whether there are diseases such as sleep breathing disorders.

[0089] Feedback and adjustment: Provide corresponding feedback and suggestions based on the recognition results, including improving sleeping posture, using ventilator, etc.

[0090] The snoring classification of different populations and sleep states can be done using a machine learning model, support vector machine (SVM) classification steps:

[0091] Step 1: Input the filtered snoring feature dataset and construct a training set (xi,yi), where xi is the feature and yi is the category label;

[0092] Step 2: Construct the objective function:

[0093]

[0094] Among them, w is the weight of the classification hyperplane, C is the penalty coefficient, and ξi is the slack variable;

[0095] Step 3: Optimize the classification hyperplane through the constraint condition yi(wTxi+b)≥1-ξ.

[0096] Step 4: Use kernel function for nonlinear mapping to classify the snoring characteristics of different groups of people.

[0097] Figure 1 : is a system block diagram of the present invention. The present invention provides a system for the method of extracting snoring characteristic parameters, comprising:

[0098] A sound collection device for collecting snoring signals;

[0099] A denoising module, used for denoising the collected signal;

[0100] Feature extraction module, used to extract time domain, frequency domain and cepstrum feature parameters of the signal;

[0101] Feature selection module, used to select the most representative and independent features from the extracted features;

[0102] The machine learning module is used to train the model and realize the recognition and classification of snoring.

[0103] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A method for extracting characteristic parameters of snoring, characterized in that: The following steps are involved: Collecting snoring signals, and obtaining snoring signals of patients through a sound collection device; Preprocess the collected signals, including denoising and signal normalization; Divide the preprocessed signal into frames, and extract the signal's time domain, frequency domain, cepstrum features and nonlinear characteristic parameters; Representative and independent features are selected from the extracted features through correlation analysis and principal component analysis; The selected features are input into the machine learning model for training to realize the recognition and classification of snoring.

2. The method for extracting characteristic parameters of snoring according to claim 1, characterized in that: The denoising process uses wavelet transform or short-time Fourier transform.

3. The method for extracting characteristic parameters of snoring according to claim 1, characterized in that: The characteristic parameters include time domain characteristics: energy, mean, variance, kurtosis, and skewness; Frequency domain features: spectral entropy, spectral center, spectral bandwidth; cepstrum features: Mel-frequency cepstrum coefficients; nonlinear features: Lyapunov exponent, correlation dimension.

4. The method for extracting characteristic parameters of snoring according to claim 1, characterized in that: The machine learning model is a support vector machine, a random forest or a neural network.

5. A method for extracting snoring characteristic parameters according to any one of claims 1 to 4, characterized in that: The feature screening method comprises the following steps: Correlation analysis: used to evaluate the correlation between feature parameters and reduce redundant features; Principal Component Analysis: used to reduce dimensionality and select the features with the most information.

6. The method for extracting characteristic parameters of snoring according to claim 5, characterized in that: The machine learning model is trained to include identification and classification of characteristics of different populations and sleep states.

7. The method for extracting characteristic parameters of snoring according to claim 6, characterized in that: The identification and classification of characteristics of different populations and sleep states include: The machine learning model, support vector machine (SVM) classification, was adopted.

8. A system for implementing the method according to any one of claims 1 to 6, comprising: A sound collection device for collecting snoring signals; A denoising module, used for denoising the collected signal; Feature extraction module, used to extract time domain, frequency domain and cepstrum feature parameters of the signal; Feature selection module, used to select the most representative and independent features from the extracted features; The machine learning module is used to train the model and realize the recognition and classification of snoring.

Images