Portable all-weather cough monitoring equipment and use method
By designing a portable all-weather cough monitoring device and using acoustic sensors and data processing units for real-time monitoring and analysis, the problem of difficulty in monitoring patients' coughs in the prior art is solved, and accurate monitoring and recording of patients' cough status is achieved, and the treatment effect is improved.
Patent Information
- Application Number
- CN202510093618.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-20
- Publication Date
- 2025-05-02
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to monitor the patient's cough in real time around the clock, especially when the patient is not in the hospital in the early stages of the onset of the disease and during the recovery period, which makes it difficult for medical staff to accurately grasp the patient's cough.
A portable all-weather cough monitoring device is designed, including housing components, acoustic sensors, data processing units, power management units and abnormal alarm units. The device collects coughing sounds through acoustic sensors, the data processing unit performs feature extraction and classification processing, the power management unit provides a stable power supply, and the abnormal alarm unit issues an alarm when an abnormality occurs.
Real-time monitoring and recording of the patient's cough status is achieved, and the patient can be tested anytime and anywhere, which improves medical staff's accurate grasp of the patient's cough status and enhances the treatment effect.
Smart Images
Figure CN119908706A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of cough monitoring equipment, and in particular to a portable all-weather cough monitoring equipment and a use method thereof. Background Art
[0002] Coughing is an important defensive reflex of the body, which can expel inhaled dust, pathogenic microorganisms, smoke and other foreign matter from the respiratory system. With the rapid development of industrialization and urbanization, pollutants and irritants in the living environment have increased, resulting in an increase in the number of patients with pathological coughs year by year. Pathological coughs are usually divided into acute coughs, subacute coughs and chronic coughs.
[0003] Cough is an important part of the risk scores of various lung diseases. The understanding of the patient's cough condition is crucial in clinical practice. Therefore, the monitoring of cough is of great medical significance. At present, most medical staff understand the patient's cough condition by asking coughing patients and obtaining their feedback on the cough condition. Due to the large differences in the cultural level, medical knowledge theory, language proficiency and self-tolerance of coughing patients, or the lack of such ability, it is difficult for them to use quantitative data to truthfully feedback their coughing conditions to medical staff. There are often vague or even wrong descriptions of coughing conditions. Especially when it comes to all-weather coughing, coughing patients are even more unable to truthfully feedback the frequency and intensity of coughing within such a large time span to medical staff. This leads to distorted coughing feedback for medical staff, which can easily mislead medical staff to make wrong judgments.
[0004] At present, some large hospitals also have equipment that can perform round-the-clock detection of patients' coughing conditions, but most of them are large desktop equipment that patients cannot carry with them. Patients can only be tested in the hospital. Once the patient leaves the hospital, the test cannot be performed. Compared with the patient's coughing condition during treatment in the hospital, the coughing condition in the early stage of the disease and the recovery period makes it particularly important for medical staff to grasp the condition. However, most patients are not in the hospital in the early stage of the disease and the recovery period, which brings great difficulties to the detection of patients' coughing and makes it difficult for medical staff to accurately grasp the patient's coughing condition.
[0005] To this end, we provide a portable all-weather cough monitoring device and a method of use to solve the above-mentioned problems. Summary of the invention
[0006] The purpose of the present invention is to provide a portable all-weather cough monitoring device and a method of use, which has the advantage of being able to monitor the patient's coughing status in real time, solving the problem that most patients are not in the hospital in the early stage of the disease and the recovery period, which brings great difficulties to the detection of the patient's cough and makes it difficult for medical staff to accurately grasp the patient's coughing situation.
[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a portable all-weather cough monitoring device, comprising a shell component, the shell component comprising a shell, the surface of the shell is installed with an acoustic sensor, a data processing unit, a power management unit and an abnormal alarm unit, the number of the acoustic sensors is two, and they are respectively installed at the two ends of the shell, the data processing unit and the power management unit are both installed in the inner cavity of the shell, the data processing unit comprises a DSP chip, a storage module, a communication module and an output interface, the DSP chip is used to process data, the storage module is used to store data, the communication module and the output interface are used to transmit data, the power management unit comprises a rechargeable battery and a charging interface, the rechargeable battery is used to provide power to the acoustic sensor, the data processing unit and the abnormal alarm unit, the abnormal alarm unit comprises a display screen, an indicator light and a buzzer, the display screen is fixed to one side of the shell surface, the indicator light and the buzzer are both fixed to the other side of the shell surface, the output end of the acoustic sensor is connected to the input end of the data processing unit, and the output end of the data processing unit is connected to the input end of the abnormal alarm unit.
[0008] Preferably, the DSP chip includes feature extraction, machine learning classification and model training, the feature extraction is used to extract cough feature information, the machine learning classification is used to classify and process cough audio, and the model training is used for cough testing.
[0009] Preferably, the storage module includes a storage hard disk and cloud storage, the storage module is used to store historical monitoring information and model training data, and the communication module is connected to an external mobile phone APP via Bluetooth or WiFi.
[0010] Preferably, the feature extraction uses VGGish to automatically extract the collected cough audio, the machine learning classification uses a Logistic regression algorithm, and the model training uses a convolutional neural network learning model.
[0011] Preferably, the housing assembly further comprises a bracket and a gasket, and two of the bracket and the gasket are provided, the bracket is fixed to two sides of the rear end of the housing, and the gasket is fixed to two sides of the front end of the housing.
[0012] Preferably, the Logistic regression algorithm uses a sigmoid activation function to convert the predicted value z generated by linear regression into T +b is mapped to a y value between 0 and 1, and the sigmoid function is calculated as follows:
[0013]
[0014] This gives the Logistic regression formula:
[0015]
[0016] Then the value of the prediction function represents the probability of y=1, so the probabilities of y=1 and y=0 are:
[0017]
[0018] The input data is substituted into z and then mapped to between 0 and 1 by the sigmoid function. After the numerical value is obtained, it is classified. Those greater than 0.5 are classified into the positive class, otherwise they are classified into the negative class.
[0019] Preferably, the model training steps are as follows:
[0020] First, preprocessing is performed. Each input cough record is divided into 6-second audio blocks, padded as needed, and processed with MFCC packages. The CNN structure consists of three parallel ResNet50s. The 7×7×2048 4-dimensional tensor output layer of each ResNet50 model is connected in parallel. These ResNet50s are not pre-trained.
[0021] The outputs of these three concatenated tensors are then merged together using a 2D layer of GlobalAveragePooling, using a deep neural network layer of 1024 neurons activated with ReLU;
[0022] Finally, the binary classification layer of the Sigmoid activation function is used. In addition to the patient's cough sound data, the network can also be trained using three types of biomarkers: muscle degeneration, vocal cord changes, and emotional changes. These three types of biomarker features are added to the CNN network. Through the feature of muscle fatigue and degeneration, a memory decay model is introduced. By using M = (I x )=Poisson(λ)I X The Poisson mask in modifies the input signal for all training and test sets, and then multiplies this value by the parameter I x and λ are randomly distributed to calculate the cough recording MFCC point I x The Poisson mask of , where λ is the average value of all values in MFCC, and the Poisson distribution formula is:
[0023]
[0024] According to the vocal cord change characteristics, a ResNet50 network with an input shape of (300, 200) from MFCC is trained, and LibriSpeech is used for distinction. By introducing emotional change features, an emotional speech classifier model is trained to learn the emotional features on the RAVDESS speech dataset.
[0025] Preferably, the method of use comprises the following steps:
[0026] a) First, the housing assembly is worn on the neck, the front end of the housing is located on both sides of the front of the neck, the bracket fits the back of the neck, the gasket fits the skin of the front of the neck, the display screen can display the working status, the rechargeable battery provides power to the monitoring device, and the charging operation can be performed through the charging interface and the external power adapter. When the wearer coughs, the acoustic sensor can collect the cough sound;
[0027] b) The acoustic sensor performs sample preprocessing, feature extraction, feature selection, and training and testing on the collected audio information through the data processing unit to obtain the monitoring results. The cloud storage of the storage module stores cough audio information and related symptom cases. The DSP chip uses the original data to establish a training model in advance, and analyzes the collected cough sounds through the training model. The real-time cough sounds collected by the acoustic sensor are automatically extracted through VGGish, and then the extracted audio information is classified through the Logistic regression algorithm. After classification, the cough information is analyzed through the established training model, so that the cough status can be monitored and recorded, and the cough information is recorded in the hard disk of the storage module;
[0028] c) The monitored cough results can be transmitted to an external mobile phone APP through the communication module using Bluetooth or WiFi. When the patient comes for a follow-up visit, the cough data during the patient's recovery period can be exported through the output interface or queried through the mobile phone APP. When the cough is more serious, the buzzer will sound an alarm and the indicator light will turn red to remind the patient and caregiver.
[0029] Compared with the prior art, the present invention has the following beneficial effects:
[0030] The present invention achieves the effect of real-time monitoring of the patient's coughing status around the clock by providing a shell component, a sound-sensitive sensor, a data processing unit, a power management unit and an abnormal alarm unit. The shell, the bracket and the gasket cooperate to make it convenient for patients to carry and use. The sound-sensitive sensor can collect coughing sounds, the data processing unit can analyze and record the coughing status, the power management unit can provide power to the monitoring equipment to ensure its stable operation, and the abnormal alarm unit can promptly notify medical staff when an abnormality occurs, so that the medical staff can promptly grasp the patient's coughing condition during the rehabilitation period, thereby effectively improving the treatment effect of the medical staff. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a schematic diagram of the main structure of the present invention;
[0032] Figure 2 It is a structural schematic diagram of the connection state between the housing component and the abnormal alarm unit of the present invention;
[0033] Figure 3 It is a schematic diagram of the system flow of the present invention;
[0034] Figure 4 Schematic diagram of the cough monitoring process of the present invention.
[0035] In the figure: 1. Shell assembly; 101. Shell; 102. Bracket; 103. Gasket; 2. Acoustic sensor; 3. Data processing unit; 301. DSP chip; 302. Storage module; 303. Communication module; 304. Output interface; 4. Power management unit; 401. Rechargeable battery; 402. Charging interface; 5. Abnormal alarm unit; 501. Display screen; 502. Indicator light; 503. Buzzer. DETAILED DESCRIPTION
[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0037] The shell assembly 1, shell 101, bracket 102, gasket 103, acoustic sensor 2, data processing unit 3, DSP chip 301, storage module 302, communication module 303, output interface 304, power management unit 4, rechargeable battery 401, charging interface 402, abnormal alarm unit 5, display screen 501, indicator light 502 and buzzer 503 of the present invention are all universal standard parts or parts known to those skilled in the art, and their structures and principles can be known to those skilled in the art through technical manuals or through conventional experimental methods.
[0038] Example 1
[0039] like Figure 1-4As shown in the figure, it is the first embodiment of the present invention, which provides a portable all-weather cough monitoring device, including a shell component 1, the shell component 1 includes a shell 101, the surface of the shell 101 is installed with an acoustic sensor 2, a data processing unit 3, a power management unit 4 and an abnormal alarm unit 5, the number of acoustic sensors 2 is two, and they are respectively installed at both ends of the shell 101, the data processing unit 3 and the power management unit 4 are both installed in the inner cavity of the shell 101, the data processing unit 3 includes a DSP chip 301, a storage module 302, a communication module 303 and an output interface 304, the DSP chip 301 is used to process data, and the storage module 302 is used to The communication module 303 and the output interface 304 are used to store data, the power management unit 4 includes a rechargeable battery 401 and a charging interface 402, the rechargeable battery 401 is used to provide power to the acoustic sensor 2, the data processing unit 3 and the abnormal alarm unit 5, the abnormal alarm unit 5 includes a display screen 501, an indicator light 502 and a buzzer 503, the display screen 501 is fixed to one side of the surface of the shell 101, the indicator light 502 and the buzzer 503 are both fixed to the other side of the surface of the shell 101, the output end of the acoustic sensor 2 is connected to the input end of the data processing unit 3, and the output end of the data processing unit 3 is connected to the input end of the abnormal alarm unit 5.
[0040] like Figure 1-4 As shown, the shell component 1 is worn on the neck, the front end of the shell 101 is located on both sides of the front of the neck, the bracket 102 fits the back of the neck, the gasket 103 fits the skin of the front of the neck, the display screen 501 can display the working status, the rechargeable battery 401 provides power to the monitoring device, and the charging interface 402 and the external power adapter can be used for charging. When the wearer coughs, the sound sensor 2 can collect the cough sound, and the sound sensor 2 obtains the monitoring result after sample preprocessing, feature extraction, feature selection, and training and testing the collected audio information through the data processing unit 3. The cloud storage of the storage module 302 contains cough audio information and related symptom cases. The DSP chip 301 uses the original data to establish a training model in advance, and the training model is used to collect the collected audio information. The collected cough sounds are analyzed, the real-time cough sounds collected by the acoustic sensor 2 are automatically extracted by VGGish, and then the extracted audio information is classified by the Logistic regression algorithm. After classification, the cough information is analyzed by the established training model, so that the cough state can be monitored and recorded, and the cough information is recorded in the hard disk of the storage module 302. The monitored cough results can be transmitted to the external mobile phone APP through the communication module 303 using Bluetooth or WiFi. When the patient has a follow-up visit, the cough data of the patient during the recovery period can be exported through the output interface 304, or queried through the mobile phone APP. When the cough is more serious, the buzzer 503 will sound an alarm and the indicator light 502 will light up red, which has reminded the patient and the nursing staff.
[0041] Example 2
[0042] Reference Figure 1 and 2 , which is the second embodiment of the present invention, and this embodiment is based on the previous embodiment.
[0043] In this embodiment, the DSP chip 301 includes feature extraction, machine learning classification and model training. Feature extraction is used to extract cough feature information, machine learning classification is used to classify and process cough audio, and model training is used for cough testing.
[0044] The storage module 302 includes a storage hard disk and a cloud storage. The storage module 302 is used to store historical monitoring information and model training data. The communication module 303 is connected to an external mobile phone APP via Bluetooth or WiFi.
[0045] The housing assembly 1 further includes a bracket 102 and a gasket 103 . Two brackets 102 and two gaskets 103 are provided. The bracket 102 is fixed to both sides of the rear end of the housing 101 , and the gasket 103 is fixed to both sides of the front end of the housing 101 .
[0046] like Figure 1 and 2 As shown, the shell 101 is made of elastic plastic, which can fit the curvature of the neck well. The bracket 102 and the gasket 103 can provide support for the neck and reduce the friction between the neck and the monitoring device. The bracket 102 and the gasket 103 can be made of silicone material, which can effectively improve the wearing comfort. The storage module 302 can store the monitored data and monitoring model, so that the sound vibration frequency and decibel parameters can be collected in real time, and the parameters and the time stamp of the collection time are packaged and stored in the hard disk of the storage module 302. When the storage space is full, the new data collected will overwrite the oldest data. The data can be charged by connecting the charging interface 402 through an external power adapter. During the charging process, the monitoring state is off and the indicator light 502 flashes red. When the rechargeable battery 401 is fully charged, the indicator light 502 is green. When the monitoring device fails, the indicator light 502 lights up yellow. The cough monitoring data stored in the storage module 302 can be exported by connecting the output interface 304 through a wire, so as to facilitate the review during the follow-up visit. If the decibel or cough frequency of the cough exceeds the warning value, an alarm message will be sent to the handheld device or the data acquisition computer.
[0047] Example 3
[0048] Reference Figure 3 and 4 , which is the third embodiment of the present invention, and this embodiment is based on the previous two embodiments.
[0049] In this embodiment, the Logistic regression algorithm uses the sigmoid activation function to convert the predicted value z generated by linear regression into T +b is mapped to a y value between 0 and 1, and the sigmoid function is calculated as follows:
[0050]
[0051] This gives the Logistic regression formula:
[0052]
[0053] Then the value of the prediction function represents the probability of y=1, so the probabilities of y=1 and y=0 are:
[0054]
[0055] The input data is substituted into z and then mapped to between 0 and 1 by the sigmoid function. After the numerical value is obtained, it is classified. Those greater than 0.5 are classified into the positive class, otherwise they are classified into the negative class.
[0056] The model training steps are as follows:
[0057] First, preprocessing is performed. Each input cough record is divided into 6-second audio blocks, padded as needed, and processed with MFCC packages. The CNN structure consists of three parallel ResNet50s. The 7×7×2048 4-dimensional tensor output layer of each ResNet50 model is connected in parallel. These ResNet50s are not pre-trained.
[0058] The outputs of these three concatenated tensors are then merged together using a 2D layer of GlobalAveragePooling, using a deep neural network layer of 1024 neurons activated with ReLU;
[0059] Finally, the binary classification layer of the Sigmoid activation function is used. In addition to the patient's cough sound data, the network can also be trained using three types of biomarkers: muscle degeneration, vocal cord changes, and emotional changes. These three types of biomarker features are added to the CNN network. Through the feature of muscle fatigue and degeneration, a memory decay model is introduced. By using M = (I x )=Poisson(λ)I X The Poisson mask in modifies the input signal for all training and test sets, and then multiplies this value by the parameter I x and λ are randomly distributed to calculate the cough recording MFCC point I x The Poisson mask of , where λ is the average value of all values in MFCC, and the Poisson distribution formula is:
[0060]
[0061] According to the vocal cord change characteristics, a ResNet50 network with an input shape of (300, 200) from MFCC is trained, and LibriSpeech is used for distinction. By introducing emotional change features, an emotional speech classifier model is trained to learn the emotional features on the RAVDESS speech dataset.
[0062] like Figure 3 and 4 As shown, firstly, a training model is established, and a big data model of the relationship between cough sound wave characteristics and cough properties, disease location, and TCM syndrome is established to achieve preliminary diagnosis of diseases by hearing sounds and treatment based on syndrome differentiation. The historical cough audio and related cough cases are analyzed and the model is established through feature extraction, feature selection, and learning classification. After the model is established, sound is first collected through the sound-sensitive sensor 2, and the collected cough sound is sample preprocessed through the DSP chip 301. The preprocessing is performed by denoising, filtering, and normalizing the cough audio, and then feature extraction is performed on the preprocessed cough audio to extract the frequency, intensity, duration, and spectral characteristics of the cough sound. The most representative and discriminative features are selected from the extracted features to reduce the dimension of the data and improve the efficiency of the model. The selected features and labeled training data are used to train the classification model, and the newly collected signals are analyzed and classified to achieve cough monitoring.
[0063] A method for using a portable all-weather cough monitoring device, the method comprising the following steps:
[0064] a) First, the housing assembly 1 is worn on the neck, the front end of the housing 101 is located on both sides of the front of the neck, the bracket 102 fits the back of the neck, the gasket 103 fits the skin of the front of the neck, the display screen 501 can display the working status, the rechargeable battery 401 provides power to the monitoring device, and the charging interface 402 and the external power adapter can be used for charging. When the wearer coughs, the acoustic sensor 2 can collect the coughing sound;
[0065] b) The acoustic sensor 2 obtains the monitoring result after performing sample preprocessing, feature extraction, feature selection, and training test on the collected audio information through the data processing unit 3. The cloud storage of the storage module 302 stores the cough audio information and related symptom cases. The DSP chip 301 uses the original data to establish a training model in advance, and analyzes the collected cough sound through the training model. The real-time cough sound collected by the acoustic sensor 2 is automatically extracted through VGGish, and then the extracted audio information is classified through the Logistic regression algorithm. After classification, the cough information is analyzed through the established training model, so that the cough state can be monitored and recorded, and the cough information is recorded in the hard disk of the storage module 302;
[0066] c) The monitored cough results can be transmitted to an external mobile phone APP via Bluetooth or WiFi through the communication module 303. When the patient returns for a follow-up visit, the cough data of the patient during the recovery period can be exported through the output interface 304, or queried through the mobile phone APP. When the cough is more serious, the buzzer 503 will sound an alarm and the indicator light 502 will light up red to remind the patient and the nursing staff.
[0067] The standard parts used in this application document can all be purchased from the market and can be customized according to the description in the specification and drawings. The specific connection methods of each part adopt conventional means such as mature bolts, rivets, welding, etc. in the prior art. The machinery, parts and equipment all adopt conventional models in the prior art. The control method is automatic control through a controller. The control circuit of the controller can be implemented through simple programming by technicians in this field, which is common knowledge in the field. This application is mainly used to protect mechanical devices, so this application no longer explains the control method and circuit connection in detail.
[0068] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0069] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A portable all-weather cough monitoring device, comprising a housing component (1), characterized in that: The housing component (1) comprises a housing (101), a sound sensor (2), a data processing unit (3), a power management unit (4) and an abnormal alarm unit (5) are installed on the surface of the housing (101), the number of the sound sensors (2) is two, and they are installed at two ends of the housing (101) respectively, the data processing unit (3) and the power management unit (4) are both installed in the inner cavity of the housing (101), the data processing unit (3) comprises a DSP chip (301), a storage module (302), a communication module (303) and an output interface (304), the DSP chip (301) is used to process data, the storage module (302) is used to store data, the communication module (303) and the output interface (304) are used to store data, and the communication module (303) and the output interface (304) are used to store data. ) is used to transmit data, the power management unit (4) comprises a rechargeable battery (401) and a charging interface (402), the rechargeable battery (401) is used to provide power to the acoustic sensor (2), the data processing unit (3) and the abnormal alarm unit (5), the abnormal alarm unit (5) comprises a display screen (501), an indicator light (502) and a buzzer (503), the display screen (501) is fixed to one side of the surface of the shell (101), the indicator light (502) and the buzzer (503) are both fixed to the other side of the surface of the shell (101), the output end of the acoustic sensor (2) is connected to the input end of the data processing unit (3), and the output end of the data processing unit (3) is connected to the input end of the abnormal alarm unit (5).
2. A portable all-weather cough monitoring device according to claim 1, characterized in that: The DSP chip (301) includes feature extraction, machine learning classification and model training, wherein the feature extraction is used to extract cough feature information, the machine learning classification is used to classify and process cough audio, and the model training is used for cough testing.
3. A portable all-weather cough monitoring device according to claim 1, characterized in that: The storage module (302) includes a storage hard disk and a cloud storage, and the storage module (302) is used to store historical monitoring information and model training data. The communication module (303) is connected to an external mobile phone APP via Bluetooth or WiFi.
4. A portable all-weather cough monitoring device according to claim 2, characterized in that: The feature extraction uses VGGish to automatically extract the collected cough audio, the machine learning classification uses a Logistic regression algorithm, and the model training uses a convolutional neural network learning model.
5. A portable all-weather cough monitoring device according to claim 1, characterized in that: The housing component (1) further comprises a bracket (102) and a gasket (103), wherein two of each of the bracket (102) and the gasket (103) are provided, wherein the bracket (102) is fixed to two sides of the rear end of the housing (101), and the gasket (103) is fixed to two sides of the front end of the housing (101).
6. A portable all-weather cough monitoring device according to claim 4, characterized in that: The Logistic regression algorithm uses the sigmoid activation function to convert the predicted value z generated by linear regression into T +b is mapped to a y value between 0 and 1, and the sigmoid function is calculated as follows: This gives the Logistic regression formula: Then the value of the prediction function represents the probability of y=1, so the probabilities of y=1 and y=0 are: The input data is substituted into z and then mapped to between 0 and 1 by the sigmoid function. After the numerical value is obtained, it is classified. Those greater than 0.5 are classified into the positive class, otherwise they are classified into the negative class.
7. A portable all-weather cough monitoring device according to claim 4, characterized in that: The model training steps are as follows: First, preprocessing is performed. Each input cough record is divided into 6-second audio blocks, padded as needed, and processed with MFCC packages. The CNN structure consists of three parallel ResNet50s. The 7×7×2048 4-dimensional tensor output layer of each ResNet50 model is connected in parallel. These ResNet50s are not pre-trained. The outputs of these three concatenated tensors are then merged together using a 2D layer of GlobalAveragePooling, using a deep neural network layer of 1024 neurons activated with ReLU; Finally, the binary classification layer of the Sigmoid activation function is used. In addition to the patient's cough sound data, the network can also be trained using three types of biomarkers: muscle degeneration, vocal cord changes, and emotional changes. These three types of biomarker features are added to the CNN network. Through the feature of muscle fatigue and degeneration, a memory decay model is introduced. By using M = (I x )=Poisson(λ)I X The Poisson mask in modifies the input signal for all training and test sets, and then multiplies this value by the parameter I x and λ are randomly distributed to calculate the cough recording MFCC point I x The Poisson mask of , where λ is the average value of all values in MFCC, and the Poisson distribution formula is: According to the vocal cord change characteristics, a ResNet50 network with an input shape of (300, 200) from MFCC is trained, and LibriSpeech is used for distinction. By introducing emotional change features, an emotional speech classifier model is trained to learn the emotional features on the RAVDESS speech dataset.
8. A method for using a portable all-weather cough monitoring device according to any one of claims 1 to 7, characterized in that: The method of use includes the following steps: a) firstly, the housing assembly (1) is worn on the neck, the front end of the housing (101) is located on both sides of the front of the neck, the bracket (102) fits against the back of the neck, the pad (103) fits against the skin of the front of the neck, the display screen (501) can display the working status, the rechargeable battery (401) provides power to the monitoring device, and the charging operation can be performed through the charging interface (402) and the external power adapter, and when the wearer coughs, the acoustic sensor (2) can collect the coughing sound; b) the acoustic sensor (2) performs sample preprocessing, feature extraction, feature selection, and training and testing on the collected audio information through the data processing unit (3) to obtain a monitoring result; the cloud storage of the storage module (302) stores cough audio information and related symptom cases; the DSP chip (301) uses the original data to establish a training model in advance, and analyzes the collected cough sound through the training model; the real-time cough sound collected by the acoustic sensor (2) is automatically extracted through VGGish, and then the extracted audio information is classified through the Logistic regression algorithm; after classification, the cough information is analyzed through the established training model, so that the cough state can be monitored and recorded, and the cough information is recorded in the hard disk of the storage module (302); c) The monitored cough results can be transmitted to an external mobile phone APP via Bluetooth or WiFi through the communication module (303). When the patient returns for a follow-up visit, the cough data of the patient during the recovery period can be exported through the output interface (304) or queried through the mobile phone APP. When the cough is more serious, the buzzer (503) will sound an alarm and the indicator light (502) will light up red to alert the patient and the nursing staff.