Snore feature extraction and analysis method

Abstract

The snore feature extraction and analysis method comprises the following steps: (a) data acquisition: acquiring snore data of different crowds in a controlled environment through a microphone or a multi-sensor device; (b) signal preprocessing: denoising, normalizing and framing the acquired snore signals; (c) feature extraction: extracting snore features from time domain, frequency domain, cepstrum and statistical features; (d) feature optimization: performing feature screening on the basis of crowd differences to screen snore features which are most representative and mutually independent; (e) classification model training: selecting a suitable machine learning model for training; (f) system integration: constructing a snore analysis system. The snore analysis method can adapt to patient groups of different ages, genders and health states, and can be widely applied to snore analysis; the analysis accuracy and the diagnosis effect are improved through feature optimization and a personalized classification model for different crowds; the kit supports early screening and diagnosis of various sleep-related diseases, and has important clinical value.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal processing, and in particular to a method for extracting and analyzing snoring features. Background Art

[0002] As a common sleep phenomenon, the characteristics of snoring are affected by many individual differences, including age, gender, body shape, and health status. This makes snoring highly diverse and complex. Although some snoring analysis methods already exist, they are often optimized for specific populations or specific conditions, so they may encounter challenges when applied to a wider patient population.

[0003] This limitation mainly stems from the following aspects:

[0004] 1. Insufficient data representation: Existing analytical methods may have been developed and validated based on limited datasets, which may not adequately represent all potential patient populations.

[0005] 2. Limitations of feature extraction: Some analysis methods may only focus on certain specific features of snoring and ignore other features that may be equally important and may vary in different populations.

[0006] 3. Model generalization ability: Even if a model performs well on a specific dataset, it may not generalize well to new, unseen data, especially when the data comes from a different population. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides a snoring feature extraction and analysis method, which can process snoring signals of people of different ages, genders, health conditions, etc., provide accurate snoring feature analysis, and support the formulation of personalized medical diagnosis and treatment plans.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A method for extracting and analyzing snoring features comprises the following steps:

[0010] (a) Data collection: Snoring data from different groups of people are collected in a controlled environment using microphones or multi-sensor devices;

[0011] (b) Signal preprocessing: denoising, normalizing, and framing the collected snoring signal;

[0012] (c) Feature extraction: extracting snoring features from time domain, frequency domain, cepstrum and statistical features;

[0013] (d) Feature optimization: feature screening is performed based on population differences to select the most representative and independent snoring features;

[0014] (e) Classification model training: Select a suitable machine learning model for training;

[0015] (f) System integration: building a snoring analysis system.

[0016] Preferably, the feature extraction step includes extracting time domain features, frequency domain features, cepstrum features and statistical features, wherein:

[0017] Time domain characteristics include energy, amplitude, and duration;

[0018] Frequency domain features are calculated through Fourier transform to calculate spectrum entropy, spectrum center and spectrum bandwidth;

[0019] The cepstral features are obtained by calculating the Mel-frequency cepstral coefficients (MFCC);

[0020] Statistical characteristics include kurtosis and skewness.

[0021] Preferably, the time domain feature extraction step further comprises calculating a zero crossing rate to capture instantaneous changes of the signal.

[0022] Preferably, the difference analysis performed in the feature optimization step adopts a one-way analysis of variance method, and the calculation steps are as follows:

[0023] (a) Calculate the mean square error between groups, the formula is:

[0024]

[0025] Among them, k is the number of groups, n i is the number of samples in the i-th group, X i is the mean of the i-th group, and X is the total mean;

[0026] (b) Calculate the within-group mean square error MSW using the formula:

[0027]

[0028] Among them, N is the total number of samples, X ij is the jth sample value of the i-th group;

[0029] (c) Calculate the F value using the formula:

[0030]

[0031] Preferably, the feature screening performed in the feature optimization step adopts the LASSO regression method, and the specific steps are as follows:

[0032] (a) Data standardization: Standardize the feature data to ensure that the contribution of each feature to the model is at the same level;

[0033] (b) Model training: Use the training data to train the LASSO regression model, and introduce the L1 regularization term for feature selection. The model formula is:

[0034]

[0035] Among them, y i is the target variable, X ij is the jth feature of the i-th sample, β j is the regression coefficient of the jth feature, and λ is the regularization parameter;

[0036] (c) Feature selection: Based on the size of the regression coefficient, features with zero regression coefficients are eliminated, and features with non-zero coefficients are retained.

[0037] Preferably, the classification model selects a support vector machine, a random forest or a deep neural network, and the generalization ability of the model is improved through cross-population model training.

[0038] Preferably, the system integration step includes constructing a large-scale snoring feature database covering different populations to support continuous training and optimization of the model.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] This application can adapt to patient groups of different ages, genders, and health conditions to achieve widely applicable snoring analysis; improve analysis accuracy and diagnostic results through feature optimization and personalized classification models for different populations; and support early screening and diagnosis of multiple sleep-related diseases. DETAILED DESCRIPTION

[0041] 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.

[0042] Those skilled in the art should understand that, in the disclosure of the present invention, the orientation or position indicated by terms such as "longitudinal", "lateral", "up", "down", "left", "right", "front", "back", "vertical", "horizontal", "top", "bottom", "inside" and "outside" are based on the orientation or position relationship shown, which is only a simplified description for the convenience of describing the present invention, and does not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the above terms should not be understood as limiting the present invention.

[0043] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0044] Example:

[0045] The present invention provides a method for extracting and analyzing snoring features, comprising the following steps:

[0046] (a) Data collection: Snoring data from different groups of people are collected in a controlled environment using microphones or multi-sensor devices;

[0047] (b) Signal preprocessing: denoising, normalizing, and framing the collected snoring signal;

[0048] (c) Feature extraction: extracting snoring features from time domain, frequency domain, cepstrum and statistical features;

[0049] (d) Feature optimization: feature screening is performed based on population differences to select the most representative and independent snoring features;

[0050] (e) Classification model training: Select a suitable machine learning model for training;

[0051] (f) System integration: building a snoring analysis system.

[0052] In the present invention, snoring data of different groups of people (classified by age, gender, health status, etc.) are collected in a controlled environment through high-quality microphones or multi-sensor devices; the collected data will include continuous snoring records in a natural sleep state.

[0053] In the present invention, feature extraction includes extracting time domain features, frequency domain features, cepstrum features and statistical features, etc., wherein:

[0054] Time domain features extract the energy, amplitude, duration and other time domain features of each frame signal to capture the transient characteristics of the signal;

[0055] Frequency domain features: The frequency domain features such as spectrum entropy, spectrum center, spectrum bandwidth, etc. of the signal are calculated through Fourier transform to analyze the spectrum distribution of the signal;

[0056] The cepstral feature obtains the pitch characteristics of the signal by calculating the Mel-frequency cepstral coefficient (MFCC);

[0057] Statistical features include kurtosis, skewness, etc., which provide a statistical description of the signal distribution.

[0058] Furthermore, the time domain feature extraction step also includes calculating the zero crossing rate to capture the instantaneous changes of the signal.

[0059] Furthermore, the difference analysis performed in the feature optimization step adopts the one-way analysis of variance method, and its calculation steps are as follows:

[0060] (a) Calculate the mean square error between groups, the formula is:

[0061]

[0062] Among them, k is the number of groups, n i is the number of samples in the i-th group, X i is the mean of the i-th group, and X is the total mean;

[0063] (b) Calculate the within-group mean square error MSW using the formula:

[0064]

[0065] Among them, N is the total number of samples, X ij is the jth sample value of the i-th group;

[0066] (c) Calculate the F value using the formula:

[0067]

[0068] In the specific implementation, by proposing the null hypothesis, assuming that the means of each group are equal, comparing the F value obtained by variance analysis with the critical value, it is determined whether to reject the null hypothesis. If the F value is significant, it means that there are significant differences in the snoring characteristics of different populations, and further analysis is conducted to determine which groups have the greatest differences.

[0069] Furthermore, the feature screening performed in the feature optimization step adopts the LASSO regression method, and the specific steps are as follows:

[0070] (a) Data standardization: Standardize the feature data to ensure that the contribution of each feature to the model is at the same level;

[0071] (b) Model training: Use the training data to train the LASSO regression model, and introduce the L1 regularization term for feature selection. The model formula is:

[0072]

[0073] Among them, y i is the target variable, X ij is the jth feature of the i-th sample, β j is the regression coefficient of the jth feature, and λ is the regularization parameter;

[0074] (c) Feature selection: Based on the size of the regression coefficient, features with zero regression coefficients are eliminated, and features with non-zero coefficients are retained.

[0075] Standardize the feature data to ensure that the contribution of each feature to the model is at the same level. Use the training data to train the LASSO regression model, and adjust the regularization parameter λ to control the sparsity of the model. After the model training is completed, observe the size of the regression coefficient βj. For features with a regression coefficient of zero, remove them; retain features with non-zero coefficients as the final screening results, verify the effectiveness of the screening features through cross-validation or other evaluation methods, and optimize the model.

[0076] The feature optimization of the present invention is based on the differences in population characteristics, and uses correlation analysis, principal component analysis (PCA) and other methods to screen the most representative and independent features. The population includes men, women, children, the elderly, healthy people, snoring patients, etc., and special feature sets are generated for each group to ensure the accuracy of the analysis.

[0077] When training the classification model of the present invention, a machine learning model is first selected, including support vector machine SVM, random forest RF, deep neural network DNN, etc., and training is performed on the characteristic data of different populations. Through multi-task learning or transfer learning methods, the characteristic data of different populations are incorporated into the training of the same model to improve the generalization ability of the model. The model performance is evaluated using cross-validation and test sets, and the classification accuracy is improved by adjusting parameters and optimizing algorithms.

[0078] Those skilled in the art should understand that the embodiments of the present invention shown in the above description 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 and analyzing snoring features, characterized in that: The following steps are involved: (a) Data collection: Snoring data from different groups of people are collected in a controlled environment using microphones or multi-sensor devices; (b) Signal preprocessing: denoising, normalizing, and framing the collected snoring signal; (c) Feature extraction: extracting snoring features from time domain, frequency domain, cepstrum and statistical features; (d) Feature optimization: feature screening is performed based on population differences to select the most representative and independent snoring features; (e) Classification model training: Select a suitable machine learning model for training; (f) System integration: building a snoring analysis system.

2. A method for extracting and analyzing snoring features according to claim 1, characterized in that: The feature extraction step includes extracting time domain features, frequency domain features, cepstrum features and statistical features, wherein: Time domain characteristics include energy, amplitude, and duration; Frequency domain features are calculated through Fourier transform to calculate spectrum entropy, spectrum center and spectrum bandwidth; The cepstral features are obtained by calculating the Mel-frequency cepstral coefficients (MFCC); Statistical characteristics include kurtosis and skewness.

3. A method for extracting and analyzing snoring features according to claim 2, characterized in that: The time domain feature extraction step also includes calculating the zero-crossing rate to capture the instantaneous changes of the signal.

4. The method for extracting and analyzing snoring features according to claim 1, characterized in that: The difference analysis performed in the feature optimization step adopts a one-way analysis of variance method, and the calculation steps are as follows: (a) Calculate the mean square error between groups, the formula is: Among them, k is the number of groups, n i is the number of samples in the i-th group, X i is the mean of the i-th group, and X is the total mean; (b) Calculate the within-group mean square error MSW using the formula: Among them, N is the total number of samples, X ij is the jth sample value of the i-th group; (c) Calculate the F value using the formula:

5. The method for extracting and analyzing snoring features according to claim 1, characterized in that: The feature screening performed in the feature optimization step adopts the LASSO regression method, and the specific steps are as follows: (a) Data standardization: Standardize the feature data to ensure that the contribution of each feature to the model is at the same level; (b) Model training: Use the training data to train the LASSO regression model, and introduce the L1 regularization term for feature selection. The model formula is: Among them, y i is the target variable, X ij is the jth feature of the i-th sample, β j is the regression coefficient of the jth feature, and λ is the regularization parameter; (c) Feature selection: Based on the size of the regression coefficient, features with zero regression coefficients are eliminated, and features with non-zero coefficients are retained.

6. A method for extracting and analyzing snoring features according to claim 1, characterized in that: The classification model selects support vector machine, random forest or deep neural network, and improves the generalization ability of the model through cross-population model training.

7. The method for extracting and analyzing snoring features according to claim 1, characterized in that: The system integration step includes building a large-scale snoring feature database covering different populations to support continuous training and optimization of the model.

Images