Dynamic killing method based on cough audio detection and analysis
By using a deep learning model based on cough audio detection and multi-parameter judgment to dynamically adjust the ultraviolet disinfection strategy, the problems of blind disinfection and lag in response in existing technologies are solved, enabling real-time and precise disinfection in hospitals, improving infection control and saving energy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGZHOU NO 2 PEOPLES HOSPITAL
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-26
AI Technical Summary
Existing ultraviolet disinfection technology in hospitals suffers from blind disinfection, delayed response, and interference in application. It cannot respond to respiratory infection risks in a real-time and intelligent manner, resulting in unsatisfactory infection control and energy waste.
By using a dynamic disinfection method based on cough audio detection and analysis, a deep learning model is used to identify cough audio. Combined with multi-parameter judgment, precise ultraviolet disinfection triggering is achieved, and the disinfection strategy is dynamically adjusted to reduce interference with normal activities.
It enables real-time and precise disinfection response, improves infection control, saves energy, reduces equipment wear and tear, and minimizes disruption to normal activities.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent disinfection technology, specifically to a dynamic disinfection method based on cough audio detection and analysis. Background Technology
[0002] In hospitals and other medical institutions, airborne pathogens (such as aerosols carrying pathogens generated by patients coughing or sneezing) are a significant route of nosocomial cross-infection. To reduce the risk of such infections, ultraviolet disinfection technology is widely used for air and surface disinfection in areas such as wards, corridors, and waiting areas.
[0003] Currently, the mainstream ultraviolet disinfection operation mode in hospitals is mostly a fixed procedure, which is mainly reflected in two ways: First, timed start-up and shutdown control based on preset time, such as turning on the ultraviolet disinfection equipment uniformly when there is less activity at night; Second, relying on manual judgment and operation, with staff manually turning on or off the disinfection equipment based on experience or inspection results.
[0004] However, the existing disinfection methods described above have significant drawbacks, including:
[0005] (1) Blindness in disinfection: Fixed disinfection schedules or manual decision-making models cannot be correlated with the real-time and dynamically changing infection risk levels in the hospital environment. For example, during high-risk periods when patients are densely packed and sudden coughing or sneezing events are frequent, effective disinfection intervention may be lacking because the preset disinfection time has not arrived or the situation is not detected in time. Conversely, during periods when there are few people and the infection risk is low, disinfection may still be carried out according to a fixed plan, resulting in energy waste and equipment wear and tear. This mismatch between supply and demand leads to low disinfection efficiency and unsatisfactory infection control results.
[0006] (2) Lag in response: Whether relying on fixed cycles or manual inspections, existing methods are insufficient to respond quickly to sudden, localized respiratory symptom events. There is a time delay between the occurrence of a risk event and the initiation of disinfection measures, making it impossible to achieve immediate and targeted risk containment and potentially missing the critical window period for blocking pathogen transmission.
[0007] (3) Interference in application: When traditional ultraviolet disinfection equipment is working, the ultraviolet radiation emitted by it (especially the UVC band) is harmful to human skin and eyes. Therefore, it is necessary to ensure "human-machine isolation", that is, to carry out the disinfection in an unmanned environment. This safety requirement severely limits the direct application of ultraviolet disinfection during peak daytime traffic hours, in continuously occupied wards or public areas (such as registration halls and corridors), and reduces the coverage and timeliness of disinfection.
[0008] In summary, current technologies lack an integrated solution capable of real-time, intelligently sensing respiratory infection risk events (such as aerosol-generating behaviors like coughing and sneezing) in specific areas of a hospital, and automatically, accurately, and safely triggering and executing ultraviolet disinfection actions accordingly. Therefore, there is an urgent need for a novel disinfection strategy and system to overcome the aforementioned blindness, lag, and interference, and to achieve adaptive, intelligent disinfection management based on real-time risk perception. Summary of the Invention
[0009] The purpose of this invention is to provide a dynamic disinfection method based on cough audio detection and analysis. This method overcomes the shortcomings of existing technologies and provides a real-time, precise, and dynamic ultraviolet disinfection method. By correlating acoustic events such as coughing with environmental parameters, it achieves "disinfection on demand," blocking aerosol transmission routes while minimizing interference with normal medical activities.
[0010] To achieve the above functions, this invention designs a dynamic disinfection method based on cough audio detection and analysis. For the target hospital environment, the following steps S1-S4 are executed to complete the disinfection of the target hospital environment:
[0011] Step S1: Collect audio from a preset range within the target hospital environment at a preset period, and after filtering, divide it into multiple audio segments of preset duration. Based on a deep learning model, identify audio frames containing cough audio within the audio segments, and merge the audio frames containing cough audio to obtain cough audio segments.
[0012] Step S2: For each cough audio segment, based on energy peak and fundamental frequency detection, classify each cough audio segment into single cough and overlapping cough, separate overlapping coughs to obtain samples of each single cough; and use an unsupervised clustering method to associate each sample of a single cough with its corresponding individual.
[0013] Step S3: For each individual, based on the number and time interval of their corresponding single cough, determine whether it is an occasional cough. Based on the number of individuals with non-occasional coughs, further determine whether the hospital environment disinfection early warning procedure needs to be activated.
[0014] Step S4: For each environmental indicator of the target hospital, preset the corresponding disinfection trigger threshold. If it is determined in step S3 that the hospital environment disinfection early warning program needs to be activated, and the environmental indicators of the target hospital reach the corresponding disinfection trigger threshold, then activate the corresponding disinfection program to complete the disinfection of the target hospital environment.
[0015] As a preferred technical solution of the present invention, the specific steps of step S1 are as follows:
[0016] Step S1.1: Deploy sensors at preset locations within the target hospital to collect audio data within a preset range of the target hospital environment;
[0017] Step S1.2: Design a bandpass filter for the acquired audio, filter the acquired audio, retain the audio in the preset frequency band, and remove ambient noise;
[0018] Step S1.3: Frame the filtered audio, divide the audio into multiple audio segments of fixed duration, and convert each audio segment into a Log-Mel spectrogram;
[0019] Step S1.4: Input the Log-Mel spectrograms corresponding to each audio segment into the CNN deep learning model pre-trained with the sample library. The CNN deep learning model uses multi-layer convolution and pooling operations to extract features, estimate the probability that each audio frame in the audio segment contains cough audio, and finally outputs the probability that each audio frame contains cough audio through a fully connected layer, forming a cough probability curve.
[0020] Step S1.5: Mark the audio frames with a probability greater than a preset threshold in the cough probability curve as cough frames;
[0021] Step S1.6: For each cough frame, if the time interval between two adjacent cough frames is less than 0.3 seconds, they are merged to obtain a cough audio segment.
[0022] As a preferred technical solution of the present invention, the specific steps of step S2 are as follows:
[0023] Step S2.1: For each cough audio segment, detect the number of fundamental frequencies and energy peaks. If one energy peak and one fundamental frequency are detected, it is determined to be a single cough. If multiple energy peaks and multiple fundamental frequencies are detected, it is determined to be an overlapping cough. If multiple energy peaks and one fundamental frequency are detected, it is determined to be an echo of a single cough. If one single energy peak and multiple fundamental frequencies are detected, it is determined to be a fundamental frequency detection error.
[0024] Step S2.2: For overlapping coughs, separate them according to each fundamental frequency to obtain multiple separated monophonic coughs;
[0025] Step S2.3: The single cough detected in Step S2.1, the echo of the single cough, and the multiple single coughs separated in Step S2.2 are all taken as single cough samples. The Mel-frequency cepstral coefficient method is used to extract the voiceprint features of each single cough sample. For the voiceprint features of each single cough sample, clustering is performed based on the K-means algorithm to associate each single cough with its corresponding individual. The specific method is as follows:
[0026] Step A. For each sample of single coughs, randomly select k initial center points;
[0027] Step B. Traverse all samples, calculate the distance between each sample and the initial centroid, and assign each sample to the initial centroid that is closest to it to form a cluster;
[0028] Step C. Calculate the average value of each cluster and use it as the new centroid;
[0029] Step D. Repeat steps A through C until the number of clusters no longer changes. At this point, the K-means algorithm converges, and each cluster is used as the single cough corresponding to each individual.
[0030] As a preferred technical solution of the present invention, the specific steps of step S3 are as follows:
[0031] Step S3.1: For each individual, detect the number of single coughs and the time interval; if the number of single coughs within a preset time is less than a preset value, the individual is determined to have occasional coughs; otherwise, it is determined to be another situation.
[0032] Step S3.2: For individuals in other situations, if there are at least two consecutive single coughs with a time interval of less than a preset value, then the individual is marked as having a non-accidental cough.
[0033] Step S3.3: Count the number of individuals with non-accidental coughs and aerosol concentration within a preset time and preset range, calculate the aerosol concentration rise rate using a sliding window algorithm, calculate the temperature difference between the outdoor environment and the target hospital environment in real time, collect the humidity change and air pressure change of the target hospital environment within 24 hours, and normalize them.
[0034] Calculate the following formula:
[0035] ;
[0036] In the formula, Indicates the disinfection warning index. This represents the number of individuals with non-accidental coughs after normalization. This represents the normalized aerosol concentration. This represents the normalized temperature difference between the outdoor environment and the target hospital environment. This represents the normalized change in humidity in the target hospital environment over 24 hours. Changes in air pressure in the target hospital environment over 24 hours; , , , , In order , , , , Weighting coefficients;
[0037] If the disinfection warning index If the value exceeds the preset threshold, it is determined that the hospital environment disinfection early warning program needs to be activated; otherwise, it is determined that the hospital environment disinfection early warning program does not need to be activated.
[0038] As a preferred embodiment of the present invention: In step S4, if it is determined that a hospital environment disinfection early warning program needs to be activated, and any of the following conditions are met, then the corresponding disinfection program is activated:
[0039] If the PM2.5 concentration in the target hospital environment is ≥60μg / m³ and lasts for 2 hours, the medium-intensity disinfection of the ZZNM-40 disinfection device will be triggered.
[0040] If the TVOC concentration in the target hospital environment is ≥550μg / m³, then the medium-intensity disinfection of the ZZNM-40 sterilizer will be triggered.
[0041] If the current target hospital's ambient humidity is ≥58% and the hospital's internal ambient humidity is <60%, then the ZZNM-40 sterilizer will be activated for medium-intensity disinfection.
[0042] If the current hospital environment humidity is ≥60%, the ZZNM-40 sterilizer will be triggered for low-intensity continuous disinfection.
[0043] As a preferred technical solution of the present invention: if the target hospital environment is a children's activity area, the volume recognition threshold of the audio collection of the target hospital environment is reduced, the disinfection intensity is reduced, and the audio collection cycle of the target hospital environment is shortened.
[0044] If the target hospital environment is a patient inpatient area, the volume recognition threshold for audio collection in the target hospital environment should be lowered, the disinfection intensity should be increased, and the audio collection cycle of the target hospital environment should be shortened.
[0045] As a preferred technical solution of the present invention: if the target hospital environment is a doctor's work area, then medium-intensity disinfection is carried out by ZZNM-40 disinfection device during the doctor's non-working peak period;
[0046] Environmental sensors are used to monitor the flow of people in real time. When multiple doctors are detected moving around at the same time, the disinfection operation is delayed.
[0047] As a preferred technical solution of the present invention: if the target hospital environment is an outpatient area, environmental sensors are used to monitor the flow of people in real time; if it is a low-flow period, a ZZNM-40 disinfection device is used for low-intensity disinfection.
[0048] During peak periods of high foot traffic, use the ZZNM-40 sterilizer for medium-intensity or high-intensity disinfection.
[0049] The present invention also designs a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the dynamic disinfection method based on cough audio detection and analysis.
[0050] The present invention also designs a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the dynamic disinfection method based on cough audio detection and analysis.
[0051] Beneficial effects: Compared with the prior art, the advantages of the present invention include:
[0052] 1. Precision and Foresight: By taking "cough acoustic events" as the core triggering factor, directly linking them to the source of respiratory droplets and aerosols, a leap from "passive timed disinfection" to "proactive risk elimination" has been achieved.
[0053] 2. High efficiency and energy saving: By linking multiple parameters (PM2.5, TVOC, humidity) with acoustic events, ineffective disinfection is avoided. It only starts when there is a clear risk, saving energy and extending the equipment life.
[0054] 3. Data-driven and adaptive: It has learning capabilities, can optimize trigger thresholds based on historical data, and can adjust disinfection strategies based on real-time feedback (such as persistent cough) to achieve personalized and adaptive infection control. Attached Figure Description
[0055] Figure 1 This is a flowchart of a dynamic disinfection method based on cough audio detection and analysis according to an embodiment of the present invention;
[0056] Figure 2 This is a schematic diagram of the sensor deployment within a target hospital according to an embodiment of the present invention;
[0057] Figure 3 This is a schematic diagram of the disinfection settings interface provided in an embodiment of the present invention. Detailed Implementation
[0058] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.
[0059] This invention provides a dynamic disinfection method based on cough audio detection and analysis, targeting a specific hospital environment, referring to... Figure 1 Perform the following steps S1-S4 to complete the disinfection of the target hospital environment:
[0060] Step S1: Collect audio from a preset range within the target hospital environment at a preset period, and after filtering, divide it into multiple audio segments of preset duration. Based on a deep learning model, identify audio frames containing cough audio within the audio segments, and merge the audio frames containing cough audio to obtain cough audio segments.
[0061] The specific steps of step S1 are as follows:
[0062] Step S1.1: Refer to Figure 2 Sensors are deployed at predetermined locations within the target hospital to collect audio data within a predetermined range of the target hospital environment.
[0063] Step S1.2: Design a bandpass filter for the acquired audio, filter the acquired audio, retain the audio in the preset frequency band, and remove ambient noise;
[0064] Step S1.3: Frame the filtered audio, divide the audio into multiple audio segments of fixed duration, and convert each audio segment into a Log-Mel spectrogram;
[0065] Step S1.4: Input the Log-Mel spectrograms corresponding to each audio segment into the CNN deep learning model pre-trained with the sample library. The CNN deep learning model uses multi-layer convolution and pooling operations to extract features, estimate the probability that each audio frame in the audio segment contains cough audio, and finally outputs the probability that each audio frame contains cough audio through a fully connected layer, forming a cough probability curve.
[0066] Step S1.5: Mark the audio frames with a probability greater than a preset threshold in the cough probability curve as cough frames;
[0067] Step S1.6: For each cough frame, if the time interval between two adjacent cough frames is less than 0.3 seconds, they are merged to obtain a cough audio segment.
[0068] Step S2: For each cough audio segment, based on energy peak and fundamental frequency detection, classify each cough audio segment into single cough and overlapping cough, separate overlapping coughs to obtain samples of each single cough; and use an unsupervised clustering method to associate each sample of a single cough with its corresponding individual.
[0069] The specific steps of step S2 are as follows:
[0070] Step S2.1: For each cough audio segment, detect the number of fundamental frequencies and energy peaks. If one energy peak and one fundamental frequency are detected, it is determined to be a single cough. If multiple energy peaks and multiple fundamental frequencies are detected, it is determined to be an overlapping cough. If multiple energy peaks and one fundamental frequency are detected, it is determined to be an echo of a single cough. If one single energy peak and multiple fundamental frequencies are detected, it is determined to be a fundamental frequency detection error.
[0071] Step S2.2: For overlapping coughs, separate them according to each fundamental frequency to obtain multiple separated monophonic coughs;
[0072] Step S2.3: The single cough detected in Step S2.1, the echo of the single cough, and the multiple single coughs separated in Step S2.2 are all taken as single cough samples. The Mel-frequency cepstral coefficient method is used to extract the voiceprint features of each single cough sample. For the voiceprint features of each single cough sample, clustering is performed based on the K-means algorithm to associate each single cough with its corresponding individual. The specific method is as follows:
[0073] Step A. For each sample of single coughs, randomly select k initial center points;
[0074] Step B. Traverse all samples, calculate the distance between each sample and the initial centroid, and assign each sample to the initial centroid that is closest to it to form a cluster;
[0075] Step C. Calculate the average value of each cluster and use it as the new centroid;
[0076] Step D. Repeat steps A through C until the number of clusters no longer changes. At this point, the K-means algorithm converges, and each cluster is used as the single cough corresponding to each individual.
[0077] Step S3: For each individual, based on the number and time interval of their corresponding single cough, determine whether it is an occasional cough. Based on the number of individuals with non-occasional coughs, further determine whether the hospital environment disinfection early warning procedure needs to be activated.
[0078] The specific steps of step S3 are as follows:
[0079] Step S3.1: For each individual, detect the number of single coughs and the time interval; if the number of single coughs within a preset time is less than a preset value, the individual is determined to have occasional coughs; otherwise, it is determined to be another situation.
[0080] Step S3.2: For individuals in other situations, if there are at least two consecutive single coughs with a time interval of less than a preset value, then the individual is marked as having a non-accidental cough.
[0081] Step S3.3: Count the number of individuals with non-accidental coughs and aerosol concentration within a preset time and preset range, calculate the aerosol concentration rise rate using a sliding window algorithm, calculate the temperature difference between the outdoor environment and the target hospital environment in real time, collect the humidity change and air pressure change of the target hospital environment within 24 hours, and normalize them.
[0082] Calculate the following formula:
[0083] ;
[0084] In the formula, Indicates the disinfection warning index. This represents the number of individuals with non-accidental coughs after normalization. This represents the normalized aerosol concentration. This represents the normalized temperature difference between the outdoor environment and the target hospital environment. This represents the normalized change in humidity in the target hospital environment over 24 hours. Changes in air pressure in the target hospital environment over 24 hours; , , , , In order , , , , Weighting coefficients;
[0085] If the disinfection warning index If the value exceeds the preset threshold, it is determined that the hospital environment disinfection early warning program needs to be activated; otherwise, it is determined that the hospital environment disinfection early warning program does not need to be activated.
[0086] Step S4: For each environmental indicator of the target hospital, preset the corresponding disinfection trigger threshold. If it is determined in step S3 that the hospital environment disinfection early warning program needs to be activated, and the environmental indicators of the target hospital reach the corresponding disinfection trigger threshold, then activate the corresponding disinfection program to complete the disinfection of the target hospital environment.
[0087] In step S4, if it is determined that the hospital environment disinfection early warning procedure needs to be activated, and any of the following conditions are met, then the corresponding disinfection procedure will be activated:
[0088] If the PM2.5 concentration in the target hospital environment is ≥60μg / m³ and lasts for 2 hours, the medium-intensity disinfection of the ZZNM-40 disinfection device will be triggered.
[0089] If the TVOC concentration in the target hospital environment is ≥550μg / m³, then the medium-intensity disinfection of the ZZNM-40 sterilizer will be triggered.
[0090] If the current target hospital's ambient humidity is ≥58% and the hospital's internal ambient humidity is <60%, then the ZZNM-40 sterilizer will be activated for medium-intensity disinfection.
[0091] If the current hospital environment humidity is ≥60%, the ZZNM-40 sterilizer will be triggered for low-intensity continuous disinfection.
[0092] In this embodiment, if the target hospital environment is a children's activity area, the volume recognition threshold for audio collection in the target hospital environment is reduced, the disinfection intensity is reduced, and the audio collection cycle of the target hospital environment is shortened.
[0093] If the target hospital environment is a patient inpatient area, the volume recognition threshold for audio collection in the target hospital environment should be lowered, the disinfection intensity should be increased, and the audio collection cycle of the target hospital environment should be shortened.
[0094] In this embodiment, if the target hospital environment is a doctor's work area, then medium-intensity disinfection is carried out using a ZZNM-40 disinfection device during the doctor's off-peak working hours.
[0095] Environmental sensors are used to monitor the flow of people in real time. When multiple doctors are detected moving around at the same time, the disinfection operation is delayed.
[0096] In the embodiment, if the target hospital environment is an outpatient area, environmental sensors are used to monitor the flow of people in real time; if it is a low-flow period, a ZZNM-40 disinfection device is used for low-intensity disinfection.
[0097] During peak periods of high foot traffic, use the ZZNM-40 sterilizer for medium-intensity or high-intensity disinfection.
[0098] In the embodiment, if the target hospital environment is an area where the elderly are active, the ZZNM-CC (CC01) sound sensor recognition algorithm is optimized to adapt to the cough / sneezing sound characteristics of the elderly, and a progressive disinfection mode is adopted, gradually increasing the intensity through the ZZNM-40 disinfection device;
[0099] Establish a model of the activity patterns of the elderly, and carry out centralized disinfection during periods of low activity (such as nighttime) using the ZZNM-40 disinfection device.
[0100] In the embodiment, if the target hospital environment is an area where adults are active, the standard recognition threshold and disinfection intensity of the ZZNM-CC (CC01) sound sensor are maintained;
[0101] Prioritize using the ZZNM-40 sterilizer for medium-intensity or high-intensity disinfection.
[0102] The disinfection period is intelligently adjusted based on the work / rest time pattern. See the illustration of the disinfection settings interface for details. Figure 3 .
[0103] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the dynamic disinfection method based on cough audio detection and analysis.
[0104] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the dynamic disinfection method based on cough audio detection and analysis.
[0105] In one embodiment, the selection is as follows:
[0106] Sensor selection:
[0107] Environmental parameter sensors:
[0108] Model: ZZNM-AS (AS01) Multifunctional Environmental Sensor, WIFI Basic Type;
[0109] Parameters: PM2.5 measurement range 0-500μg / m³, accuracy ±10μg / m³; TVOC measurement range 0-5000ppb; temperature range -40°C~80°C; humidity range 0-100%RH.
[0110] Features: Response time <3 seconds, low power consumption design, good stability, long service life, suitable for long-term monitoring;
[0111] Applications: Real-time monitoring of airborne fine particulate matter concentration, volatile organic compounds, changes in ambient temperature and humidity, and aerosol particle size distribution; monitoring data is directly uploaded to the cloud server via mobile Wi-Fi.
[0112] Acoustic sensors:
[0113] Model: ZZNM-CC (CC01) Dual-channel breath sound acoustic recognition device (sound sensor);
[0114] Parameters: 2 channels, 48kHz sampling rate, 16-bit precision;
[0115] Features: Beamforming technology, high directional sensitivity, and strong noise suppression capability;
[0116] Applications: Accurately locate sound sources and enhance target sound recognition. The collected breathing sound data is first transmitted to an edge computing device via WIFI for anomaly detection, and then uploaded to the cloud server.
[0117] Interface unit selection:
[0118] Edge computing devices:
[0119] Model: ZZNM-S6 Pro Edge Computing and Control Unit;
[0120] Specifications: Quad-core Cortex-A35 processor, 2GB RAM / 8GB ROM, built-in Wi-Fi and Bluetooth;
[0121] Features: Supports local machine learning, highly scalable, low power consumption, high performance, and supports multiple interface protocols;
[0122] Applications: Preprocessing of respiratory sound data, sound recognition and judgment, and uploading the judgment results to the cloud server;
[0123] Mobile WiFi communication module:
[0124] Model: ZZNM-4G-WIFI (WM01) communication module;
[0125] Specifications: Supports multiple network standards, ensuring stable data transmission;
[0126] Features: Full network compatibility, supports location services, low power mode;
[0127] Application: Data transmission;
[0128] Connection relationships:
[0129] The sensor is connected to the core unit of the system;
[0130] The ZZNM-AS (AS01) multi-functional environmental sensor directly uploads data to the cloud server via the ZZNM-4G-WIFI (WM01) communication module.
[0131] The temperature and humidity sensor is integrated into the ZZNM-AS (AS01) multi-functional environmental sensor and is uploaded synchronously with its data.
[0132] The ZZNM-CC (CC01) sound sensor transmits data to the ZZNM-S6 Pro edge computing and control unit via WIFI;
[0133] Edge computing and cloud servers:
[0134] The ZZNM-S6 Pro edge computing and control unit will collect respiratory sound data after determining whether an individual's cough is accidental, and upload it to the cloud server via the ZZNM-4G-WIFI (WM01) communication module through the local network or mobile communication network; data from the multi-functional environmental sensors will also be uploaded to the server synchronously through this communication module.
[0135] Use RESTful API for data interaction;
[0136] Implement data encryption and authentication mechanisms to ensure communication security;
[0137] System integration topology:
[0138] Each region is equipped with one ZZNM-S6 Pro edge computing unit and multiple sensor nodes; the ZZNM-4G-WIFI (WM01) communication module serves as the core data transmission hub.
[0139] Multifunctional environmental sensor data is directly uploaded to the cloud via the WIFI module. Breath sound sensor data is first transmitted to the edge computing unit to determine whether an individual's cough is accidental, and then uploaded to the cloud via the WIFI module. The edge computing unit is also responsible for the preprocessing of sensor data within the area.
[0140] The cloud server acts as the system's brain, performing global rule integration calculations and disinfection decision-making, and finally issuing disinfection instructions to the ZZNM-40 disinfection unit.
[0141] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A dynamic disinfection method based on cough audio detection and analysis, characterized in that, For the target hospital environment, perform the following steps S1-S4 to complete the disinfection of the target hospital environment: Step S1: Collect audio from a preset range within the target hospital environment at a preset period, and after filtering, divide it into multiple audio segments of preset duration. Based on a deep learning model, identify audio frames containing cough audio within the audio segments, and merge the audio frames containing cough audio to obtain cough audio segments. Step S2: For each cough audio segment, based on energy peak and fundamental frequency detection, classify each cough audio segment into single cough and overlapping cough, separate overlapping coughs to obtain samples of each single cough; and use an unsupervised clustering method to associate each sample of a single cough with its corresponding individual. Step S3: For each individual, based on the number and time interval of their corresponding single cough, determine whether it is an occasional cough. Based on the number of individuals with non-occasional coughs, further determine whether the hospital environment disinfection early warning procedure needs to be activated. Step S4: For each environmental indicator of the target hospital, preset the corresponding disinfection trigger threshold. If it is determined in step S3 that the hospital environment disinfection early warning program needs to be activated, and the environmental indicators of the target hospital reach the corresponding disinfection trigger threshold, then activate the corresponding disinfection program to complete the disinfection of the target hospital environment.
2. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, The specific steps of step S1 are as follows: Step S1.1: Deploy sensors at preset locations within the target hospital to collect audio data within a preset range of the target hospital environment; Step S1.2: Design a bandpass filter for the acquired audio, filter the acquired audio, retain the audio in the preset frequency band, and remove ambient noise; Step S1.3: Frame the filtered audio, divide the audio into multiple audio segments of fixed duration, and convert each audio segment into a Log-Mel spectrogram; Step S1.4: Input the Log-Mel spectrograms corresponding to each audio segment into the CNN deep learning model pre-trained with the sample library. The CNN deep learning model uses multi-layer convolution and pooling operations to extract features, estimate the probability that each audio frame in the audio segment contains cough audio, and finally outputs the probability that each audio frame contains cough audio through a fully connected layer, forming a cough probability curve. Step S1.5: Mark the audio frames with a probability greater than a preset threshold in the cough probability curve as cough frames; Step S1.6: For each cough frame, if the time interval between two adjacent cough frames is less than 0.3 seconds, they are merged to obtain a cough audio segment.
3. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, The specific steps of step S2 are as follows: Step S2.1: For each cough audio segment, detect the number of fundamental frequencies and energy peaks. If one energy peak and one fundamental frequency are detected, it is determined to be a single cough. If multiple energy peaks and multiple fundamental frequencies are detected, it is determined to be an overlapping cough. If multiple energy peaks and one fundamental frequency are detected, it is determined to be an echo of a single cough. If one single energy peak and multiple fundamental frequencies are detected, it is determined to be a fundamental frequency detection error. Step S2.2: For overlapping coughs, separate them according to each fundamental frequency to obtain multiple separated monophonic coughs; Step S2.3: The single cough detected in Step S2.1, the echo of the single cough, and the multiple single coughs separated in Step S2.2 are all taken as single cough samples. The Mel-frequency cepstral coefficient method is used to extract the voiceprint features of each single cough sample. For the voiceprint features of each single cough sample, clustering is performed based on the K-means algorithm to associate each single cough with its corresponding individual. The specific method is as follows: Step A. For each sample of single coughs, randomly select k initial center points; Step B. Traverse all samples, calculate the distance between each sample and the initial centroid, and assign each sample to the initial centroid that is closest to it to form a cluster; Step C. Calculate the average value of each cluster and use it as the new centroid; Step D. Repeat steps A through C until the number of clusters no longer changes. At this point, the K-means algorithm converges, and each cluster is used as the single cough corresponding to each individual.
4. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, The specific steps of step S3 are as follows: Step S3.1: For each individual, detect the number of single coughs and the time interval; if the number of single coughs within a preset time is less than a preset value, the individual is determined to have occasional coughs; otherwise, it is determined to be another situation. Step S3.2: For individuals in other situations, if there are at least two consecutive single coughs with a time interval of less than a preset value, then the individual is marked as having a non-accidental cough. Step S3.3: Count the number of individuals with non-accidental coughs and aerosol concentration within a preset time and preset range, calculate the aerosol concentration rise rate using a sliding window algorithm, calculate the temperature difference between the outdoor environment and the target hospital environment in real time, collect the humidity change and air pressure change of the target hospital environment within 24 hours, and normalize them. Calculate the following formula: ； In the formula, Indicates the disinfection warning index. This represents the number of individuals with non-accidental coughs after normalization. This represents the normalized aerosol concentration. This represents the normalized temperature difference between the outdoor environment and the target hospital environment. This represents the normalized change in humidity in the target hospital environment over 24 hours. Changes in air pressure in the target hospital environment over 24 hours; , , , , In order , , , , Weighting coefficients; If the disinfection warning index If the value exceeds the preset threshold, it is determined that the hospital environment disinfection early warning program needs to be activated; otherwise, it is determined that the hospital environment disinfection early warning program does not need to be activated.
5. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, In step S4, if it is determined that the hospital environment disinfection early warning procedure needs to be activated, and any of the following conditions are met, then the corresponding disinfection procedure will be activated: If the PM2.5 concentration in the target hospital environment is ≥60μg / m³ and lasts for 2 hours, the medium-intensity disinfection of the ZZNM-40 disinfection device will be triggered. If the TVOC concentration in the target hospital environment is ≥550μg / m³, then the medium-intensity disinfection of the ZZNM-40 sterilizer will be triggered. If the current target hospital's ambient humidity is ≥58% and the hospital's internal ambient humidity is <60%, then the ZZNM-40 sterilizer will be activated for medium-intensity disinfection. If the current hospital environment humidity is ≥60%, the ZZNM-40 sterilizer will be triggered for low-intensity continuous disinfection.
6. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, If the target hospital environment is a children's activity area, then lower the volume recognition threshold for audio collection of the target hospital environment, reduce the disinfection intensity, and shorten the audio collection cycle of the target hospital environment; If the target hospital environment is a patient inpatient area, the volume recognition threshold for audio collection in the target hospital environment should be lowered, the disinfection intensity should be increased, and the audio collection cycle of the target hospital environment should be shortened.
7. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, If the target hospital environment is a doctor's work area, then medium-intensity disinfection should be carried out using the ZZNM-40 disinfection device during the doctor's off-peak working hours; Environmental sensors are used to monitor the flow of people in real time. When multiple doctors are detected moving around at the same time, the disinfection operation is delayed.
8. The dynamic disinfection method based on cough audio detection and analysis according to claim 1, characterized in that, If the target hospital environment is the outpatient area, environmental sensors are used to monitor the flow of people in real time. If it is during the off-peak period, the ZZNM-40 disinfection device is used for low-intensity disinfection. During peak periods of high foot traffic, use the ZZNM-40 sterilizer for medium-intensity or high-intensity disinfection.
9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the dynamic disinfection method based on cough audio detection and analysis as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements each step of the dynamic disinfection method based on cough audio detection and analysis as described in any one of claims 1 to 8.