Intelligent breath sound monitoring and diagnosing system

Abstract

The invention belongs to the technical field of breath sound intelligent monitoring, and particularly relates to a breath sound intelligent monitoring and diagnosis system, which comprises a data acquisition system for acquiring an output signal of a breath sound sensor to obtain breath sound data; and the data analysis system is used for analyzing the breath sound data by adopting the breath sound characteristic quantity processing system based on the EEMD algorithm, and the breath sound characteristic quantity processing system based on the EEMD algorithm comprises a preprocessing unit, a denoising unit, a characteristic extraction unit, a classification unit and an output unit. The system can solve the problems that a traditional breath sound monitoring technology is low in accuracy, poor in anti-interference capacity, incapable of achieving long-term monitoring and the like, can be applied to breath sound state monitoring and intelligent diagnosis, greatly relieves the labor intensity of medical workers, improves the disease diagnosis efficiency and accuracy, ensures that the body function decline characteristics of a patient are found in time, and improves the diagnosis accuracy of the patient. A reliable basis is provided for implementation of intervention treatment, and the probability of deterioration of the patient is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring of respiratory sounds, and in particular relates to an intelligent monitoring and diagnosis system for respiratory sounds. Background Art

[0002] Breath sounds refer to the sound signals generated by airflow through the human airway under respiration and transmitted to the body surface in a vibrating mode. They contain physiological and pathological information of the respiratory system. They can preliminarily judge the location of the patient's lung disease lesions and the approximate type of disease. They can also quickly detect the progression of the disease and the treatment status after medication. They are an important diagnostic method for respiratory diseases.

[0003] Auscultation of breath sounds can diagnose respiratory diseases earlier than pulmonary function tests, blood gas analysis, X-ray diagnosis, etc.

[0004] For example, dry and wet rales in lung sounds are far more useful than X-rays in diagnosing pleural adhesions. However, studies have shown that the most sensitive frequency of the human ear is between 2000Hz and 5000Hz, while the frequency of respiratory sounds is usually between 50Hz and 2500Hz. This difference causes the human ear to only be able to distinguish some abnormal sounds, but not all of them. For example, the frequency of some abnormal wheezes is between 100Hz and 2500Hz, and the frequency of some abnormal rales is between 100Hz and 500Hz. These abnormal sounds are difficult to distinguish with the human ear alone.

[0005] Wheezing is an important basis for detecting asthma. The number, distribution and time of occurrence of rales are closely related to many lung diseases such as bronchopneumonia, chronic bronchitis, and pulmonary interstitial fibrosis. Obviously, the lack of this information will cause doctors to miss the best time to judge the condition, and ultimately delay the diagnosis and treatment of the condition. Secondly, traditional respiratory sound auscultation relies on the doctor's independent analysis and is easily affected by subjective factors such as the doctor's auscultation experience, and it is not possible to collect respiratory sounds continuously, in real time, and objectively. For patients who have been draped during surgery, real-time auscultation is even more impossible. For example, some heart surgeries often have pleural rupture and compression of the lungs. Laparoscopic surgery and head-down surgery often have too deep tracheal intubation, resulting in one side of the lung being blocked. These critical situations require real-time monitoring of respiratory sounds to facilitate doctors to deal with them in a timely manner and avoid missing the best treatment opportunity.

[0006] Therefore, the introduction of signal processing, computers, artificial intelligence and other technologies to record, quantitatively analyze, detect and identify respiratory sound signals to assist in diagnosis is of great significance both for improving the accuracy of actual clinical diagnosis and for theoretical research.

[0007] With the development of modern digital signal processing technology, people began to use frequency, amplitude and time to analyze respiratory sound signals. Initially, people could only obtain sound signals by comparing them with set sine waves, which would lead to the loss of intermediate frequencies. Fourier transform, especially fast Fourier transform, increases the components of identifiable signals and greatly reduces the occurrence of spectrum leakage. Since then, spectrum analysis methods based on Fourier analysis have become a powerful tool for respiratory sound analysis.

[0008] However, traditional time-frequency analysis methods such as Fourier transform, Wigner-Ville distribution, and wavelet transform have their own limitations when processing non-stationary signals. In practical applications, most signals are non-stationary and nonlinear. At present, some scholars have proposed the empirical mode decomposition (EMD) method in the prior art. However, the main defect of EMD is that if the time series does not meet the definition of completely white noise, then mode aliasing will occur; further, as a newer time-frequency analysis tool, the EEMD method still has some problems to be solved, such as endpoint effect, mode aliasing, envelope overshoot, etc. Summary of the invention

[0009] The purpose of the embodiments of the present invention is to provide a respiratory sound intelligent monitoring and diagnosis system, aiming to solve the technical problems of low accuracy and poor anti-interference ability of traditional respiratory sound monitoring technology.

[0010] To achieve the above object, the present invention provides the following technical solutions.

[0011] A respiratory sound intelligent monitoring and diagnosis system, comprising:

[0012] A data acquisition system, used for acquiring output signals of a respiratory sound sensor to obtain respiratory sound data;

[0013] A data analysis system, for analyzing respiratory sound data using a respiratory sound feature quantity processing system based on an EEMD algorithm, wherein the respiratory sound feature quantity processing system based on an EEMD algorithm comprises a preprocessing unit, a denoising unit, a feature extraction unit, a classification unit and an output unit; wherein:

[0014] A preprocessing unit, used for preprocessing the collected respiratory sound signal to obtain a separated respiratory gas phase;

[0015] A denoising unit, used to denoise the respiratory sound data using an EEMD algorithm and a blind source separation algorithm;

[0016] The feature extraction unit is used to calculate the MFCC value of one dimension in the MFCC for the breath sound and breath gas phase obtained after denoising, and use the mean, standard deviation, variance, minimum value, maximum value calculated from the MFCC of all dimensions and normalize them to form a feature value vector as a classification feature;

[0017] A classification unit, used to classify the respiratory sound data using SVM as a classifier;

[0018] The output unit is used to output the classification result of the respiratory sound data.

[0019] Furthermore, the data acquisition system includes a respiratory sound sensor and a data acquisition instrument, wherein the respiratory sound sensor adopts a piezoelectric sound sensor with a frequency response range of 60-4500Hz and a sensitivity of 4mV / Pa. The collected human respiratory sound signal is amplified and conditioned to output an analog voltage signal of 0.5-1.5V.

[0020] Furthermore, the piezoelectric sound sensor includes a shell, a plastic disk is arranged in the shell, a pressure-sensitive element is arranged on the plastic disk, and the pressure-sensitive element is made of PVDF piezoelectric material; a preamplifier is also arranged on the shell, and the preamplifier is connected to the data acquisition instrument through a wire.

[0021] Furthermore, the data acquisition system also includes an EDC system and an audio decoder, wherein the EDC system is used to record the respiratory sound data collected by the data acquisition instrument, and the audio decoder is used to perform analog-to-digital conversion on the respiratory sound data recorded by the EDC system. The audio decoder integrates ADC and DAC inside the chip, adopts Sigma-Delta oversampling technology, has a sampling frequency covering 8KHz to 96KHz, and transmits 16 / 20 / 24 / 32bit sampling data.

[0022] Furthermore, the process of denoising the respiratory sound data using the EEMD algorithm and the blind source separation algorithm includes:

[0023] Add Gaussian white noise to the breathing sound signal of the breathing sound data to generate multiple noisy signals: i (t) = x(t) + n i (t), i = 1, 2, 3, ..., M, where x(t) represents the original breathing sound signal of the breathing sound data when white noise is added for the i-th time, n i (t) represents the Gaussian white noise added at time t for the i-th time;

[0024] Perform EMD decomposition on each noisy signal to obtain a series of intrinsic mode functions IMFs, expressed as: r n (t) represents the decomposition residue, cj (t) represents the component that meets the IMF conditions;

[0025] The IMFs obtained by multiple decompositions are averaged together to screen out the IMFs with high correlation with the original signal; the screened IMFs are superimposed to reconstruct the denoised breathing sound signal;

[0026] The EEMD denoised signal is input as the observation signal into the blind source separation algorithm, and the FastICA algorithm is used to perform blind source separation to separate independent signal sources. The signal source representing the denoised breathing sound is selected from the separated signal sources to achieve denoising.

[0027] Furthermore, in the MFCC calculation, the frame length is 256, the number of sampling points in the overlapping area is 128, the Hamming window is used, the Gaussian mixed mode order is set to 16, the pre-emphasis coefficient is 0.93, and the MFCC order is 12.

[0028] Furthermore, in the process of classifying the respiratory sound data, the kernel function of the SVM model uses the Gauss kernel function.

[0029] Furthermore, the output unit adopts a CTMS data management system.

[0030] Compared with the prior art, the beneficial effects of the intelligent respiratory sound monitoring and diagnosis system of the present invention are:

[0031] First, the present invention can solve the problems of low accuracy, poor anti-interference ability and inability to monitor for a long time in traditional respiratory sound monitoring technology. It can be applied to the state monitoring and intelligent diagnosis of respiratory sounds, greatly reducing the labor intensity of medical staff, improving the efficiency and accuracy of disease diagnosis, ensuring the timely discovery of the characteristics of the patient's physical function decline, providing a reliable basis for the implementation of intervention treatment, and helping to reduce the probability of worsening of the patient's condition;

[0032] Second, in the respiratory sound feature processing system based on the EEMD algorithm of the present invention, the statistical characteristic of Gaussian white noise with uniform frequency distribution is utilized, and different white noises of the same amplitude are added each time to change the extreme point characteristics of the signal, and then the corresponding IMF obtained by multiple EMDs is averaged to offset the added white noise, thereby effectively suppressing the generation of modal aliasing;

[0033] Third, in the feature extraction of respiratory sounds in the present invention, the MFCC value calculated from the denoised signal eliminates a part of the high-frequency noise and retains the signal part; the present invention finds that if MFCC is directly used as a feature, the effect is not ideal. In order to better utilize MFCC features to realize the recognition of three types of respiratory sounds, the present invention uses the mean, standard deviation, variance, minimum value, and maximum value calculated by MFCC of all dimensions and normalizes them to form a eigenvalue vector as a classification feature. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0035] Figure 1 A schematic diagram of a system for intelligent monitoring and diagnosis of respiratory sounds provided by the present invention;

[0036] Figure 2 This is a block diagram of the architecture of a respiratory sound intelligent monitoring and diagnosis system of the present invention;

[0037] Figure 3 This is a structural block diagram of a respiratory sound intelligent monitoring and diagnosis system of the present invention;

[0038] Figure 4 A schematic diagram of the structure of the respiratory sound sensor provided by the present invention;

[0039] Figure 5 It is a schematic diagram of the arrangement of the respiratory sound sensor of the present invention on the human body;

[0040] Figure 6 It is a schematic diagram of the flow of the data analysis system of the present invention;

[0041] Figure 7 This is a schematic diagram of the 12th MFCC value calculated from the respiratory phase of the breath sound after denoising using the traditional EMD method;

[0042] Figure 8 This is a schematic diagram of the 12th MFCC value calculated from the respiratory sound and respiratory gas phase after EEMD denoising in the present invention.

[0043] The reference numerals are as follows:

[0044] 101. Data acquisition system; 102. Data analysis system;

[0045] 201. Plastic disk; 202. Pressure-sensitive element; 203. Housing; 204. Wire; 205. Preamplifier. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] Breath sounds refer to the sound signals generated by airflow through the human airway under respiration and transmitted to the body surface in a vibrating mode. They contain physiological and pathological information of the respiratory system. They can preliminarily judge the location of the patient's lung disease lesions and the approximate type of disease. They can also quickly detect the progression of the disease and the treatment status after medication. They are an important diagnostic method for respiratory diseases.

[0048] The present invention collects the patient's breathing sound signals by installing sensors on the patient's chest (front chest or back), and displays and saves a large amount of data in real time. Secondly, data analysis is carried out on a large number of patients' breathing sound signals, and key information that can reflect the disease in the data is extracted through intelligent algorithms, providing a basis for diagnosing a variety of diseases. Finally, the breathing state monitoring device and the intelligent algorithm are combined to form the breathing sound intelligent monitoring and diagnosis system of the present invention.

[0049] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0050] like Figure 1-Figure 3 As shown, in one embodiment of the present invention, a respiratory sound intelligent monitoring and diagnosis system is provided, comprising:

[0051] The data acquisition system 101 is used to collect the output signal of the respiratory sound sensor to obtain the respiratory sound data;

[0052] Breathing sounds are produced by the alveoli and bronchi in the human body. They are very weak when transmitted from the lungs to the body surface. In order to collect weak breathing sound signals, the sound sensor that collects breathing sounds needs to have high sensitivity, high amplification factor and strong resistance to environmental noise.

[0053] At present, acceleration sensors have strong anti-noise ability and high sensitivity, but their cost is also high; capacitive sound sensors are prone to introduce other sound interference. Therefore, the respiratory sound sensor of the present invention adopts a piezoelectric sensor. The piezoelectric sound sensor converts the respiratory sound signal into an electrical signal through a pressure-sensitive element, and has the characteristics of high reliability, high sensitivity and good noise resistance.

[0054] Furthermore, the data acquisition system includes a respiratory sound sensor and a data acquisition instrument, wherein the respiratory sound sensor adopts a piezoelectric sound sensor with a frequency response range of 60-4500Hz and a sensitivity of 4mV / Pa. The collected human respiratory sound signal is amplified and conditioned to output an analog voltage signal of 0.5-1.5V.

[0055] Further, such as Figure 4 As shown, the piezoelectric sound sensor provided by the embodiment of the present invention includes a housing 203, a plastic disk 201 is arranged in the housing 203, a pressure-sensitive element 202 is arranged on the plastic disk 201, and the pressure-sensitive element 202 is made of PVDF piezoelectric material; a preamplifier 205 is also arranged on the housing 203, and the preamplifier 205 is connected to the data acquisition instrument through a wire 204.

[0056] Furthermore, the data acquisition system also includes an EDC system and an audio decoder, wherein the EDC system is used to record the respiratory sound data collected by the data acquisition instrument, and the audio decoder is used to perform analog-to-digital conversion on the respiratory sound data recorded by the EDC system. The audio decoder integrates ADC and DAC inside the chip, adopts Sigma-Delta oversampling technology, has a sampling frequency covering 8KHz to 96KHz, and transmits 16 / 20 / 24 / 32bit sampling data.

[0057] The analog-to-digital conversion of the present invention uses a high-performance stereo audio codec TLV320AIC23 to perform analog-to-digital conversion on the data collected by the sensor, which is convenient for further analysis of the breathing sound at a later stage. VC5509A can configure the internal registers of AIC23 through the I2C bus interface. After each signal is collected by the AD chip, the data is sent to the McBSP (multi-channel buffered serial port) interface of the DSP. The DSP can read the breathing sound data. Each data is a 16-bit unsigned integer, and each left channel has a value.

[0058] Figure 5 The figure shows a schematic diagram of the arrangement of the respiratory sound sensor of the present invention in the human body. The present invention installs multiple sensors in the lungs on the back of the patient, collects the respiratory sound signals of the two left lung lobes and the three right lung lobes, filters out interfering noise, finds the best installation position and sensor combination, and displays the respiratory sound in a digital image, which is conducive to discovering the early symptoms of the patient's disease, guiding medical staff to carry out intervention treatment in time, and avoiding the deterioration of human body function.

[0059] For further information, please refer to Figure 1-Figure 3 , the respiratory sound intelligent monitoring and diagnosis system of the embodiment of the present invention further includes a data analysis system 102, which is used to analyze the respiratory sound data using a respiratory sound feature quantity processing system based on the EEMD algorithm;

[0060] like Figure 6 As shown, the method for processing the respiratory sound feature quantity based on the EEMD algorithm includes signal acquisition, preprocessing, EEMD method denoising, MFCC feature extraction, SVM method classification and output results;

[0061] The respiratory sound feature processing system based on the EEMD algorithm provided by the present invention comprises a preprocessing unit, a denoising unit, a feature extraction unit, a classification unit and an output unit; wherein:

[0062] A preprocessing unit, used for preprocessing the collected respiratory sound signal to obtain a separated respiratory gas phase;

[0063] During the preprocessing process, in order to better characterize the signal and reduce the impact of respiratory phase conversion on recognition, it is necessary to first preprocess the collected respiratory sound signal to obtain separate respiratory phases. The experimental preprocessing divides the collected respiratory sound signal into inspiratory phase and expiratory phase. Since each of the three types of respiratory sounds collected has 10 respiratory cycles, there are a total of 300 respiratory sound sample data. Each type of respiratory sound has two hundred samples (100 expiratory phases and 100 inspiratory phases), for a total of 600 sample data. During the test, 150 samples of each type of respiratory sound are randomly selected as training samples, and 50 samples are used as positive test samples (50 samples of other categories are randomly selected as negative test samples);

[0064] A denoising unit, used to denoise the respiratory sound data using an EEMD algorithm and a blind source separation algorithm;

[0065] The feature extraction unit is used to calculate the MFCC value of one dimension in the MFCC for the breath sound and breath gas phase obtained after denoising, and use the mean, standard deviation, variance, minimum value, maximum value calculated from the MFCC of all dimensions and normalize them to form a feature value vector as a classification feature;

[0066] Mel Frequency Cepstrum Coefficient (MFCC) can reflect the characteristics of speech from the perspective of the nonlinear psychological feeling of the human ear to high and low frequencies, and has good recognition and noise resistance. The present invention uses MFCC as a feature. In the MFCC calculation experiment, the frame length is 256, the number of sampling points in the overlapping area is 128, the Hamming window is used, the Gaussian mixed mode order is set to 16, the pre-emphasis coefficient is 0.93, and the MFCC order is 12.

[0067] In the feature extraction of respiratory sounds in the present invention, the MFCC value calculated from the denoised signal eliminates a part of the high-frequency noise and retains the signal part; the present invention finds that if MFCC is directly used as a feature, the effect is not ideal. In order to better utilize MFCC features to realize the recognition of three types of respiratory sounds, the present invention uses the mean, standard deviation, variance, minimum value, maximum value calculated from MFCC of all dimensions and normalizes them to form a eigenvalue vector as a classification feature.

[0068] Figure 7It is the MFCC value of one dimension in the MFCC calculated for each of the three respiratory phases of the breath sounds obtained after denoising using the traditional EMD method;

[0069] Figure 8 It is the MFCC value of one dimension in the MFCC calculated by the three respiratory sound and respiratory gas phase after denoising using the method of the present invention. As can be seen from the figure, the MFCC value calculated from the denoised signal eliminates a part of the high-frequency noise and retains the signal part.

[0070] Experiments have found that if MFCC is used directly as a feature, the effect is not ideal. In order to better utilize MFCC features to realize the recognition of three types of respiratory sounds, the present invention uses the mean, standard deviation, variance, minimum value, and maximum value calculated by MFCC of all dimensions and normalizes them to form a eigenvalue vector as a classification feature.

[0071] A classification unit, used to classify the respiratory sound data using SVM as a classifier;

[0072] The output unit is used to output the classification result of the respiratory sound data.

[0073] The respiratory sound signal is a typical nonlinear signal with complex and diverse signal details and large differences between individuals. There is little data available for research. The present invention uses SVM as a classifier to perform recognition experiments.

[0074] The object of the recognition experiment of the present invention is the three normal breath sounds of 10 healthy adults collected above. The goal is to classify the input breath sounds into three categories: alveolar breath sounds, bronchial breath sounds and bronchoalveolar breath sounds, which belongs to a multi-classification problem. The "one-to-the-rest" combined classification method was used during the test. The strategy of this method is to select one category from the sample as the positive class and the rest as the negative class. For example, when doing the experiment of identifying alveolar breath sounds, alveolar breath sounds are used as the positive class, and the other two bronchial breath sounds and bronchoalveolar breath sounds are used as negative classes. The recognition of the other two breath sounds is similar. The test used the open source library LIBSVM, and the kernel function used was the Gauss kernel function.

[0075] Furthermore, the process of denoising the respiratory sound data using the EEMD algorithm and the blind source separation algorithm includes:

[0076] Add Gaussian white noise to the breathing sound signal of the breathing sound data to generate multiple noisy signals: i (t) = x(t) + n i (t), i = 1, 2, 3, ..., M, where x(t) represents the original breathing sound signal of the breathing sound data when white noise is added for the i-th time, n i(t) represents the Gaussian white noise added at time t for the i-th time;

[0077] Perform EMD decomposition on each noisy signal to obtain a series of intrinsic mode functions IMFs, expressed as: r n (t) represents the decomposition residue, c j (t) represents the component that meets the IMF conditions;

[0078] The IMFs obtained by multiple decompositions are averaged together to screen out the IMFs with high correlation with the original signal; the screened IMFs are superimposed to reconstruct the denoised breathing sound signal;

[0079] The EEMD denoised signal is input as the observation signal into the blind source separation algorithm, and the FastICA algorithm is used to perform blind source separation to separate independent signal sources. The signal source representing the denoised breathing sound is selected from the separated signal sources to achieve denoising.

[0080] In the respiratory sound feature processing system based on the EEMD algorithm of the present invention, the statistical characteristic of Gaussian white noise with uniform frequency distribution is utilized. Different white noises of equal amplitude are added each time to change the extreme point characteristics of the signal. Then, the corresponding IMFs obtained by multiple EMDs are averaged to offset the added white noise, thereby effectively suppressing the generation of modal aliasing.

[0081] Furthermore, the output unit adopts the CTMS data management system, and the data management adopts the CTMS (ClinicalTrial Management System) clinical trial information management system. CTMS is a project management tool that can effectively plan, manage and track clinical research projects. It consists of different tools. CTMS can select centers and monitoring, patient recruitment, contact management, financial management and regulatory documents. CTMS focuses on the management of the research project itself and does not focus on data collection. CTMS and EDC are complementary. Most clinical studies use these two systems. EDC focuses on collecting patient data, and CTMS focuses on the project management of clinical trials. The systems can be directly connected, and part of the EDC data can be integrated into CTMS.

[0082] Furthermore, the recognition results after using the traditional EMD denoising method and the EEMD denoising method of the present invention, from the experimental results in the table, after using the new method for denoising, the recognition rate of alveolar breath sounds has increased by 6%, the recognition rate of bronchial breath sounds has increased by 12%, the recognition rate of bronchoalveolar breath sounds has increased by 8%, and the average recognition rate has increased by 9.67%.

[0083]

[0084] Table 1. Recognition rates after traditional EMD denoising method.

[0085]

[0086] Table 2. Recognition rates after EEMD denoising method.

[0087] In summary, the present invention can solve the problems of low accuracy, poor anti-interference ability and inability to monitor for a long time in traditional respiratory sound monitoring technology, and can be applied to the state monitoring and intelligent diagnosis of respiratory sounds, which greatly reduces the labor intensity of medical staff, improves the efficiency and accuracy of disease diagnosis, ensures the timely discovery of the characteristics of the patient's physical function decline, provides a reliable basis for the implementation of intervention treatment, and is conducive to reducing the probability of worsening of the patient's condition;

[0088] The above solutions are only an illustration of a preferred example, but are not limited thereto. When implementing the present invention, appropriate replacement and / or modification can be performed according to user needs.

[0089] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily realized. Therefore, without departing from the general concept defined by the claims and equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. A respiratory sound intelligent monitoring and diagnosis system, characterized in that: The respiratory sound intelligent monitoring and diagnosis system includes: A data acquisition system (101) is used to acquire output signals of a respiratory sound sensor to obtain respiratory sound data; A data analysis system (102) is used to analyze the respiratory sound data using a respiratory sound feature quantity processing system based on the EEMD algorithm, wherein the respiratory sound feature quantity processing system based on the EEMD algorithm comprises a preprocessing unit, a denoising unit, a feature extraction unit, a classification unit and an output unit; wherein: A preprocessing unit, used for preprocessing the collected respiratory sound signal to obtain a separated respiratory gas phase; A denoising unit, used to denoise the respiratory sound data using an EEMD algorithm and a blind source separation algorithm; The feature extraction unit is used to calculate the MFCC value of one dimension in the MFCC for the breath sound and breath gas phase obtained after denoising, and use the mean, standard deviation, variance, minimum value, maximum value calculated from the MFCC of all dimensions and normalize them to form a feature value vector as a classification feature; A classification unit, used to classify the respiratory sound data using SVM as a classifier; The output unit is used to output the classification result of the respiratory sound data.

2. The intelligent respiratory sound monitoring and diagnosis system according to claim 1, characterized in that: The data acquisition system includes a respiratory sound sensor and a data acquisition instrument, wherein the respiratory sound sensor adopts a piezoelectric sound sensor with a frequency response range of 60-4500Hz and a sensitivity of 4mV / Pa. The collected human respiratory sound signal is amplified and conditioned to output an analog voltage signal of 0.5-1.5V.

3. The intelligent respiratory sound monitoring and diagnosis system according to claim 2, characterized in that: The piezoelectric sound sensor comprises a housing (203), a plastic disk (201) is arranged in the housing (203), a pressure-sensitive element (202) is arranged on the plastic disk (201), and the pressure-sensitive element (202) is made of PVDF piezoelectric material; a preamplifier (205) is also arranged on the housing (203), and the preamplifier (205) is connected to a data acquisition instrument via a wire (204).

4. The intelligent respiratory sound monitoring and diagnosis system according to claim 3 is characterized in that: The data acquisition system also includes an EDC system and an audio decoder, wherein the EDC system is used to record the respiratory sound data collected by the data acquisition instrument, and the audio decoder is used to perform analog-to-digital conversion on the respiratory sound data recorded by the EDC system. The audio decoder integrates ADC and DAC inside the chip, adopts Sigma-Delta oversampling technology, has a sampling frequency covering 8KHz to 96KHz, and transmits 16 / 20 / 24 / 32bit sampling data.

5. The intelligent respiratory sound monitoring and diagnosis system according to claim 4, characterized in that: The process of denoising the respiratory sound data by using the EEMD algorithm and the blind source separation algorithm includes: Add Gaussian white noise to the breathing sound signal of the breathing sound data to generate multiple noisy signals: i (t) = x(t) + n i (t), i = 1, 2, 3, ..., M, where x(t) represents the original breathing sound signal of the breathing sound data when white noise is added for the i-th time, n i (t) represents the Gaussian white noise added at time t for the i-th time; Perform EMD decomposition on each noisy signal to obtain a series of intrinsic mode functions IMFs, expressed as: r n (t) represents the decomposition residue, c j (t) represents the component that meets the IMF conditions; The IMFs obtained by multiple decompositions are averaged together to screen out the IMFs with high correlation with the original signal; the screened IMFs are superimposed to reconstruct the denoised breathing sound signal; The EEMD denoised signal is input as the observation signal into the blind source separation algorithm, and the FastICA algorithm is used to perform blind source separation to separate independent signal sources. The signal source representing the denoised breathing sound is selected from the separated signal sources to achieve denoising.

6. The intelligent respiratory sound monitoring and diagnosis system according to claim 5, characterized in that: In the MFCC calculation, the frame length is 256, the number of sampling points in the overlapping area is 128, the Hamming window is used, the Gaussian mixed mode order is set to 16, the pre-emphasis coefficient is 0.93, and the MFCC order is 12.

7. The intelligent respiratory sound monitoring and diagnosis system according to claim 6, characterized in that: In the process of classifying respiratory sound data, the kernel function of the SVM model uses the Gauss kernel function.

8. The intelligent respiratory sound monitoring and diagnosis system according to claim 7, characterized in that: The output unit adopts a CTMS data management system.

Images