Adaptive noise reduction method for positive pressure type air breathing machine based on airflow frequency spectrum characteristics

Abstract

The invention discloses a self-adaptive noise reduction method for a positive pressure type air respirator based on airflow spectrum characteristics, and relates to the technical field of positive pressure type air respirators.The self-adaptive noise reduction method comprises the steps that a respirator air path pressure signal and an environment noise signal are synchronously collected through an airflow sensor and a microphone array; performing time-frequency analysis on the collected signals, and extracting airflow spectrum features; and establishing a mapping relation between the working state of the breathing machine and the noise spectrum based on the historical record according to the mapping relation between the working state of the breathing machine and the noise spectrum. Through real-time acquisition and analysis of airflow frequency spectrum characteristics, accurate identification of useful airflow signals and environmental noise signals, dynamic adjustment of noise reduction parameters, effective suppression of high-frequency noise and low-frequency vibration, and improvement of noise reduction effect, compared with a traditional fixed frequency band filtering or passive sound insulation material, the method can better cope with a non-stationary noise environment, and has good application prospects. The noise reduction strategy is adjusted in real time, clear auditory information is provided in a complex noise environment, and therefore the auditory perception ability of firemen is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of positive pressure air breathing apparatuses, and particularly to an adaptive noise reduction method for positive pressure air breathing apparatuses based on airflow spectrum characteristics. Background Art

[0002] Firefighters need to wear positive pressure air breathing apparatuses at the fire scene to ensure breathing safety. Usually, there are high-decibel noises at the fire scene, such as the sound of flame combustion, the operation sound of fire-fighting equipment, the sound of building collapse, etc. These noises will seriously interfere with the auditory system of firefighters, affecting their judgment of the surrounding environment and communication ability. Moreover, the breathing apparatus will generate noises such as airflow fluctuations and mechanical vibrations during operation, further exacerbating the on-site noise environment and reducing the comfort and work efficiency of firefighters. In fire rescue, clear auditory information is crucial for the command and dispatch of firefighters, teamwork, and rapid response to emergencies.

[0003] Traditional noise reduction methods usually adopt fixed-frequency band filtering or passive sound insulation materials, but they have poor adaptability to non-stationary noises such as sudden changes in breathing airflow and sudden environmental noises, and cannot adjust the noise reduction parameters in real time to cope with the dynamically changing noise environment. Although traditional spectrum analysis methods can analyze noises to a certain extent, they often cannot effectively suppress high-frequency noises such as motor harmonics and low-frequency vibrations such as valve opening and closing impacts at the same time. In addition, the dynamic influence of the working state of the breathing apparatus, including the supply air pressure and the user's breathing frequency, on the noise spectrum is not fully considered. Therefore, an adaptive noise reduction method for positive pressure air breathing apparatuses based on airflow spectrum characteristics is proposed to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide an adaptive noise reduction method for positive pressure air breathing apparatuses based on airflow spectrum characteristics to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] An adaptive noise reduction method for positive pressure air breathing apparatuses based on airflow spectrum characteristics includes the following steps:

[0007] Step 1: Synchronously collect the air pressure signal of the breathing apparatus and the environmental noise signal through an airflow sensor and a microphone array;

[0008] Step 2: Perform time-frequency analysis (short-time Fourier transform or wavelet transform) on the collected signals to extract the airflow spectrum characteristics;

[0009] Step 3: Establish the mapping relationship between the working state of the ventilator and the noise spectrum based on historical records, and perform targeted data matching and differential adjustment according to the mapping relationship between the working state of the ventilator and the noise spectrum, combined with the differences between the uplink and downlink audio;

[0010] Step 4: Under the multi-task learning framework of polymorphic conversion, use a deep learning model of a CNN-LSTM hybrid network to construct a noise recognition model to distinguish useful airflow signals from environmental noise signals;

[0011] Step 5: Introduce a multi-module supervision strategy and design a grouping strategy to optimize model deployment to ensure that the algorithm can still achieve a leading noise reduction effect under limited computing power;

[0012] Step 6: Introduce a feedback mechanism, analyze the user's wearing comfort through ear pressure sensor data, and optimize the model parameters to ensure the balance between noise reduction effect and wearing experience.

[0013] A further improvement of the technical solution of the present invention lies in that: the specific content of the said Step 1 includes:

[0014] Deploy high-precision airflow sensors and microphone arrays in a positive-pressure air ventilator. Among them, the airflow sensors are installed at the position of the air supply valve to monitor the changes in airflow velocity and flow rate. The microphone arrays are arranged inside the mask to collect internal airflow noise and external environmental noise, and ear pressure sensors are installed at the ear position to monitor the user's wearing comfort;

[0015] Initialize all sensors, set the sampling frequency (44.1kHz or 48kHz), ensure that the sensors work properly and perform calibration to eliminate system errors, and then synchronously collect airflow pressure signals and environmental noise signals through a data acquisition card;

[0016] Preprocess the collected airflow pressure signals and environmental noise signals, including filtering, normalization, and feature extraction. Among them, use a low-pass filter to remove high-frequency noise and retain the main airflow characteristics, and adjust the signal amplitude to a unified range through normalization processing for subsequent analysis;

[0017] Synchronize the airflow pressure signals and environmental noise signals in time to ensure that the airflow pressure signals and environmental noise signals are aligned in time, and then use data fusion technology to integrate the airflow signals and noise signals into a comprehensive data set for subsequent noise recognition and noise reduction processing.

[0018] A further improvement of the technical solution of the present invention lies in that: the specific content of the said Step 2 includes:

[0019] The short-time Fourier transform is used to perform time-frequency analysis on the preprocessed airflow pressure signal and ambient noise signal. Among them, the short-time Fourier transform is suitable for processing quasi-stationary signals and can provide good frequency resolution;

[0020] The signal is segmented into multiple short time periods through a sliding window, and Fourier transform is performed on each time period to analyze the variation of the signal's spectrum over time, obtaining the spectrum diagram of the signal;

[0021] After obtaining the time-frequency analysis result, airflow spectrum features are extracted from the spectrum diagram of the signal, including the main noise frequency components, noise bandwidth, and airflow fluctuation frequency, and the airflow spectrum features are used to reflect the characteristics of the airflow signal and ambient noise;

[0022] A multi-task learning framework with multi-state conversion is constructed, and multiple task branches are set up, each responsible for different processing tasks, including multiple processing tasks such as airflow spectrum feature extraction, noise recognition, and noise reduction parameter adjustment.

[0023] A further improvement of the technical solution of the present invention lies in that: the process of obtaining the spectrum diagram of the signal includes:

[0024] A sliding window with a fixed length is used to segment the preprocessed airflow pressure signal and ambient noise signal, so that the sliding window slides on the signal, moving a fixed step length each time, and the entire signal is segmented into multiple overlapping short signal segments, approximating the non-stationary signal as a series of short-time stationary signal segments for subsequent analysis;

[0025] Fourier transform is performed on the short signal segments within each sliding window, converting them from the time domain to the frequency domain. Through Fourier transform, the spectrum of each short signal segment is obtained, including the amplitude spectrum and phase spectrum, and the amplitude spectrum reflects the energy distribution of the signal at different frequencies;

[0026] The amplitude spectra of each short signal segment are arranged in chronological order to generate the spectrum diagram of the signal. Among them, the horizontal axis of the spectrum diagram represents time, the vertical axis represents frequency, and the gray scale represents the amplitude size. Through the spectrum diagram of the signal, the variation of the signal's spectrum over time can be observed, clearly showing the appearance and disappearance of different frequency components in the signal and the intensity change.

[0027] A further improvement of the technical solution of the present invention lies in that: the specific steps of step three include:

[0028] Historical airflow pressure signals and ambient noise signals of the ventilator in different working states are collected, and preprocessing and extraction operations of airflow spectrum features are performed on the signals, and the historical records are integrated. Among them, the working states include supply pressure, user breathing frequency, and breathing mode (inhalation, exhalation);

[0029] Analyze the changes in the noise spectrum under different working conditions in combination with historical records, construct a noise database, and establish a mapping relationship between the working state of the ventilator and the noise spectrum;

[0030] Analyze the differences between the uplink and downlink audio, compare the spectral characteristics of the uplink and downlink audio, and identify the noise differences caused by different respiratory airflow directions. The noise differences include amplitude differences, frequency differences, and phase differences. Here, the uplink and downlink audio respectively refer to the audio signals generated by the user's exhaled and inhaled airflows;

[0031] According to the established mapping relationship and the differences between the uplink and downlink audio, perform targeted data matching and differential adjustment. For each working state, match the corresponding noise reduction parameters, including the bandwidth and filtering intensity of the filter, and optimize the audio fidelity by dynamically adjusting the noise reduction parameters. The differential adjustment is the separate processing of different frequency components, enhancing the low-frequency airflow signal and suppressing high-frequency noise.

[0032] A further improvement in the technical solution of the present invention lies in that: the analysis process of the differences between the uplink and downlink audio includes:

[0033] By arranging high-precision airflow sensors and microphone arrays inside the air supply valve and mask of the ventilator, respectively collect the uplink and downlink audio signals generated by the user's exhaled and inhaled airflows, and preprocess the collected signals, including filtering to remove high-frequency noise and normalizing and adjusting the signal amplitude to a unified range. Segment the signal through a sliding window to ensure the continuity and stability of the signal;

[0034] Perform time-frequency analysis on the preprocessed uplink and downlink audio signals. Use the short-time Fourier transform to convert the signal from the time domain to the frequency domain, and extract the amplitude spectrum, frequency components, and phase information of each short signal segment. Among them, the amplitude spectrum reflects the energy distribution of the signal at different frequencies, the frequency components show the main frequencies of the signal, and the phase information provides the time characteristics of the signal;

[0035] Compare the spectral characteristics of the uplink and downlink audio, and identify the noise differences of amplitude differences, frequency differences, and phase differences caused by different respiratory airflow directions. Among them, analyze the amplitude differences, observe the energy changes of the uplink and downlink audio at the same frequency, and analyze their relationship with the respiratory airflow direction. Analyze the frequency differences, identify the different frequency components that appear in the uplink and downlink audio, analyze the phase differences, study the phase changes of the uplink and downlink audio over time, and determine the influence of the respiratory airflow direction on the noise characteristics through the difference analysis;

[0036] According to the identified differences between the uplink and downlink audio, perform targeted data matching. For each working state (supply pressure, respiratory rate, respiratory mode), match the corresponding noise reduction parameters. Among them, in the inhalation stage, enhance the low-frequency airflow signal, and in the exhalation stage, suppress high-frequency noise;

[0037] Audio fidelity is optimized through differentiated adjustments. During the inhalation phase, the low-frequency airflow signal is enhanced to ensure that the user can clearly feel the airflow. During the exhalation phase, the high-frequency noise is suppressed.

[0038] A further improvement of the technical solution of the present invention is that the step 4 specifically includes:

[0039] Extract audio data sets containing useful airflow signals and environmental noise signals from historical records to ensure that the data covers different working conditions and noise types. Annotate the audio data sets to distinguish useful airflow signals from environmental noise signals, and then divide the data sets into training sets, validation sets, and test sets to ensure data distribution consistency.

[0040] The deep learning model of the CNN-LSTM hybrid network is used as the basic architecture to convert the signals in the audio dataset into spectrograms as the hybrid input of the CNN-LSTM network;

[0041] Use the training set to train the CNN-LSTM hybrid network, define the cross entropy loss function to measure the difference between the prediction and the true label, use the Adam optimizer to adjust the network parameters, dynamically adjust the learning rate based on the performance of the validation set, and then use the test set to evaluate the trained model, calculate the accuracy, recall rate, and F1 score indicators to quantify the performance of the model, analyze the misclassification through the confusion matrix, optimize the model in a targeted manner, and then obtain a trained noise recognition model;

[0042] The trained noise recognition model is integrated into the positive pressure air ventilator to ensure low-latency reasoning capability. The real-time airflow pressure signal and environmental noise signal are input, converted into a spectrum graph and then input into the noise recognition model. The noise recognition model then outputs the classification result, calculates the classification score, analyzes the probability that the signal belongs to a useful airflow signal, and distinguishes the useful airflow signal from the environmental noise signal in real time.

[0043] A further improvement of the technical solution of the present invention is that the calculation process of the classification score includes:

[0044] The airflow pressure signal and the ambient noise signal are collected in real time, the signal is converted into a spectrum diagram of the signal through short-time Fourier transform, and the energy value of each time-frequency point is calculated from the spectrum diagram of the signal;

[0045] The energy value of each time-frequency point is normalized using a reference energy value, wherein the reference energy value is obtained through historical data statistics, and the normalized energy value is weighted and summed;

[0046] Introduce a time decay factor to simulate the decay of the signal over time, and combine the weighted summation result. Map the result of the weighted summation to the range [0, 1] through an exponential function and a logistic regression function to obtain a classification score.

[0047] A further improvement of the technical solution of the present invention lies in: The specific steps of step five include:

[0048] Divide the noise recognition model into multiple sub-modules, including a feature extraction module, a feature fusion module, and a classification module, and set supervision points at the output end of each sub-module to monitor the output quality and performance of the sub-module. Among them, the supervision point uses the cross-entropy loss function to measure the difference between the intermediate output and the target value, and updates the parameters through the backpropagation algorithm. Through the multi-module supervision strategy, ensure that each module can be independently optimized, improving the generalization ability and robustness of the overall model;

[0049] Evaluate the computing power limitations of the positive pressure air respirator, including the available resources of the CPU, GPU, and memory. Determine the computing power and memory limitations of different hardware through performance testing. According to the evaluation results, allocate the computationally intensive feature extraction module to the GPU, allocate the time-series processing feature fusion module to the CPU, and allocate the lightweight classification module to a dedicated FPGA to ensure that each sub-module runs on the most suitable hardware, reducing the computing delay;

[0050] According to the computing power requirements and grouping strategy, divide the sub-modules into multiple groups. Divide the feature extraction module and the feature fusion module into one group and run them on the GPU, and divide the classification module into another group and run it on the CPU.

[0051] A further improvement of the technical solution of the present invention lies in: The specific steps of step six include:

[0052] Combine an ear pressure sensor to collect ear pressure data in real time during the user's wearing process, and preprocess the collected ear pressure data, including filtering to remove high-frequency noise and normalizing to adjust the signal amplitude to a unified range;

[0053] Calculate statistical indicators such as the mean, variance, and peak value of the ear pressure as quantitative indicators of the user's wearing comfort, analyze the change trend of the ear pressure statistical indicators, calculate the wearing comfort score, and evaluate the user's wearing comfort under different noise reduction intensities;

[0054] According to the evaluation results of the wearing comfort, dynamically adjust the noise reduction parameters. If the mean value of the ear pressure is too high, reduce the noise reduction intensity to reduce the stimulation to the user's ears. If the peak value is too high, adjust the bandwidth of the filter.

[0055] Due to the adoption of the above technical solution, the technical progress achieved by the present invention compared with the prior art is:

[0056] 1. The present invention provides an adaptive noise reduction method for a positive pressure air breathing apparatus based on airflow spectrum characteristics. By collecting and analyzing airflow spectrum characteristics in real time, useful airflow signals and environmental noise signals are accurately identified, and the noise reduction parameters are dynamically adjusted to effectively suppress high-frequency noise and low-frequency vibration, improving the noise reduction effect. Compared with traditional fixed-frequency filtering or passive sound insulation materials, it can better cope with non-stationary noise environments, adjust the noise reduction strategy in real time, ensure clear auditory information in complex noise environments, and thus enhance the auditory perception ability of firefighters.

[0057] 2. The present invention provides an adaptive noise reduction method for a positive pressure air breathing apparatus based on airflow spectrum characteristics. By introducing a feedback mechanism, an ear pressure sensor is used to monitor the user's wearing comfort in real time, and the noise reduction parameters are dynamically adjusted. According to the change of ear pressure, the noise reduction intensity can be appropriately reduced or the bandwidth of the filter can be adjusted to reduce the stimulation to the user's ears. By balancing the noise reduction effect and the wearing experience, it is ensured that the user feels comfortable during long-term use and avoids ear discomfort or fatigue caused by excessive noise reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0059] Figure 1 It is a flowchart of the present invention;

[0060] Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0062] Example 1, as Figure 1 、 Figure 2 shown, the present invention provides an adaptive noise reduction method for a positive pressure air breathing apparatus based on airflow spectrum characteristics, including the following steps:

[0063] Step 1: Synchronously collect the pressure signal of the ventilator air circuit and the ambient noise signal through an airflow sensor and a microphone array. Deploy a high-precision airflow sensor and a microphone array in a positive-pressure air ventilator. Among them, the airflow sensor is installed at the position of the air supply valve to monitor the changes in airflow velocity and flow rate. The microphone array is arranged inside the mask to collect internal airflow noise and external ambient noise. An ear pressure sensor is installed at the ear position to monitor the user's wearing comfort. Initialize all sensors, set the sampling frequency (44.1 kHz or 48 kHz), ensure the normal operation of the sensors and perform calibration to eliminate system errors. Then, synchronously collect the airflow pressure signal and the ambient noise signal through a data acquisition card. Preprocess the collected airflow pressure signal and ambient noise signal, including filtering, normalization, and feature extraction. Among them, use a low-pass filter to remove high-frequency noise and retain the main airflow characteristics. Adjust the signal amplitude to a unified range through normalization processing for subsequent analysis. Perform time synchronization on the airflow pressure signal and the ambient noise signal to ensure that the airflow pressure signal and the ambient noise signal are aligned in time. Then, use data fusion technology to integrate the airflow signal and the noise signal into a comprehensive data set for subsequent noise identification and noise reduction processing;

[0064] Step 2: Perform time-frequency analysis (short-time Fourier transform or wavelet transform) on the collected signals, extract the airflow spectrum characteristics, and use a multi-task learning framework with multi-state conversion to synchronously perform lightweight processing to improve the model processing efficiency. Use the short-time Fourier transform to perform time-frequency analysis on the preprocessed airflow pressure signal and ambient noise signal. Among them, the short-time Fourier transform is suitable for processing quasi-stationary signals and can provide good frequency resolution. The signal is segmented into multiple short time periods through a sliding window, and the Fourier transform is performed on each time period to analyze the change of the signal spectrum over time and obtain the spectrum diagram of the signal. After obtaining the time-frequency analysis result, extract the airflow spectrum characteristics from the spectrum diagram of the signal, including the main noise frequency components, the noise frequency band width, and the airflow fluctuation frequency. Use the airflow spectrum characteristics to reflect the characteristics of the airflow signal and the ambient noise. Construct a multi-task learning framework with multi-state conversion, set multiple task branches, which are respectively responsible for different processing tasks, including multiple processing tasks such as airflow spectrum feature extraction, noise identification, and noise reduction parameter adjustment. Through multi-task learning, simultaneously learn the characteristics of the airflow signal and the noise signal, and dynamically adjust the processing strategy according to the task requirements. Among them, under the multi-task learning framework, use pruning and quantization compression techniques to perform lightweight processing on the model. Remove the unimportant connections in the model through the pruning technique, and the quantization technique converts the model parameters from high-precision representation to low-precision representation, thereby reducing the storage space and computational amount of the model. Through lightweight processing, ensure that the model can still achieve efficient noise identification and noise reduction processing under limited computing power;

[0065] In addition, the process of obtaining the spectrogram of the signal includes:

[0066] Segment the preprocessed air flow pressure signal and ambient noise signal using a sliding window of a fixed length. Let the sliding window slide over the signal, moving a fixed step length each time, and divide the entire signal into multiple overlapping short signal segments. Approximate the non-stationary signal as a series of short-time stationary signals for subsequent analysis. Here, the length of the sliding window is set to 25 - 50 milliseconds, and the overlapping part between windows is set to about 50% to ensure the continuity of the signal and avoid information loss. The sliding window is used to divide the long signal into multiple short signal segments, and each short signal segment can be approximately regarded as a quasi-stationary signal, thus being suitable for short-time Fourier transform. By means of the sliding window, the window position can be gradually moved to cover the entire signal, ensuring that the signal in each time period can be analyzed. Perform Fourier transform on the short signal segments within each sliding window, converting them from the time domain to the frequency domain. Through Fourier transform, obtain the spectrum of each short signal segment, including the amplitude spectrum and the phase spectrum. The amplitude spectrum reflects the energy distribution of the signal at different frequencies. Arrange the amplitude spectra of each short signal segment in chronological order to generate the spectrogram of the signal. Here, the horizontal axis of the spectrogram represents time, the vertical axis represents frequency, and the gray scale represents the amplitude magnitude. Through the spectrogram of the signal, the change of the signal spectrum over time can be observed, clearly showing the appearance and disappearance of different frequency components in the signal, as well as the intensity change;

[0067] Step 3: Establish the mapping relationship between the working state of the ventilator and the noise spectrum based on the historical records, and perform targeted data matching and differential adjustment in combination with the differences between the upstream and downstream audio according to the mapping relationship between the working state of the ventilator and the noise spectrum. Collect the historical air flow pressure signals and ambient noise signals of the ventilator under different working states, and perform preprocessing on the signals and extraction operations on the air flow spectrum characteristics. Integrate to obtain the historical records. Here, the working state includes the supply pressure, the user's breathing frequency, and the breathing mode (inhalation, exhalation). Combine the historical records, analyze the changes in the noise spectrum under different working states, construct a noise database, and establish the mapping relationship between the working state of the ventilator and the noise spectrum. Analyze the differences between the upstream and downstream audio, compare the spectrum characteristics of the upstream and downstream audio, and identify the noise differences caused by different breathing air flow directions. The noise differences include amplitude differences, frequency differences, and phase differences. Here, the upstream and downstream audio respectively refer to the audio signals generated by the user's exhaled and inhaled air flows. According to the established mapping relationship and the differences between the upstream and downstream audio, perform targeted data matching and differential adjustment. For each working state, match the corresponding noise reduction parameters, including the bandwidth and filtering intensity of the filter, and optimize the audio fidelity by dynamically adjusting the noise reduction parameters. The differential adjustment is the separate processing of different frequency components. By enhancing the low-frequency air flow signal and suppressing the high-frequency noise, ensure that the noise-reduced audio signal minimizes noise interference while retaining useful information;

[0068] In addition, the analysis process of the differences between the uplink and downlink audio includes:

[0069] By arranging a high-precision airflow sensor and a microphone array inside the air supply valve and mask of the ventilator, the uplink and downlink audio signals generated by the user's exhaled and inhaled airflow are collected respectively, and the collected signals are preprocessed, including filtering to remove high-frequency noise and normalizing to adjust the signal amplitude to a unified range. The signal is processed in segments through a sliding window to ensure the continuity and stability of the signal. The time-frequency analysis is performed on the preprocessed uplink and downlink audio signals. The short-time Fourier transform is used to convert the signal from the time domain to the frequency domain, and the amplitude spectrum, frequency components, and phase information of each short signal segment are extracted. Among them, the amplitude spectrum reflects the energy distribution of the signal at different frequencies, the frequency components show the main frequencies of the signal, and the phase information provides the time characteristics of the signal. Compare the spectral characteristics of the uplink and downlink audio to identify the noise differences in amplitude, frequency, and phase caused by the different directions of the respiratory airflow. Among them, analyze the amplitude difference, observe the energy change of the uplink and downlink audio at the same frequency, and analyze its relationship with the direction of the respiratory airflow. Analyze the frequency difference, identify the different frequency components that appear in the uplink and downlink audio. Analyze the phase difference, study the phase change of the uplink and downlink audio over time. Through the difference analysis, determine the influence of the respiratory airflow direction on the noise characteristics. According to the identified differences between the uplink and downlink audio, perform targeted data matching. For each working state (air supply pressure, respiratory rate, respiratory mode), match the corresponding noise reduction parameters. Among them, in the inhalation stage, enhance the low-frequency airflow signal. In the exhalation stage, suppress the high-frequency noise. By dynamically adjusting the bandwidth and filtering intensity of the filter, ensure that noise can be effectively suppressed in different respiratory stages while retaining useful information. Optimize the audio fidelity through differential adjustment. In the inhalation stage, focus on enhancing the low-frequency airflow signal to ensure that the user can clearly feel the airflow. In the exhalation stage, focus on suppressing the high-frequency noise to reduce the interference of environmental noise, and use the POLQA (Perceptual Objective Listening Quality Assessment) score to evaluate the noise-reduced audio to ensure that it is better than the traditional algorithm;

[0070] Step 4. Under the multi-task learning framework of polymorphic transformation, a deep learning model of a CNN-LSTM hybrid network is used to construct a noise recognition model to distinguish useful airflow signals from environmental noise signals. An audio dataset containing useful airflow signals and environmental noise signals is extracted from historical records, ensuring that the data covers different working states and noise types, and the audio dataset is labeled to distinguish useful airflow signals from environmental noise signals. Then, the dataset is divided into a training set, a validation set, and a test set to ensure data distribution consistency. The deep learning model of the CNN-LSTM hybrid network is used as the basic architecture, and the signals in the audio dataset are converted into spectrograms as the hybrid input of the CNN-LSTM network. Among them, the CNN layer is used to automatically extract local features (frequency distribution, energy change) in the spectrogram of the signal, reducing parameter redundancy. The LSTM layer is used to capture long-term dependencies in the feature sequence, model the temporal dynamics of the airflow signal, and dynamically adjust the network structure (convolution kernel size, number of LSTM units) through the polymorphic transformation mechanism to adapt to different noise patterns. The training set is used to train the CNN-LSTM hybrid network. The cross-entropy loss function is defined to measure the difference between the prediction and the true label. The Adam optimizer is used to adjust the network parameters, and the learning rate is dynamically adjusted in combination with the performance of the validation set. The batch size and regularization parameters are adjusted through grid search or random search to improve the generalization ability of the model, and the early stopping method is introduced to terminate the training when the validation set loss does not decrease continuously to prevent overfitting. Then, the test set is used to evaluate the trained model, and the accuracy, recall rate, and F1-score metrics are calculated to quantify the performance of the model. The misclassification situation is analyzed through the confusion matrix, and the model is optimized accordingly. Finally, a trained noise recognition model is obtained. The trained noise recognition model is integrated into the positive pressure air ventilator to ensure low-latency inference ability. The real-time airflow pressure signal and environmental noise signal are input, converted into a spectrogram and then input into the noise recognition model. Then, the noise recognition model outputs the classification result, calculates the classification score, analyzes the probability that the signal belongs to the useful airflow signal, and distinguishes the useful airflow signal from the environmental noise signal in real time;

[0071] In addition, the calculation process of the classification score includes:

[0072] The airflow pressure signal and environmental noise signal are collected in real time, the signal is converted into a spectrogram of the signal through short-time Fourier transform, and the energy value of each time-frequency point is calculated from the spectrogram of the signal. The energy value of each time-frequency point is normalized using the reference energy value, where the reference energy value is obtained through statistical analysis of historical data. The normalized energy values are weighted and summed, a time decay factor is introduced to simulate the attenuation of the signal over time, and combined with the weighted sum result, the weighted sum result is mapped to the range [0, 1] through an exponential function and a logistic regression function to obtain the classification score;

[0073] The expression of the classification score is:

[0074]

[0075] E(t, f) = |X(t, f)| 2 ;

[0076]

[0077] In the formula, S(t) is the classification score, representing the probability that the signal belongs to the useful airflow signal at time t, F is the number of frequency points in the spectrogram, w f is the weight of frequency f, representing the importance of this frequency in classification, E(t, f) is the energy value at time t and frequency f, representing the energy value of each time-frequency point in the spectrogram calculated by the short-time Fourier transform (STFT), B(f) is the reference energy value at frequency f for normalization, α is the time decay factor, representing the attenuation degree of the signal over time, |X(t, f)| is the amplitude of the spectrum, x(n) is the input signal, w(n - t) is the sliding window function, N is the number of points for Fourier transform, t is the time index, f is the frequency index, j is the imaginary unit. The value range of S(t) is between 0 and 1. When S(t) > 0.5, the signal is classified as a useful airflow signal. When S(t) ≤ 0.5, the signal is classified as environmental noise signal;

[0078] Step Five: Introduce a multi-module supervision strategy and design a grouping strategy to optimize model deployment to ensure that the algorithm can still achieve a leading noise reduction effect under limited computing power;

[0079] Step Six: Introduce a feedback mechanism, analyze the user's wearing comfort through ear pressure sensor data and optimize the model parameters to ensure the balance between noise reduction effect and wearing experience.

[0080] Example 2, as Figure 1 , Figure 2 shown. Based on Example 1, the present invention provides a technical solution: Preferably, Step Five specifically includes:

[0081] The noise recognition model is divided into multiple sub-modules, including a feature extraction module, a feature fusion module, and a classification module. Supervision points are set at the output end of each sub-module to monitor the output quality and performance of the sub-module. Among them, the supervision point uses the cross-entropy loss function to measure the difference between the intermediate output and the target value, and updates the parameters through the backpropagation algorithm. Through the multi-module supervision strategy, it is ensured that each module can be independently optimized, improving the generalization ability and robustness of the overall model. Evaluate the computing power limitations of the positive pressure air ventilator, including the available resources of the CPU, GPU, and memory. Determine the computing power and memory limitations of different hardware through performance testing. According to the evaluation results, allocate the computationally intensive feature extraction module to the GPU, allocate the feature fusion module for time series processing to the CPU, and allocate the lightweight classification module to a dedicated FPGA to ensure that each sub-module runs on the most suitable hardware, reducing the computing latency. According to the computing power requirements and grouping strategy, divide the sub-modules into multiple groups. Group the feature extraction module and the feature fusion module together and run them on the GPU, and divide the classification module into another group and run it on the CPU. Optimize the communication mechanism between sub-modules, adopt data compression technology and asynchronous communication mechanism to reduce the overhead of data transmission, and ensure efficient and low-latency data transmission between modules through shared memory and message queues;

[0082] Step six specifically includes:

[0083] In combination with the ear pressure sensor, the ear pressure data during the user's wearing process is collected in real time, and the collected ear pressure data is preprocessed, including filtering to remove high-frequency noise and normalizing to adjust the signal amplitude to a unified range. Calculate the statistical indicators of the mean, variance, and peak value of the ear pressure as the quantitative indicators of the user's wearing comfort, and analyze the change trend of the ear pressure statistical indicators, calculate the wearing comfort score, and evaluate the user's wearing comfort at different noise reduction intensities. According to the evaluation results of the wearing comfort, dynamically adjust the noise reduction parameters. If the mean value of the ear pressure is too high, reduce the noise reduction intensity to reduce the stimulation to the user's ears. If the peak value is too high, adjust the bandwidth of the filter to avoid excessive amplification of high-frequency noise. Through the feedback mechanism, ensure the balance between the noise reduction effect and the wearing experience;

[0084] The expression of the wearing comfort score is:

[0085]

[0086] In the formula, C(t) is the wearing comfort score, indicating the comfort of the user's wearing at time t, P k(t) is the ear pressure value collected by the k-th ear pressure sensor at time t, with the unit of Pascal (Pa). B is the reference pressure value, obtained through historical data statistics, with the unit of Pascal (Pa). μ is the mean value of the ear pressure, reflecting the comfort level during long-term wearing. σ is the variance of the ear pressure, reflecting the stability of the pressure change. K is the number of ear pressure sensors, taking the value of 2, one for each of the left and right ears. β is the peak weight, indicating the degree of influence of the peak on the comfort level. max(P k (t)) is the peak value of the ear pressure at time t, reflecting the instantaneous discomfort, with the unit of Pascal (Pa). The value range of C(t) is between 0 and 1. When C(t) is close to 1, it indicates a higher wearing comfort level. When C(t) is close to 0, it indicates a lower wearing comfort level.

[0087] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.

Claims

1. A positive pressure air ventilator adaptive noise reduction method based on the airflow spectrum characteristics, characterized in that, It includes the following steps: Step 1: Synchronously collect the pressure signal of the ventilator air circuit and the ambient noise signal through an airflow sensor and a microphone array; Step 2: Perform time-frequency analysis on the collected signals to extract the airflow spectrum characteristics; Step 3: Establish a mapping relationship between the working state of the ventilator and the noise spectrum based on historical records, and perform targeted data matching and differential adjustment according to the mapping relationship between the working state of the ventilator and the noise spectrum, combined with the differences in uplink and downlink audio; Step 4: Use a deep learning model of a CNN-LSTM hybrid network to construct a noise recognition model to distinguish useful airflow signals from ambient noise signals; Step 5: Introduce a multi-module supervision strategy and design a grouping strategy to optimize model deployment; Step 6: Introduce a feedback mechanism to analyze the user's wearing comfort through ear pressure sensor data and optimize model parameters.

2. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 1, characterized in that: The specific content of Step 1 includes: Deploy a high-precision airflow sensor and a microphone array in a positive-pressure air ventilator. Among them, the airflow sensor is installed at the position of the air supply valve, the microphone array is arranged inside the mask, and an ear pressure sensor is installed at the ear position; Initialize all sensors, set the sampling frequency, and then synchronously collect the airflow pressure signal and the ambient noise signal through a data acquisition card; Preprocess the collected airflow pressure signal and ambient noise signal, including filtering, normalization, and feature extraction; Synchronize the airflow pressure signal and the ambient noise signal in time, and then use data fusion technology to integrate the airflow signal and the noise signal into a comprehensive data set.

3. The adaptive noise reduction method for a positive pressure air ventilator based on airflow spectrum characteristics according to claim 2, wherein: The specific content of Step 2 includes: Perform time-frequency analysis on the preprocessed airflow pressure signal and ambient noise signal using the short-time Fourier transform; Divide the signal into multiple short time periods through a sliding window, and perform Fourier transform on each time period to analyze the change of the signal spectrum over time, and obtain the spectrogram of the signal; After obtaining the time-frequency analysis result, extract the airflow spectrum characteristics from the spectrogram of the signal, including the main noise frequency components, the noise frequency band width, and the airflow fluctuation frequency; Construct a multi-task learning framework with multi-state conversion, set multiple task branches, and be responsible for different processing tasks respectively, including multiple processing tasks such as airflow spectrum feature extraction, noise recognition, and noise reduction parameter adjustment.

4. The positive pressure air ventilator adaptive noise reduction method based on airflow spectrum characteristics according to claim 3, characterized in that: The acquisition process of the spectrogram of the signal includes: Use a sliding window with a fixed length to segment the preprocessed airflow pressure signal and ambient noise signal, make the sliding window slide on the signal, move a fixed step length each time, divide the entire signal into multiple overlapping short signal segments, and approximate the non-stationary signal as a series of short-time stationary signal segments; Perform Fourier transform on the short signal segments within each sliding window, convert it from the time domain to the frequency domain, and obtain the spectrum of each short signal segment through Fourier transform, including the amplitude spectrum and the phase spectrum; Arrange the amplitude spectra of each short signal segment in chronological order to generate the spectrogram of the signal. Among them, the horizontal axis of the spectrogram represents time, the vertical axis represents frequency, and the gray scale represents the amplitude size.

5. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 4, wherein: The specific content of Step 3 includes: Collect the historical airflow pressure signals and ambient noise signals of the ventilator under different working conditions, preprocess the signals and extract the airflow spectrum characteristics, and integrate to obtain the historical record. Among them, the working conditions include the supply pressure, the user's breathing frequency, and the breathing mode; Combine the historical record, analyze the changes in the noise spectrum under different working conditions, construct a noise database, and establish a mapping relationship between the working state of the ventilator and the noise spectrum; Analyze the differences between the upstream and downstream audio, compare the spectrum characteristics of the upstream and downstream audio, and identify the noise differences caused by different breathing airflow directions. The noise differences include amplitude differences, frequency differences, and phase differences. Among them, the upstream and downstream audio respectively refer to the audio signals generated by the user's exhaled and inhaled airflow; According to the established mapping relationship and the differences between the upstream and downstream audio, perform targeted data matching and differential adjustment. For each working state, match the corresponding noise reduction parameters, including the bandwidth and filtering intensity of the filter, and optimize the audio fidelity by dynamically adjusting the noise reduction parameters. The differential adjustment is the separate processing of different frequency components, by enhancing the low-frequency airflow signal and suppressing the high-frequency noise.

6. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 5, wherein: The analysis process of the differences between the upstream and downstream audio includes: Arrange high-precision airflow sensors and microphone arrays inside the supply valve and mask of the ventilator to collect the upstream and downstream audio signals generated by the user's exhaled and inhaled airflow respectively, and preprocess the collected signals; Perform time-frequency analysis on the preprocessed upstream and downstream audio signals. Use the short-time Fourier transform to convert the signals from the time domain to the frequency domain, and extract the amplitude spectrum, frequency components, and phase information of each short signal segment. Among them, the amplitude spectrum reflects the energy distribution of the signal at different frequencies, the frequency components show the main frequencies of the signal, and the phase information provides the time characteristics of the signal; Compare the spectrum characteristics of the upstream and downstream audio, and identify the noise differences of amplitude differences, frequency differences, and phase differences caused by different breathing airflow directions. Among them, analyze the amplitude differences, observe the energy changes of the upstream and downstream audio at the same frequency, and analyze their relationship with the breathing airflow direction. Analyze the frequency differences, identify the different frequency components that appear in the upstream and downstream audio, and analyze the phase differences, study the phase changes of the upstream and downstream audio in time; According to the identified differences between the upstream and downstream audio, perform targeted data matching. For each working state, match the corresponding noise reduction parameters. Among them, in the inhalation phase, enhance the low-frequency airflow signal, and in the exhalation phase, suppress the high-frequency noise; Optimize the audio fidelity through differential adjustment. In the inhalation phase, focus on enhancing the low-frequency airflow signal, and in the exhalation phase, focus on suppressing the high-frequency noise.

7. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 6, characterized in that: The specific steps of step four include: Extract the audio data set containing useful airflow signals and ambient noise signals from the historical record, annotate the audio data set to distinguish useful airflow signals from ambient noise signals, and then divide the data set into a training set, a validation set, and a test set; Use the deep learning model of the CNN-LSTM hybrid network as the basic architecture, convert the signals in the audio data set into spectrograms, and use them as the hybrid input of the CNN-LSTM network; The CNN-LSTM hybrid network is trained using the training set. The cross-entropy loss function is defined to measure the difference between the prediction and the true label. The Adam optimizer is used to adjust the network parameters, and the learning rate is dynamically adjusted in combination with the performance of the validation set. Then, the trained model is evaluated using the test set, and the accuracy, recall, and F1-score metrics are calculated to quantify the performance of the model, thereby obtaining a trained noise recognition model. The trained noise recognition model is integrated into the positive pressure air ventilator. The real-time air flow pressure signal and the environmental noise signal are input, converted into a spectrogram and then input into the noise recognition model. Then, the noise recognition model outputs the classification result, calculates the classification score, analyzes the probability that the signal belongs to the useful air flow signal, and distinguishes the useful air flow signal and the environmental noise signal in real time.

8. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 7, characterized in that: The calculation process of the classification score includes: The air flow pressure signal and the environmental noise signal are collected in real time, and the signal is converted into a spectrogram of the signal through short-time Fourier transform, and the energy value of each time-frequency point is calculated from the spectrogram of the signal. The energy value of each time-frequency point is normalized using the reference energy value, where the reference energy value is obtained by statistical analysis of historical data, and the normalized energy values are weighted and summed. A time decay factor is introduced to simulate the attenuation of the signal over time, and in combination with the weighted sum result, the weighted sum result is mapped to the range [0, 1] through an exponential function and a logistic regression function to obtain the classification score.

9. The adaptive noise reduction method of the positive pressure air ventilator based on the airflow spectrum characteristics according to claim 8, characterized in that: The specific content of step five includes: The noise recognition model is divided into multiple sub-modules, including a feature extraction module, a feature fusion module, and a classification module, and supervision points are set at the output end of each sub-module to monitor the output quality and performance of the sub-module. Among them, the supervision point uses the cross-entropy loss function to measure the difference between the intermediate output and the target value, and updates the parameters through the backpropagation algorithm. The computing power limitations of the positive pressure air ventilator are evaluated, including the available resources of the CPU, GPU, and memory. The computing capabilities and memory limitations of different hardware are determined through performance testing. According to the evaluation results, the computationally intensive feature extraction module is assigned to the GPU, the time-series processing feature fusion module is assigned to the CPU, and the lightweight classification module is assigned to a dedicated FPGA. According to the computing power requirements and the grouping strategy, the sub-modules are divided into multiple groups. The feature extraction module and the feature fusion module are divided into one group and run on the GPU, and the classification module is divided into another group and run on the CPU.

10. The positive pressure air ventilator adaptive noise reduction method based on airflow spectrum characteristics according to claim 9, wherein: The specific content of step six includes: Combined with an ear pressure sensor, the ear pressure data during the user's wearing process is collected in real time, and the collected ear pressure data is preprocessed, including filtering to remove high-frequency noise and normalizing to adjust the signal amplitude to a unified range. The statistical indicators of the mean, variance, and peak value of the ear pressure are calculated as the quantitative indicators of the user's wearing comfort, and the change trend of the ear pressure statistical indicators is analyzed, the wearing comfort score is calculated, and the user's wearing comfort under different noise reduction intensities is evaluated. According to the evaluation result of the wearing comfort, the noise reduction parameters are dynamically adjusted. If the mean value of the ear pressure is too high, the noise reduction intensity is reduced. If the peak value is too high, the bandwidth of the filter is adjusted.

Images