Cough recognition system based on sleep process monitoring

Abstract

The invention discloses a cough recognition system based on sleep process monitoring, which relates to the technical field of sleep monitoring and comprises a sensor control subsystem, a signal processing subsystem, an auxiliary data integration subsystem, a cough recognition subsystem and a data storage subsystem. The sensor control subsystem is used for directly controlling each unit sensor, collecting data signals collected by each sensor in a unified manner, performing signal data processing on the data signals, identifying cough signals by adopting corresponding algorithms according to different data types, and realizing monitoring operation in a sleep process; and the sensor control subsystem is used for selecting, setting and acquiring signals of the sensors. According to the cough identification system, automatic classification of different types of coughs is realized from the subtle change, the sensitivity and specificity of a cough monitoring system are improved, the interference of the environment on cough signal identification is reduced, and the cough identification system can be really applied to clinical diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep monitoring, and specifically to a cough recognition system based on sleep process monitoring. Background Art

[0002] Cough is a common symptom. Frequent or chronic cough may indicate the presence of a disease, and persistent cough at night will affect the sleep quality of patients, making them lethargic during the day. Therefore, effectively quantifying characteristics such as the occurrence frequency and intensity of cough can enable doctors to diagnose diseases more accurately, which is also a key point in current cough research.

[0003] In the current sleep environment, there may be various noises and interference signals, such as external noises and signals generated by actions such as turning over of the human body. These may all affect the accurate acquisition and recognition of cough signals by sensors. Moreover, there are differences in cough sounds and breathing patterns among different people, which requires the cough recognition system to have good adaptability and generalization ability to accurately identify the cough conditions of different individuals. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a cough recognition system based on sleep process monitoring, which solves the problems raised in the above background art.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A cough recognition system based on sleep process monitoring includes a sensor control subsystem, a signal processing subsystem, an auxiliary data integration subsystem, a cough recognition subsystem, and a data storage subsystem;

[0006] The sensor control subsystem directly controls each unit sensor, uniformly collects the data signals collected by each sensor, and performs signal data processing on them. According to different data types, corresponding algorithms are used to identify cough signals to achieve the monitoring operation during the sleep process;

[0007] The sensor control subsystem is used for the selection, setting, and signal acquisition operations of sensors, including a photoplethysmography sensor, an acceleration sensor, a pressure sensor, and an acoustic sensor;

[0008] The photoplethysmography sensor is worn on the body or placed near the body to monitor the change in blood volume to detect the physical actions of cough, including inhalation, exhalation against a closed glottis, opening the glottis, or relaxation, and is used to collect non-invasive signals corresponding to physiological data;

[0009] The signal of the acceleration sensor is associated with the body sign acceleration data of the user to assist in detecting cough actions; and the acceleration sensor is also a component of the sensor control subsystem, which is used to record and collect the acceleration and vibration signals of the abdomen during cough;

[0010] The pressure sensor is an equally spaced pressure sensing element built into the mattress, which is used to measure various human physiological signals in the natural lying position of the user, including chest impact, leg impact and respiratory wave. The biocompatible sensor array is connected to the signal acquisition circuit and transmits the sleep posture and respiratory state data to the signal processing subsystem in real time wirelessly. The working pressure monitoring range of the pressure sensing element is 20 Pa to 30 kPa;

[0011] The acoustic sensor is used to collect cough sound signals. Coughing produces specific sounds, and the acoustic sensor is introduced to directly collect cough sound signals. The acoustic sensor converts the sound signals into electrical signals, and then identifies the cough by extracting and analyzing the characteristics of the electrical signals, including the frequency, amplitude and duration of the sound. Combining with machine learning algorithms, a large number of cough sound samples are trained to improve the accuracy of cough recognition;

[0012] The signal processing subsystem preprocesses the original signals of the mattress-type physiological signal monitoring system and various data signals collected in the sensor control subsystem. Combining the working principle of the digital filter and the actual application needs, a zero-phase filter is selected and designed for filtering to obtain an output signal with precise zero-phase distortion. The original signals of the upper body and the lower body are decomposed by using wavelet analysis method to obtain information in different frequency ranges, and signals that meet the actual application are selected;

[0013] The auxiliary data integration subsystem. The auxiliary data includes any information related to the occurrence and / or detection of coughs, providing the background for collecting non-invasive signals at one time, thereby allowing modification or temporary suspension of cough detection operations. The integration of the auxiliary data is used to change the cough detection status to best adapt to the measurement conditions, adjust the detection parameters to increase or decrease the sensitivity, or suspend the detection work when the measurement conditions are poor or the interference is too high. It can also classify the detected coughs contextually;

[0014] The cough recognition subsystem identifies the data of the physical actions indicating coughs through signal amplitude thresholding, signal amplitude deviation outside the statistical benchmark, time-domain analysis method, frequency-domain analysis method, signal decomposition, statistical method or machine learning method. Matching recognition algorithms are designed respectively for the signal monitoring results of the sensor control subsystem, making full use of the unique characteristics of acceleration, respiratory wave, vibration signal and sound envelope signal during coughing to identify cough events from interference events, and the interference events include throat clearing, speaking, breathing sounds;

[0015] The data storage subsystem is used to store and manage the monitoring data during the user's sleep and the relevant data of the cough analysis results, and packs and stores the data in equal time units.

[0016] Optionally, the sensor further includes a millimeter-wave radar, which uses millimeter waves with a working frequency band of 30 - 300 GHz for detection and ranging, detects the breathing movement of the human body, and has short-pulse time-domain characteristics, which can improve the detection distance and spatial resolution. By extracting the characteristics of the sleep signal, including the breathing movement cycle and breathing depth, the sleep apnea situation of the subject can be obtained.

[0017] Optionally, the auxiliary data includes physiological state data, environmental data, and measurement condition data.

[0018] Optionally, the physiological state data covers whether the user has exercised recently, whether the user is sick or the information about recovering from illness; if the user has exercised recently, there is a possibility of coughing related to exercise-induced asthma, and such coughing has different characteristics from the coughing caused by the onset of the disease; secondly, understanding the physiological state of the user can assist the system to more accurately identify the type of cough and adjust the detection parameters to adapt to the coughing characteristics under different physiological conditions; when the user is sick, the coughing characteristics will change due to the change of the disease condition, and the system will increase the detection sensitivity of the coughing related to the disease according to the real-time information.

[0019] Optionally, the environmental data includes potential hazards in the environment where the user is located, including air pollution, pollen concentration, and dust pollution; environmental factors have an important impact on the occurrence of coughing, and exposure to a polluted environment or a high-pollen environment can cause coughing; after the system integrates the environmental data, it can associate the coughing with environmental factors to help identify the coughing caused by environmental hazards; in areas with severe air pollution, if it is detected that the coughing frequency of the user increases, the environmental data can be combined to determine whether the coughing is related to air pollution.

[0020] Optionally, the measurement condition data is related to the change in signal quality caused by changing the measurement conditions; during the sleep monitoring process, the position and posture of the device will affect the signal quality; the system will modify / suspend the cough detection technology according to the measurement condition data. For example, when the signal quality is poor or interfered, the detection will be suspended to avoid misjudgment, and the detection will be resumed after the conditions improve.

[0021] Optionally, a mattress with a pressure sensor as the core component; the patient can measure a variety of human physiological signals in a natural lying position without wearing any instrument; the system can successfully separate the chest impact, leg impact, and breathing wave from the original signal through preprocessing the original signal; in the filtering link, according to the actual application needs, a zero-phase filter is selected and designed for filtering to obtain an output signal with precise zero-phase distortion; the wavelet analysis method is used to decompose the original signals of the upper body and the lower body to obtain information in different frequency ranges.

[0022] Optionally, the acoustic sensor consists of a belt for fixation, an acceleration sensor, a microphone, a pressure sensor on the inner side of the belt, and a case for collecting and storing data; it is used to record the acceleration, vibration signal of the abdomen and the respiratory wave during coughing, and extract the envelope of the collected sound as the identification feature of the coughing event; perform envelope extraction processing on the sound signal collected by the microphone.

[0023] Optionally, the cough recognition adopts the Convolutional Neural Network (CNN) algorithm. It uses a Butterworth high-pass filter for preprocessing, then uses MFCC for feature extraction, and then uses the training data set of cough features to complete the classification of cough sounds through an improved CNN; this model is superior to the existing cough detection models in terms of F1 score, sensitivity, specificity, precision and accuracy; it can effectively distinguish productive coughs and non-productive coughs, thereby detecting deformed lung function and promoting the diagnosis of pneumonia.

[0024] The present invention provides a cough recognition system based on sleep process monitoring, which has the following beneficial effects:

[0025] This cough recognition system based on sleep process monitoring starts from its subtle changes to realize the automatic classification of different types of coughs, improves the sensitivity and specificity of the cough monitoring system, and reduces the interference of the environment on the recognition of cough signals; among them, the auxiliary data can help the system more accurately recognize coughs under the background of human activities, potential diseases, environmental conditions, etc.; by combining physiological and non-physiological auxiliary data, the system can exclude signal interference caused by some non-cough factors and improve the detection specificity; after the user exercises, changes in heart rate and respiration may generate signals similar to coughs. At this time, by combining exercise information, the system can more accurately judge whether it is a cough, and the system can contextualize and classify the detected coughs; for example, recognize coughs that can indicate the onset or deterioration of diseases, asthma coughs caused by exercise, coughs caused by environmental hazards, etc.; by classifying coughs, the system can provide more targeted health advice and warnings for users; if it is detected that the cough may be related to the onset of a disease, the system can remind the user to seek medical attention in time. Detailed implementation

[0026] Next, the technical solutions in the embodiments of the present invention will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0027] The present invention provides a technical solution: a cough recognition system based on sleep process monitoring, including a sensor control subsystem, a signal processing subsystem, an auxiliary data integration subsystem, a cough recognition subsystem, and a data storage subsystem;

[0028] The sensor control subsystem directly controls each unit sensor, uniformly collects the data signals collected by each sensor, processes the signal data, and uses corresponding algorithms to identify cough signals according to different data types to achieve monitoring operations during sleep;

[0029] The sensor control subsystem is used for the selection, setting, and signal acquisition operations of sensors, including a photoplethysmography (PPG) sensor, an acceleration sensor, a pressure sensor, and an acoustic sensor;

[0030] The photoplethysmography (PPG) sensor is worn on the body or placed near the body to monitor changes in blood volume to detect the physical actions of coughing, including inhalation, exhalation against a closed glottis, opening the glottis, or relaxation, and is used to collect non-invasive signals corresponding to physiological data;

[0031] The signal of the acceleration sensor is associated with the body sign acceleration data of the user to assist in detecting coughing actions; and the acceleration sensor is also a component of the sensor control subsystem, which is used to record and collect the acceleration and vibration signals of the abdomen during coughing;

[0032] The pressure sensor is set as a pressure sensing element evenly placed inside the mattress. When the user is in a natural lying position, it is used to measure various human physiological signals, including chest impact, leg impact, and respiratory waves. The biocompatible sensor array is connected to the signal acquisition circuit to wirelessly transmit the sleeping posture and respiratory state data to the signal processing subsystem in real time. The working pressure monitoring range of the pressure sensing element is 20 Pa to 30 kPa;

[0033] The acoustic sensor is used to collect cough sound signals. Coughing produces specific sounds, and the acoustic sensor is introduced to directly collect cough sound signals; the acoustic sensor converts the sound signal into an electrical signal, and then identifies the cough by extracting and analyzing the characteristics of the electrical signal, including the frequency, amplitude, and duration of the sound; and combined with machine learning algorithms, a large number of cough sound samples are trained to improve the accuracy of cough recognition;

[0034] The sensor also includes a millimeter-wave radar, which uses millimeter waves with a working frequency band of 30 - 300 GHz for detection and ranging, detects the respiratory movement of the human body, and has short-pulse time-domain characteristics, which can improve the detection distance and spatial resolution. By extracting the characteristics of the sleep signal, including the respiratory movement cycle and respiratory depth, the sleep apnea situation of the subject can be obtained;

[0035] The signal processing subsystem preprocesses the original signals of the mattress-type physiological signal monitoring system and various data signals collected in the sensor control subsystem. Combining the working principle of digital filters and the actual application requirements, a zero-phase filter is selected and designed for filtering to obtain an output signal with precise zero-phase distortion. The original signals of the upper body and the lower body are decomposed using wavelet analysis to obtain information in different frequency ranges, and signals that meet the actual application are selected.

[0036] The auxiliary data integration subsystem. The auxiliary data includes any information related to the occurrence and / or detection of coughs, providing the background for collecting non-invasive signals at one time, thus allowing the modification or temporary suspension of cough detection operations. The integration of auxiliary data is used to change the cough detection status to best adapt to the measurement conditions, adjust the detection parameters to increase or decrease the sensitivity, or suspend the detection work when the measurement conditions are poor or the interference is too high. It can also classify the detected coughs contextually.

[0037] The auxiliary data includes physiological state data, environmental data, and measurement condition data.

[0038] The physiological state data covers whether the user has exercised recently, whether the user is ill or information about recovering from an illness. If the user has exercised recently, there is a possibility of cough related to exercise-induced asthma, and such coughs have different characteristics from those caused by disease attacks. Secondly, understanding the user's physiological state can assist the system in more accurately identifying the cough type and adjusting the detection parameters to adapt to the cough characteristics under different physiological conditions. When the user is ill, the cough characteristics will change due to the change of the illness condition, and the system will increase the detection sensitivity for disease-related coughs according to the real-time information.

[0039] The environmental data includes potential hazards in the environment where the user is located, including air pollution, pollen concentration, and dust pollution. Environmental factors have an important impact on the occurrence of coughs. Exposure to a polluted environment or a high-pollen environment can cause coughs. After the system integrates the environmental data, it can associate coughs with environmental factors to help identify coughs caused by environmental hazards. In areas with severe air pollution, if an increase in the user's cough frequency is detected, the environmental data is combined to determine whether the cough is related to air pollution.

[0040] The measurement condition data is related to the change in signal quality caused by changing the measurement conditions. During sleep monitoring, the position and posture of the device will affect the signal quality. The system modifies / suspends the cough detection technology according to the measurement condition data. For example, when the signal quality is poor or interfered, the detection is suspended to avoid misjudgment, and the detection is resumed after the conditions improve.

[0041] The cough recognition subsystem identifies data of physical actions indicating coughs through signal amplitude thresholding, statistical deviation of signal amplitudes outside the benchmark, time-domain analysis methods, frequency-domain analysis methods, signal decomposition, statistical methods, or machine learning methods; designs supporting recognition algorithms respectively for the signal monitoring results of the sensor control subsystem, makes full use of the unique features of acceleration, respiratory waves, vibration signals, and sound envelope signals during coughs, and identifies cough events from interference events, where the interference events include throat clearing, speaking, and breathing sounds;

[0042] A mattress with a pressure sensor as the core component; a patient can measure various human physiological signals in a natural lying position without wearing any instrument devices; this system can successfully separate chest impacts, leg impacts, and respiratory waves from the original signals through preprocessing of the original signals; in the filtering link, combined with the actual application needs, a zero-phase filter is selected and designed for filtering to obtain an output signal with precise zero-phase distortion; the original signals of the upper body and the lower body are decomposed using wavelet analysis to obtain information in different frequency ranges;

[0043] The acoustic sensor consists of a belt for fixation and an acceleration sensor, a microphone, a pressure sensor, and a case for collecting and storing data inside the belt; it is used to record the acceleration, vibration signals, and respiratory waves of the abdomen during coughs, and extract the envelope of the collected sound as an identification feature for cough events; perform envelope extraction processing on the sound signals collected by the microphone;

[0044] Cough recognition uses a convolutional neural network (CNN) algorithm, performs preprocessing using a Butterworth high-pass filter, then extracts features using MFCC, and then uses a training dataset of cough features to complete the classification of cough sounds through an improved CNN; this model is superior to existing cough detection models in terms of F1 score, sensitivity, specificity, precision, and accuracy; effectively distinguish productive coughs and non-productive coughs, thereby detecting deformed lung function and promoting the diagnosis of pneumonia;

[0045] The data storage subsystem is used to store and manage the monitoring data during the user's sleep and the relevant data for cough analysis results, and pack and store the data in equal time units.

[0046] As described above, it is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. A cough recognition system based on sleep process monitoring, characterized in that: It includes sensor control subsystem, signal processing subsystem, auxiliary data integration subsystem, cough recognition subsystem and data storage subsystem; The sensor control subsystem is used to directly control each unit sensor, collect the data signals collected by each sensor, and process the signal data. The corresponding algorithm is used according to different data types to identify cough signals and realize monitoring operations during sleep. The sensor control subsystem is used for the selection, setting, and signal acquisition of sensors, including photoelectric solvent mapping sensors, acceleration sensors, pressure sensors, and acoustic sensors; Photoplethysmographic sensors are worn on or placed near the body to monitor changes in blood volume to detect the physical motion of a cough, including inhalation, exhalation against a closed glottis, opening the glottis, or relaxation, for collecting non-invasive signals corresponding to physiological data; The signal of the acceleration sensor is associated with the user's physical acceleration data to assist in detecting coughing movements; And the acceleration sensor is also a component of the sensor control subsystem, which is used to record and collect the acceleration and vibration signals of the abdomen during coughing; The pressure sensor is set as a pressure sensing element equidistantly built into the mattress, which is used to measure a variety of human physiological signals, including chest impact, leg impact and respiratory wave, when the user is lying in a natural position. The sensor array with biocompatibility is connected to the signal acquisition circuit to transmit the sleeping posture and respiratory state data to the signal processing subsystem in real time wirelessly. The working pressure monitoring range of the pressure sensing element is 20Pa to 30kPa; Acoustic sensors are used to collect cough sound signals. Coughing produces specific sounds. Acoustic sensors are introduced to directly collect cough sound signals. Acoustic sensors convert sound signals into electrical signals, and then identify coughs by extracting and analyzing the features of the electrical signals, including the frequency, amplitude, and duration of the sound. In addition, they are combined with machine learning algorithms to train a large number of cough sound samples to improve the accuracy of cough recognition. The signal processing subsystem pre-processes the original signal of the mattress-type physiological signal monitoring system and the various data signals collected in the sensor control subsystem. In combination with the working principle of the digital filter and the actual application needs, a zero-phase filter is selected and designed for filtering to obtain an output signal with precise zero phase distortion; The original signals of the upper body and lower body are decomposed by wavelet analysis to obtain information in different frequency ranges and select signals that meet practical applications; Auxiliary data integration subsystem, auxiliary data includes any information related to the occurrence and / or detection of cough, which provides the background of the collection of non-invasive signal conditions, thereby allowing the cough detection operation to be modified or temporarily suspended. The integration of auxiliary data is used to change the cough detection state to best adapt it to the measurement conditions, adjust the detection parameters to increase or decrease the sensitivity, or suspend the detection work when the measurement conditions are poor or the confounding is too high, and can also contextualize and classify the detected cough; The cough recognition subsystem identifies data indicating the physical action of coughing by signal amplitude thresholding, signal amplitude deviation outside the statistical benchmark, time domain analysis method, frequency domain analysis method, signal decomposition, statistical method or machine learning method; a matching recognition algorithm is designed for the signal monitoring results of the sensor control subsystem, making full use of the unique characteristics of acceleration, respiratory waves, vibration signals and sound envelope signals during coughing, and identifying coughing events from interference events, including throat clearing, talking, and breathing sounds; The data storage subsystem is used to store and manage the monitoring data of the user during sleep and the relevant data of the cough analysis results, and package and store the data in equal time units.

2. A cough recognition system based on sleep process monitoring according to claim 1, characterized in that: The sensor also includes a millimeter wave radar, which uses millimeter waves with an operating frequency band of 30-300GHz for detection and ranging, detects the respiratory movement of the human body, and has short pulse time domain characteristics, which can improve the detection distance and spatial resolution. By extracting the characteristics of the sleep signal, including the respiratory movement cycle and breathing depth, the sleep apnea condition of the subject is obtained.

3. A cough recognition system based on sleep process monitoring according to claim 1, characterized in that: The auxiliary data includes physiological state data, environmental data, and measurement condition data.

4. A cough recognition system based on sleep process monitoring according to claim 3, characterized in that: The physiological status data covers information about whether the user has exercised recently, whether he is sick or recovered from an illness; if the user has exercised recently, there is a possibility of inducing cough related to exercise-induced asthma, and the characteristics of such cough are different from those caused by an illness attack; secondly, understanding the user's physiological status can assist the system in more accurately identifying the type of cough and adjusting the detection parameters to adapt to the cough characteristics under different physiological conditions; when the user is sick, the cough characteristics will change due to changes in the condition, and the system will improve the detection sensitivity of disease-related coughs based on real-time information.

5. A cough recognition system based on sleep process monitoring according to claim 3, characterized in that: The environmental data includes potential hazards in the user's environment, including air pollution, pollen concentration, and dust pollution. Environmental factors have an important impact on the occurrence of coughs, and exposure to polluted environments or high-pollen environments causes coughs. After the system integrates the environmental data, it can associate coughs with environmental factors to help identify coughs caused by environmental hazards. In areas with severe air pollution, if an increase in the user's coughing frequency is detected, the environmental data is combined to determine whether the cough is related to air pollution.

6. A cough recognition system based on sleep process monitoring according to claim 3, characterized in that: The measurement condition data is related to the change in signal quality caused by changing the measurement conditions; during the sleep monitoring process, the position and posture of the device will affect the quality of the signal; the system modifies / suspends the cough detection technology based on the measurement condition data. If the signal quality is poor or interfered with, the detection is suspended to avoid misjudgment, and the detection is resumed when the conditions improve.

7. The cough recognition system based on sleep process monitoring according to claim 1, characterized in that: The mattress with the pressure sensor as the core component can measure a variety of human physiological signals in a natural lying position without wearing any instrument or device. The system can successfully separate chest impact, leg impact and respiratory wave from the original signal by preprocessing the original signal. In the filtering stage, a zero-phase filter is selected and designed for filtering in combination with actual application needs to obtain an output signal with precise zero phase distortion; the original signals of the upper and lower body are decomposed using wavelet analysis to obtain information in different frequency ranges.

8. The cough recognition system based on sleep process monitoring according to claim 1, characterized in that: The acoustic sensor is composed of a waist belt for fixing, an acceleration sensor inside the waist belt, a microphone, a pressure sensor, and a box for collecting and storing data; it is used to record the acceleration, vibration signal and respiratory wave of the abdomen during coughing, and extract the envelope of the collected sound as the identification feature of the coughing event; Extract the envelope of the sound signal collected by the microphone.

9. The cough recognition system based on sleep process monitoring according to claim 1, characterized in that: The cough recognition adopts a convolutional neural network algorithm, uses a Butterworth high-pass filter for preprocessing, then uses MFCC for feature extraction, and then uses a training data set of cough features to complete the classification of cough sounds through an improved CNN; the model is superior to existing cough detection models in terms of F1 score, sensitivity, specificity, precision and accuracy; it effectively distinguishes between productive coughs and non-productive coughs, thereby detecting deformed lung function and promoting the diagnosis of pneumonia.