A portable aerosol detection system

Abstract

The application discloses a portable droplet detection system, belonging to the technical field of respiratory droplet detection, comprising a droplet detection device and a data processing platform. The droplet detection device comprises a mounting shell. The back of the mounting shell is provided with a wearing part. The front of the mounting shell is provided with a window and a baffle; the window is provided with a droplet sensor, and the window is provided with a protective net, and the droplet sensor is arranged in the protective net. The inside of the mounting shell is provided with a droplet detection alarm circuit board, a charging module and a power module which are electrically connected with each other; the droplet detection alarm circuit board is integrated with an FPGA logic control module, an analog-digital conversion module, a posture motion module, an alarm prompt module, a wireless communication module, a signal input end and a power input end. The data processing platform is arranged with a droplet recognition detection model and a visual interface.

Description

Technical Field

[0001] This invention belongs to the field of respiratory droplet detection technology, and in particular relates to a portable droplet detection system. Background Technology

[0002] Numerous studies have shown that droplet transmission is widely considered the primary mode of transmission for respiratory pathogens. When pathogens are released from the respiratory tract of an infected person and spread from person to person through droplets, they can infect the respiratory tracts of exposed and susceptible individuals within the transmission area, leading to illness or death. Therefore, the ability to detect droplets would, to some extent, reduce the risk of infection, pathogenicity, and mortality from respiratory infectious diseases, and save public health resources, time, and costs for respiratory epidemiological investigations and tracing.

[0003] To date, the mainstream methods for detecting droplets by researchers both domestically and internationally can generally be divided into sampling and collection methods and observation methods. Simha et al., in their paper *Universal trends in human cough airflows at large distance*, used schlieren photography to visualize droplet detection, thereby exploring the propagation trends of droplet airflow generated by coughing. However, the above methods have many shortcomings: complex detection processes, high detection costs, large and expensive detection equipment, high environmental requirements, and difficulty in portability. Furthermore, Chinese patent CN112683753A discloses an automatic microbial droplet aerosol detector and system, which pumps droplet aerosols in using an air pump, and a photon counter in a ventilation detection device counts the droplet aerosols. However, this automatic droplet aerosol detector and system is bulky, difficult to carry, and hard to apply to routine droplet detection for prevention. Therefore, there is an urgent need in the field for a droplet detection device to detect droplets. Summary of the Invention

[0004] To address the shortcomings of the aforementioned background technology, the present invention aims to provide a portable droplet detection system that solves the problems of existing droplet detection systems being unable to detect droplets accurately in real time and being inconvenient to carry.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A portable droplet detection system is provided, which includes a droplet detection device and a data processing platform.

[0007] The droplet detection device includes a mounting housing with wearable components on the back and a window and baffle on the front. A droplet sensor is installed inside the window, and a protective net is installed on the window, with the droplet sensor located within the protective net. Inside the mounting housing are electrically connected droplet detection alarm circuit board, charging module, and power supply module. The droplet detection alarm circuit board integrates an FPGA logic control module, an analog-to-digital conversion module, a posture motion module, an alarm indication module, a wireless communication module, and a signal input terminal. The signal input terminal is connected to the droplet sensor via a signal line. The FPGA logic control module integrates a dual-threshold algorithm for extracting the start signal.

[0008] The data processing platform is electrically connected to the wireless communication module in the droplet detection device. The data processing platform is equipped with a droplet recognition and detection model and a visualization interface for determining whether droplets are detected and displaying the detection results. The droplet recognition and detection model includes a data preprocessing module, a feature extraction module, a classification module, and a model optimization module.

[0009] The basic principle of the portable droplet detection system in this invention is as follows: the charging module replenishes the power supply module with electrical energy. The power supply module consists of a battery, a voltage regulator chip, and a voltage regulator circuit, and is used to provide the required rated operating voltage to each module of the droplet detection alarm circuit board. The FPGA logic control module is used for driving each working module and for data acquisition, calculation, processing, integration, transmission, and reception. The dual-threshold algorithm for initial signal extraction integrated in the FPGA logic control module is used to obtain the effective initial signals sensed by the droplet sensor and the posture motion module. The analog-to-digital conversion module is used to convert and transmit the signal changes of the droplet sensor. The posture motion module is used to acquire human motion state information to determine the current airflow environment, thereby assisting in the detection and judgment of droplets. The droplet recognition and detection model and visualization interface arranged in the data processing platform are used to determine whether droplets are detected and to display the detection results. In use, the user can wear the entire droplet detection device on their body through wearable components. After the data processing platform identifies and detects droplets in real time, it will send a command to the droplet detection device, causing the droplet detection device to emit an alarm sound and flash an alarm indicator light, thereby reminding the wearer to maintain a safe distance for respiratory disease transmission and to take disinfection and avoidance measures.

[0010] Furthermore, the droplet detection alarm circuit board also integrates an image acquisition module electrically connected to the FPGA logic control module; specifically, the image acquisition method is a miniature camera acquisition, used to obtain the scene where the device wearer is located.

[0011] Furthermore, the droplet detection alarm circuit board also integrates an infrared temperature measurement module electrically connected to the FPGA logic control module; specifically, the temperature acquisition method is non-contact infrared thermometer acquisition, used to obtain the temperature of the droplet source.

[0012] Furthermore, two data acquisition holes are provided on the front side of the mounting housing; the image acquisition module and the infrared temperature measurement module acquire image signals and temperature signals respectively through the two data acquisition holes.

[0013] Furthermore, the droplet detection alarm circuit board also integrates a data storage module electrically connected to the FPGA logic control module. The data storage module uses SD card offline storage to store sensor data from each working module, facilitating later traceability.

[0014] Furthermore, the alarm prompt module on the droplet detection alarm circuit board is electrically connected to the FPGA logic control module, including an audible alarm device and a visual flashing alarm device.

[0015] Furthermore, a protective net is installed on the window, and the droplet sensor is placed inside the protective net. The protective net consists of multiple crisscrossing strip structures with a diameter of 0.8mm to 1.2mm. The length and width of each grid in the protective net are not less than 5mm × 5mm. The protective net is designed to protect the droplet sensor placed inside from contamination, while ensuring good contact between the droplets and the droplet sensor.

[0016] Furthermore, the extension width of the baffle on the housing is between 1cm and 3cm, which is used to reduce the interference of the wearer's breathing through the mouth and nose on the droplet detection results.

[0017] Furthermore, the droplet sensor is a capacitive sensor with a response recovery time in the millisecond range; the droplet sensor has an interdigitated electrode structure and a droplet-sensitive thin film attached to its surface for sensing droplets.

[0018] Furthermore, the posture motion module is a six-axis sensor, including a three-axis accelerometer and a three-axis angular velocity meter, which is electrically connected to the FPGA logic control module on the droplet detection alarm circuit board to sense the wearer's motion state and assess the current airflow environment.

[0019] Furthermore, the wearable components can be of the collar clip type, the lanyard type, or the magnetic type, and the wearable components can be flexibly selected according to needs.

[0020] Furthermore, the dual-threshold algorithm for initial signal extraction includes:

[0021] Step 1: Set the sampling frequency f of the FPGA logic control module req Set the acquired stable sensor sequence as the baseline sequence {x(m), 1≤m}, and set the baseline sequence buffer depth H. bl A sliding window is set during real-time sampling, and the depth H of the sliding window buffer is set. swThe sliding window sequence is {y(n), 1≤n}, and the relative mean square error threshold of the sliding window sequence is set to T. σ Set the comparison factor C for the first-order rate of change of the sliding window sequence;

[0022] Step 2: Calculate and store the baseline mean of the baseline sequence x(m), i.e.:

[0023]

[0024] Step 3: Calculate the relative mean square error of the current sliding window sequence y(n) relative to the baseline sequence mean baseline_av. And deposit, that is:

[0025]

[0026] Step 4: Calculate the relative mean square error of the current sliding window sequence y(n). With threshold T σ Compare; if relative mean square error Less than threshold T σ If the new value is recursively applied to the sliding window, and the sliding window sequence y(n) is updated, return to step 3; if the relative mean square error Greater than threshold T σ If so, retain the current sliding window sequence y(n) and proceed to step 5;

[0027] Step 5: Calculate the first-order rate of change Δx of the current sliding window sequence y(n). n ,Right now:

[0028] Δx n =f req (x n -x n-1 )

[0029] Step 6: Apply the first-order rate of change Δx of the signal to the current sliding window sequence y(n). n The positive and negative values ​​are counted, that is:

[0030] Count + =Count + +1

[0031] Count - =Count - +1

[0032] Step 7: Compare the number of positive rates of change (Count) + Count of negative rates of change -1. Compare the magnitudes of the number factors C; if the number of positive rates of change is less than the sum of the number of negative rates of change and the number factors compared, then recursively push the new value to the sliding window and update the sliding window sequence y(n) and return to step 3; otherwise, determine the current sliding window sequence y(n) as the starting signal of the valid signal, and then move the data of the starting signal and the subsequent acquired signals to the valid signal transmission buffer in a queue manner.

[0033] The algorithm described above mainly integrates the relative mean square error and the first-order rate of change as dual features to extract the effective starting signals of the droplet sensor and the attitude motion module. This aims to improve the utilization rate and efficiency of transmission bandwidth, reduce power consumption, and provide a reliable data foundation for the determination and identification of droplet signals and non-droplet signals.

[0034] Furthermore, the droplet recognition and detection model includes:

[0035] Step 1: Perform mean filtering on the droplet sensor signal received by the data processing platform to obtain x(n); perform Kalman filtering on the triaxial acceleration signal and triaxial angular velocity signal received by the data processing platform to obtain A. x (n), A y (n), A z (n), ω x (n), ω y (n), ω z (n);

[0036] Step 2: Calculate the resultant acceleration and resultant angular velocity signals based on the filtered triaxial acceleration and angular velocity signals;

[0037]

[0038] Step 3: Combine the filtered droplet sensor signal, resultant acceleration signal, and resultant angular velocity signal to establish a three-channel dataset, and then standardize, label, and tensorize the dataset.

[0039] Step 4: Create a data loader and divide the training set, validation set, and test set proportionally;

[0040] Step 5: Create a bidirectional gated recurrent unit neural network model and a 1D convolutional neural network model on the training set to extract signal features, and set and adjust hyperparameters;

[0041] Step 6: Perform feature fusion, PCA dimensionality reduction, and standardization on the features extracted in parallel;

[0042] Step 7: Input the processed features into the SVM classifier for classification;

[0043] Step 8: Train the droplet recognition and detection model, perform forward propagation and loss calculation; determine if the model has converged; if the model has not converged, perform backpropagation and use the Adam optimizer to optimize the droplet recognition and detection model; if the model has converged, complete the training and save the model and parameters.

[0044] Step 9: Input the validation set to evaluate the model, and introduce a learning rate decay strategy and an early stopping mechanism to improve the training effect;

[0045] Step 10: Input the test set into the droplet recognition and detection model to obtain the output droplet judgment result.

[0046] The aforementioned model primarily integrates algorithms such as mean filtering, Kalman filtering, gated recurrent unit neural networks, 1D convolutional neural networks, and support vector machines to filter droplet signals and eliminate interference signals. The model exhibits good classification performance and high recognition accuracy.

[0047] The beneficial effects of this invention are as follows:

[0048] 1. The portable droplet detection system of the present invention integrates a dual-threshold algorithm for initial signal extraction and a droplet recognition and detection model, which can detect droplets accurately in real time and has the advantages of long battery life, stable operation and strong anti-interference ability.

[0049] 2. The portable droplet detection system of the present invention is based on the parallel processing of the FPGA logic control module and the millisecond-level response recovery time of the droplet sensor. The response time of the portable droplet detection system is within 1 second. Once droplets are detected, an alarm is triggered quickly. It is sensitive to droplets and can detect them at a distance of more than 2 meters. It is also low in cost, small in size, and can be worn in a variety of ways.

[0050] 3. A portable droplet detection system of the present invention supports the visual display of droplet signal changes by setting up a data processing terminal. Attached Figure Description

[0051] Figure 1 Three-dimensional explosion of a droplet detection device Figure 1 .

[0052] Figure 2 Three-dimensional explosion of a droplet detection device Figure 2 .

[0053] Figure 3 This is a schematic diagram of the structure of the droplet detection host computer platform.

[0054] Figure 4 This is a schematic diagram of a droplet detection device.

[0055] Figure 5 This is a schematic diagram of the interface of the droplet detection host computer platform.

[0056] Figure 6 This is a flowchart illustrating the processing of the effective signal start frame from the droplet sensor and attitude motion module.

[0057] Figure 7 This is a flowchart for identifying and processing droplet signals and non-droplet signals.

[0058] The components include: 1. Mounting housing; 2. Droplet detection and alarm circuit board; 3. Data processing platform; 4. Partition; 5. Protective net; 6. Droplet sensor; 8. Analog-to-digital conversion module; 9. Attitude motion module; 10. Infrared temperature measurement module; 11. Image acquisition module; 12. FPGA logic control module; 13. Alarm prompting module; 14. Data storage module; 15. Wireless communication module; 16. Charging module; 17. Signal input terminal; 18. Power cord; 19. Power module; 20. Power input terminal; 21. Wearable components; 22. Baffle. Detailed Implementation

[0059] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0060] like Figure 1 and Figure 2 As shown, the present invention provides a droplet detection device, including a droplet detection device and a data processing platform.

[0061] The droplet detection device includes a mounting housing 1. A window and a baffle 22 are provided on the front of the mounting housing 1. The baffle 22 is located above the window, and a droplet sensor 6 is disposed within the window for detecting droplets generated by breathing, coughing, sneezing, or blowing one's nose. Preferably, the extension width of the baffle 22 is between 1 cm and 3 cm to reduce interference from the wearer's breathing during droplet detection. Preferably, a protective net 5 is provided on the window, and the droplet sensor 6 is disposed within the protective net 5. The protective net 5 includes multiple crisscrossing strip structures with a diameter of 0.8 mm to 1.2 mm; the length and width dimensions of a single grid in the protective net 5 are greater than 5 mm × 5 mm. The protective net 5 aims to protect the droplet sensor 6 placed within it from contamination while ensuring good contact between droplets and the droplet sensor 6. Further, the droplet sensor 6 has an interdigitated electrode structure with a response recovery time in the millisecond range, and a droplet-sensitive thin film is attached to its surface for sensing droplets.

[0062] The back of the mounting housing 1 is provided with a wearable component 21; the wearable component 21 can be a collar clip, a lanyard, or a magnetic structure, and the wearable component 21 can be flexibly selected as needed. In this embodiment, the wearable component 21 is a clip, which can be easily worn by the user at the collar or chest.

[0063] The mounting housing 1 contains a droplet detection alarm circuit board 2, a charging module 16, and a power module 19 that are electrically connected to each other. Specifically, a partition 4 is detachably installed inside the mounting housing 1, which can divide the interior of the mounting housing 1 into a control compartment and a power compartment. The droplet detection alarm circuit board 2 is installed in the control compartment, while the charging module 16 and the power module 19 are installed in the power compartment.

[0064] Specifically, the droplet detection alarm circuit board 2 integrates an FPGA logic control module 12, an analog-to-digital conversion module 8, an attitude motion module 9, an alarm prompt module 13, a wireless communication module 15, a signal input terminal 17, and a power input terminal 20; the signal input terminal 17 is connected to the droplet sensor 6 via a signal line; the power input terminal 20 is electrically connected to the power module 19 via a power line 18; the FPGA logic control module 12 integrates a dual-threshold algorithm for extracting the start signal.

[0065] The charging module 16 replenishes the power supply module 19 with electrical energy. The power supply module 19 consists of a battery, a voltage regulator chip, and a voltage regulator circuit, and is used to provide the required rated operating voltage to each module of the droplet detection alarm circuit board 2. The FPGA logic control module 12 is used for driving each working module and for data acquisition, calculation, processing, integration, transmission, and reception. The analog-to-digital converter module 8 is used to convert and transmit the signal changes of the droplet sensor 6. The posture and motion module 9 is used to acquire human motion state information to determine the current airflow environment, thereby assisting in the detection and judgment of droplets. The dual-threshold algorithm for initial signal extraction integrated in the FPGA logic control module 12 is used to acquire the effective initial signals sensed by the droplet sensor 8 and the posture and motion module 9.

[0066] like Figure 1 , Figure 2 , Figure 3 and Figure 4 As shown, when in use, the user can wear the entire droplet detection device on their body through the wearable component 21. After the data processing platform 3 identifies and detects droplets in real time, it will send a command to the droplet detection device, causing the alarm prompt module 13 of the droplet detection device to emit an alarm sound and flash an alarm indicator light, thereby reminding the wearer to maintain a safe distance for respiratory disease transmission and to take disinfection and avoidance measures.

[0067] like Figure 3 and Figure 5As shown, the data processing platform 3 is electrically connected to the wireless communication module 15 on the droplet detection alarm circuit board to realize information interaction; the data processing platform 3 is equipped with a droplet recognition and detection model and a visualization interface, which are used to determine whether droplets are detected and to display the detection results; the droplet recognition and detection model includes a data preprocessing module, a feature extraction module, a classification module and a model optimization module.

[0068] Preferably, but not limited to, the droplet detection alarm circuit board 2 also integrates a data storage module 14 electrically connected to the FPGA logic control module 12. The data storage module 14 uses SD card offline storage to store sensor data from each working module, which is convenient for later traceability.

[0069] Preferably, but not limited to, the droplet detection alarm circuit board 2 also integrates an infrared temperature measurement module 10 electrically connected to the FPGA logic control module 12, used to acquire the temperature of the droplet source; specifically, the temperature acquisition method is non-contact infrared thermometer acquisition.

[0070] Preferably, but not limited to, the droplet detection alarm circuit board 2 also integrates an image acquisition module 11 electrically connected to the FPGA logic control module 12, used to acquire the scene where the device wearer is located; the image acquisition method is a miniature camera acquisition.

[0071] Preferably, but not limited to, two data acquisition holes are provided on the front of the mounting housing 1; the infrared temperature measurement module 10 and the image acquisition module 11 acquire temperature signals and image signals respectively through the two data acquisition holes.

[0072] like Figure 6 As shown, the dual-threshold algorithm for initial signal extraction includes:

[0073] Step 1: Set the sampling frequency f of the FPGA logic control module req Set the acquired stable sensor sequence as the baseline sequence {x(m), 1≤m}, and set the baseline sequence buffer depth H. bl A sliding window is set during real-time sampling, and the depth H of the sliding window buffer is set. sw The sliding window sequence is {y(n), 1≤n}, and the relative mean square error threshold of the sliding window sequence is set to T. σ Set the comparison factor C for the first-order rate of change of the sliding window sequence;

[0074] Step 2: Calculate and store the baseline mean of the baseline sequence x(m), i.e.:

[0075]

[0076] Step 3: Calculate the relative mean square error of the current sliding window sequence y(n) relative to the baseline sequence mean baseline_av. And deposit, that is:

[0077]

[0078] Step 4: Calculate the relative mean square error of the current sliding window sequence y(n). With threshold T σ Compare; if relative mean square error Less than threshold T σ If the new value is recursively applied to the sliding window, and the sliding window sequence y(n) is updated, return to step 3; if the relative mean square error Greater than threshold T σ If so, retain the current sliding window sequence y(n) and proceed to step 5;

[0079] Step 5: Calculate the first-order rate of change Δx of the current sliding window sequence y(n). n ,Right now:

[0080] Δx n =f req (x n -x n-1 )

[0081] Step 6: Apply the first-order rate of change Δx of the signal to the current sliding window sequence y(n). n The positive and negative values ​​are counted, that is:

[0082] Count + =Count + +1

[0083] Count - =Count - +1

[0084] Step 7: Compare the number of positive rates of change (Count) + Count of negative rates of change - 1. Compare the magnitudes of the number factors C; if the number of positive rates of change is less than the sum of the number of negative rates of change and the number factors compared, then recursively push the new value to the sliding window and update the sliding window sequence y(n) and return to step 3; otherwise, determine the current sliding window sequence y(n) as the starting signal of the valid signal, and then move the data of the starting signal and the subsequent acquired signals to the valid signal transmission buffer in a queue manner.

[0085] The algorithm described above mainly integrates the relative mean square error and the first-order rate of change as dual features to extract the effective starting signals of the droplet sensor and the attitude motion module. This aims to improve the utilization rate and efficiency of transmission bandwidth, reduce power consumption, and provide a reliable data foundation for the determination and identification of droplet signals and non-droplet signals.

[0086] like Figure 7 As shown, the droplet recognition and detection model includes:

[0087] Step 1: Perform mean filtering on the droplet sensor signal received by the data processing platform to obtain x(n); perform Kalman filtering on the triaxial acceleration signal and triaxial angular velocity signal received by the data processing platform to obtain A. x (n), A y (n), A z (n), ω x (n), ω y (n), ω z (n);

[0088] Step 2: Calculate the resultant acceleration and resultant angular velocity signals based on the filtered triaxial acceleration and angular velocity signals;

[0089]

[0090] Step 3: Combine the filtered droplet sensor signal, resultant acceleration signal, and resultant angular velocity signal to establish a three-channel dataset, and then standardize, label, and tensorize the dataset.

[0091] Step 4: Create a data loader and divide the training set, validation set, and test set proportionally;

[0092] Step 5: Create parallel bidirectional gated recurrent unit neural network models and 1D convolutional neural network models on the training set, extract signal features, and set and adjust hyperparameters;

[0093] Step 6: Perform feature fusion, PCA dimensionality reduction, and standardization on the features extracted in parallel;

[0094] Step 7: Input the processed features into the SVM classifier for classification;

[0095] Step 8: Train the droplet recognition and detection model, perform forward propagation and loss calculation; determine if the model has converged; if the model has not converged, perform backpropagation and use the Adam optimizer to optimize the droplet recognition and detection model; if the model has converged, complete the training and save the model and parameters.

[0096] Step 9: Input the validation set to evaluate the model, and introduce a learning rate decay strategy and an early stopping mechanism to improve the training effect;

[0097] Step 10: Input the test set into the droplet recognition and detection model to obtain the output droplet judgment result.

[0098] The aforementioned model primarily integrates algorithms such as mean filtering, Kalman filtering, gated recurrent unit neural networks, 1D convolutional neural networks, and support vector machines to filter droplet signals and eliminate interference signals. The model exhibits good classification performance and high recognition accuracy.

Claims

1. A portable droplet detection system, characterized in that, Includes droplet detection devices and data processing platforms; The droplet detection device includes a mounting housing; a wearable component is provided on the back of the mounting housing, and a window and a baffle are provided on the front of the mounting housing; the baffle is located above the window, a droplet sensor is provided inside the window, a protective net is provided on the window, and the droplet sensor is located inside the protective net; the interior of the mounting housing contains a droplet detection alarm circuit board, a charging module, and a power supply module that are electrically connected to each other; the droplet detection alarm circuit board integrates an FPGA logic control module, an analog-to-digital conversion module, an attitude motion module, an alarm prompt module, a wireless communication module, and a signal input terminal; the signal input terminal is connected to the droplet sensor via a signal line; the FPGA logic control module integrates a dual-threshold algorithm for extracting the start signal; The data processing platform is electrically connected to the wireless communication module. The data processing platform includes a droplet recognition and detection model and a visualization interface for determining whether droplets are detected and displaying the detection results. The droplet recognition and detection model includes a data preprocessing module, a feature extraction module, a classification module, and a model optimization module. The droplet detection alarm circuit board also integrates an image acquisition module and an infrared temperature measurement module, which are electrically connected to the FPGA logic control module. Two data acquisition holes are also provided on the front of the mounting housing. The image acquisition module and the infrared temperature measurement module acquire image signals and temperature signals respectively through the two data acquisition holes. The dual-threshold algorithm for initial signal extraction is used to extract valid initial signals sensed by the droplet sensor and the attitude motion module. The dual-threshold algorithm for initial signal extraction includes: Step 1: Set the sampling frequency of the FPGA logic control module Set the collected signal stable sensor sequence as the baseline sequence { x ( m ), 1≤ m Set the baseline sequence buffer depth. A sliding window is set during real-time sampling, and the depth of the sliding window buffer is set. The sliding window sequence is { y ( n ), 1≤ n Set the relative mean square error threshold of the sliding window sequence to} Set the comparison factor C for the first-order rate of change of the sliding window sequence; Step 2: Calculate the baseline sequence x ( m The baseline average of ) is stored, i.e.: Step 3: Calculate the current sliding window sequence y ( n ) relative to baseline sequence mean relative mean square error And deposit, that is: Step 4: Set the current sliding window sequence y ( n The relative mean square error With threshold Compare; if relative mean square error Less than the threshold Then, the new value is recursively applied to the sliding window, and the sliding window sequence is updated. y ( n Return to step 3; if the relative mean square error Greater than the threshold Then retain the current sliding window sequence. y ( n ), and proceed to step 5; Step 5: Calculate the current sliding window sequence y ( n The first-order rate of change of the signal ,Right now: Step 6: For the current sliding window sequence y ( n The first-order rate of change of the signal The positive and negative values ​​are counted, that is: Step 7: Compare the number of positive rates of change Number of negative rates of change Compare the magnitudes of the number factors C; if the number of positive rates of change is less than the sum of the number of negative rates of change and the number factors being compared, then recursively apply the new value to the sliding window and update the sliding window sequence. y ( n Return to step 3; otherwise, determine the current sliding window sequence. y ( n The start signal is used as the valid signal, and then the data of the start signal and the subsequent acquired signals are moved to the valid signal transmission buffer in a queue manner.

2. The portable droplet detection system according to claim 1, characterized in that, The droplet detection alarm circuit board also integrates a data storage module that is electrically connected to the FPGA logic control module.

3. The portable droplet detection system according to claim 1, characterized in that, The analog-to-digital conversion module is electrically connected to the FPGA logic control module and the signal input terminal on the droplet detection alarm circuit board, respectively, and is used to convert and transmit the signal changes sensed by the droplet sensor; the attitude motion module is electrically connected to the FPGA logic control module, and the attitude motion module is a six-axis sensor, which includes a three-axis accelerometer and a three-axis angular velocity meter.

4. The portable droplet detection system according to claim 1, characterized in that, The alarm notification module is electrically connected to the FPGA logic control module, and the alarm notification module includes an audible alarm device and a visual flashing alarm device.

5. The portable droplet detection system according to claim 1, characterized in that, The protective netting comprises multiple crisscrossing strip structures, with the diameter of each strip structure ranging from 0.8mm to 1.2mm. The length and width of each individual grid in the protective netting are not less than 5mm × 5mm. The extension width of the baffle ranges from 1cm to 3cm, which is used to reduce the interference of the wearer's breathing through the mouth and nose on the droplet detection results.

6. The portable droplet detection system according to claim 1, characterized in that, The droplet sensor has an interdigitated electrode structure and is a capacitive sensor with a droplet-sensitive thin film attached to its surface for sensing droplets.

7. The portable droplet detection system according to claim 1, characterized in that, The wearable component can be a collar clip, a lanyard, or a magnetic attachment.

8. The portable droplet detection system according to claim 1, characterized in that, The droplet identification and detection model includes: Step 1: Perform mean filtering on the droplet sensor signal received by the data processing platform to obtain... x ( n The triaxial acceleration and triaxial angular velocity signals received by the data processing platform are processed by Kalman filtering to obtain... , , , , , ; Step 2: Calculate the resultant acceleration and resultant angular velocity signals based on the filtered triaxial acceleration and angular velocity signals; Step 3: Combine the filtered droplet sensor signal, resultant acceleration signal, and resultant angular velocity signal to establish a three-channel dataset, and then standardize, label, and tensorize the dataset. Step 4: Create a data loader and divide the training set, validation set, and test set proportionally; Step 5: Create a parallel bidirectional gated recurrent unit neural network model and a 1D convolutional neural network model on the training set, extract signal features, and set and adjust hyperparameters; Step 6: Perform feature fusion, PCA dimensionality reduction, and standardization on the features extracted by the parallel network; Step 7: Input the processed features into the SVM classifier for classification; Step 8: Train the droplet recognition and detection model, perform forward propagation and loss calculation; determine if the model has converged; if the model has not converged, perform backpropagation and use the Adam optimizer to optimize the droplet recognition and detection model; if the model has converged, complete the training and save the model and parameters. Step 9: Input the validation set to evaluate the model, and introduce a learning rate decay strategy and an early stopping mechanism to improve the training effect; Step 10: Input the test set into the droplet recognition and detection model to obtain the output droplet judgment result.

Images