Method and system for analyzing high-altitude flight respiratory characteristic parameters and predicting respiratory change trend based on deep learning

Abstract

The application provides a method and system for high-altitude flight respiratory characteristic parameter analysis and respiratory change trend prediction based on deep learning, a respiratory characteristic parameter analysis and change trend prediction model based on deep learning, extraction of relevant time sequence waveform parameter features according to respiratory data collected by a pressure sensor and a displacement sensor, and then prediction of the change trend of human respiratory characteristics through a feature fusion algorithm in multiple modes based on discriminant canonical correlation analysis of respiratory features. Meanwhile, in order to realize the prediction of the respiratory change trend, each feature vector of the respiratory parameter sequence is converted into the input and output of the network in a sliding window mode, then deep learning network is adopted for data modeling and prediction, and the iterative prediction mode is used to obtain a long-term prediction fitting result, and finally a respiratory index prediction model based on deep learning is established, thereby providing a standardized reference for the monitoring and adjustment of the respiratory signal of the pilot in high-altitude flight.

Description

Technical Field

[0001] This invention relates to the fields of machine learning and artificial intelligence, and in particular to a method and system for analyzing respiratory characteristic parameters and predicting respiratory change trends in high-altitude flight based on deep learning. Background Technology

[0002] The impact of high-altitude air pressure on human flight activities is significant, as changes in air pressure can trigger a series of respiratory physiological responses in the human body. Intensive changes, prolonged exposure, or rapid shifts can lead to severe physiological effects. For high-altitude flight, exceeding a pilot's tolerance limits can result in flight incapacitation and impaired flight control. More seriously, changes in air pressure can cause pathological changes in respiration, leading to functional impairments or even organic damage.

[0003] During high-altitude flights, aviation oxygen masks play an immeasurable role in preventing the threat of hypoxia and ensuring pilot safety. Among the many tasks of oxygen masks, the most important is recording the pilot's breathing signals. Therefore, continuous monitoring and automatic analysis of breathing signals, understanding the changing trends of human respiration, and proactively predicting subsequent breathing processes based on these trends, allows for proactive regulation of the breath. This not only improves the oxygen utilization rate of positive pressure oxygen masks but also enhances wearing comfort. It is also crucial for achieving respiratory protection, providing significant reference value for ensuring the recovery and maintenance of pilots' combat effectiveness and accurately monitoring respiratory physiology. Furthermore, the analysis and prediction of respiratory characteristic parameters provide timely and effective data and decision support for the establishment of intelligent respiratory monitoring and protection systems, possessing significant practical application value. Summary of the Invention

[0004] The technical problem to be solved by this invention is to address the issue of high-altitude respiratory monitoring by using pressure and displacement sensors to collect respiratory data, establishing an active prediction model for the subsequent respiratory process of the human body based on respiratory feature learning and adaptive prediction algorithms, revealing the trend of respiratory changes, achieving the goal of active regulation of breathing, and providing a standardized reference for the monitoring and regulation of respiratory signals by pilots during high-altitude flight.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] This invention proposes a method for analyzing and predicting the changing trends of respiratory characteristic parameters in high-altitude flight based on deep learning, including:

[0007] Step S1: Collect the pilot's breathing data during high-altitude flight and extract relevant time series waveform parameter features;

[0008] Step S2: Perform canonical correlation analysis on the parameter features extracted in step S1, and then perform feature fusion under two modes, including manual feature extraction and STFT transform deep learning feature extraction, and obtain the human respiratory characteristic transformation curve through predictive analysis.

[0009] Step S3: Convert each feature vector of the respiratory parameter sequence into the input and output of a deep learning network through a sliding window, and then use a deep learning network CNN for data modeling and prediction to obtain an active prediction model for respiratory changes.

[0010] Step S4: Based on the active prediction model of respiratory changes, a respiratory index prediction model based on the gated recurrent unit (GRU) is further established, and a longer-term prediction fitting result is obtained by using iterative prediction, which is used for monitoring and regulating the respiratory signals of pilots during high-altitude flight.

[0011] This invention also proposes a deep learning-based system for analyzing respiratory characteristic parameters and predicting their changing trends, comprising:

[0012] The feature extraction module is used to collect the pilot's breathing data during high-altitude flight and extract relevant time series waveform parameter features;

[0013] The analysis and fusion module is used to perform canonical correlation analysis on the extracted parameter features, and then perform feature fusion in two modes, including manual feature extraction and STFT transform deep learning feature extraction, to obtain the human respiratory characteristic transformation curve through predictive analysis.

[0014] The deep learning module is used to transform the feature vectors of the respiratory parameter sequence into the input and output of a deep learning network through a sliding window. Then, a deep learning network (CNN) is used for data modeling and prediction to obtain an active prediction model of respiratory changes.

[0015] The prediction and regulation module is used to further establish a respiratory index prediction model based on the active prediction model of respiratory changes and GRU, and to obtain longer-term prediction fitting results through iterative prediction, which is used for respiratory signal monitoring and regulation of pilots during high-altitude flight.

[0016] The present invention also proposes an electronic system comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method proposed in the present invention.

[0017] Finally, the present invention proposes a computer-readable storage medium storing computer instructions, characterized in that the computer instructions are used to cause the computer to execute the method proposed in the present invention.

[0018] The present invention adopts the above technical solution, and its beneficial technical effects compared with the prior art are as follows:

[0019] This invention uses pressure and displacement sensors to collect respiratory data and employs artificial intelligence algorithms to analyze respiratory signals, thereby accurately and effectively predicting respiratory change trends and revealing these trends. This achieves the goal of actively regulating respiration, which will provide important guidance for the monitoring and regulation of respiratory signals during high-altitude flight by pilots in aerospace development. At the same time, it can also provide timely and effective basis and decision support for the establishment of intelligent respiratory monitoring systems and intelligent respiratory protection systems, and has significant practical application significance. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of a ventilator.

[0021] Figure 2 This is the training progress diagram of the model proposed in this invention.

[0022] Figure 3 This is the training completion diagram of the present invention.

[0023] Figure 4 It is a comparison chart of predicted values ​​and true values ​​and an error chart obtained using the present invention.

[0024] Figure 5 This is a comparison chart of the predicted and true values ​​and a magnified view of the error in this invention.

[0025] Figure 6 This is a flowchart of the feature fusion process of the present invention. Detailed Implementation

[0026] The following is a detailed description of the invention in conjunction with the accompanying drawings and specific embodiments: In order to better understand the above technical solutions, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] This invention proposes a deep learning-based method for analyzing and predicting respiratory characteristic parameters and their changing trends. First, a deep learning-based model for analyzing and predicting respiratory characteristic parameters and their changing trends is established. This model extracts relevant time-series waveform parameter features from respiratory data collected by pressure and displacement sensors. Then, a feature fusion algorithm based on discriminant canonical correlation analysis is designed using respiratory features to predict the changing trends of human respiratory characteristics. Simultaneously, to achieve the prediction of respiratory change trends, this invention transforms the feature vectors of the respiratory parameter sequence into network inputs and outputs using a sliding window approach. A deep learning network is then used for data modeling and prediction, and an iterative prediction method is employed to obtain long-term prediction fitting results. Finally, a deep learning-based respiratory index prediction model is established, providing a standardized reference for the monitoring and regulation of respiratory signals during high-altitude flight by pilots.

[0028] The present invention includes the following steps in its implementation:

[0029] I. Data Analysis

[0030] Pressure sensors are used to collect signals of airway ventilation pressure to ensure proper airway pressure output. In ventilators, the principle of continuous positive airway pressure (CPAP) is to maintain the astronaut's airway patency during breathing by providing appropriate continuous positive airway pressure. The purpose of the ventilator pressure sensor is to collect the pressure difference signal during the patient's breathing through a specially designed differential pressure tube, thus serving as the front-end signal for respiratory detection. The principle diagram for data acquisition using pressure and displacement sensors is shown below. Figure 1 As shown.

[0031] During inhalation, the upper airway pressure is lower than the ventilator's ventilation pressure due to the inhalation action, causing the valve opening to increase and allowing outside air to flow into the mask through the mask tubing. Conversely, during exhalation, the airway pressure exceeds the ventilator's output pressure, causing the valve to shift less, reducing the ventilator's output pressure to ensure it matches the preset setting.

[0032] Outlet pressure data is positively correlated with ventilatory flow rate. Respiratory tracking and respiratory event detection are ultimately achieved through the airflow velocity at outlet pressure, which is a crucial foundation for respiratory monitoring. Therefore, accurately obtaining outlet pressure data is essential, as predicting outlet pressure values ​​can help predict respiratory trends.

[0033] II. Analysis of Respiratory Characteristic Parameters

[0034] 2.1 Time-domain respiratory characteristic parameters

[0035] Time-domain signals are the original representation of signals and the most intuitive way to observe their time-domain waveforms. They have clear physical meaning and require minimal computation. Therefore, this invention employs time-domain analysis methods for studying breathing patterns. The time-domain characteristics of the breathing pattern classification used in this invention are as follows:

[0036] (1) Respiratory rate: The respiratory rate is characterized by the interval f between two respiratory peaks. b (i) Determined, defined as:

[0037]

[0038] Where t is peak The time of the peak respiratory signal, i.e., the respiratory characteristic curve C composed of the respiratory rate. br That is:

[0039] C br =[f b (1),f b (2),f b (3),…,f b (N)]

[0040] Where i = 1, 2, ... N, and N represents the number of times the breath was detected.

[0041] (2) Peak value of respiratory signal: The more rapid the breathing, the higher the peak value. The respiratory characteristic curve defined by the peak value of the respiratory signal is C. bp :

[0042] C bp =[V prak (1),V peak (2),V peak (3),…,V peak (N)]

[0043] Among them, V peak This represents the peak value of the respiratory signal.

[0044] (3) Respiratory airflow: Define the area of ​​the respiratory pulse as S, and the airflow C of the test subject in one breath. s for:

[0045] C s =[S(1),S(2),S(3),…,S(N)]

[0046] (4) Respiratory signal pulse width: The respiratory signal is represented by pulses with varying intervals and widths. The variation in intervals is introduced by changes in the pause time between breaths, while the width Ω b The change is introduced by the alteration of the duration of respiration:

[0047] Ω b=FFT[s b (t)]

[0048] Where s b For the acquired respiratory signal, the pulse width is measured by the distance between two characteristic peaks in the frequency domain, and is defined as:

[0049] D ω =ω1-ω2

[0050] Where ω 1(i＝1,2) The characteristic peaks are due to the shape of the respiratory signal. Therefore, the frequency domain characteristic curve C of the corresponding respiratory signal pulse width is... D That is, it can be expressed as:

[0051] C D =[D Ω (1),D Ω (2),D Ω (3),…,D Ω (N)]

[0052] Among them, D Ω This indicates the pulse width of the respiratory signal.

[0053] (5) Tidal volume (TV): The amount of air inhaled or exhaled each time at rest, which is related to age, sex, body surface area, breathing habits, and body metabolism.

[0054] (6) Tidal volume (VE): refers to the total volume of air inhaled or exhaled from the respiratory organs per minute at rest. MV = tidal volume (VT) × respiratory rate (f).

[0055] (7) Inspiratory and expiratory rate: close to the physiological respiratory rate. Tidal volume * Inspiratory and expiratory rate = Minute ventilation.

[0056] (8) Inspiratory-to-expiratory ratio: generally 1:1.5 to 2. For obstructive ventilatory disorders, it can be adjusted to 1:3 or longer expiratory time. For restrictive ventilatory disorders, it can be adjusted to 1:1.

[0057] (9) Number of breaths:

[0058] Different abnormal breathing patterns exhibit varying respiratory rates within the same time frame, and these rates also differ across different time intervals within a minute. Therefore, respiratory rate effectively reflects the differences between different breathing patterns. A respiratory event consists of one exhalation and one inhalation; thus, the respiratory rate *num* is half the number of breathing processes detected by the algorithm.

[0059] (10) Mean respiratory cycle

[0060] (Breathe Cycle, or average respiratory cycle) is the time interval between the start points of adjacent respiratory events. The respiratory cycle is calculated as follows: the start position of a respiratory event is detected by a respiratory event detection algorithm, and the time interval between the start positions of adjacent respiratory events is the respiratory cycle. The formula for calculating the average respiratory cycle is as follows:

[0061] bc(k) = nx(k+1) - nx(k)

[0062]

[0063] Where nx(k) represents the starting position of the k-th breathing event, and N is the total number of detected breathing events.

[0064] 2.2 Frequency Domain Respiratory Characteristic Parameters

[0065] Frequency domain analysis can describe the energy distribution of complex respiratory rate fluctuation signals according to different frequency bands, and can analyze the effects of various physiological factors after appropriate separation, thus exhibiting sensitivity and accuracy. Therefore, this invention performs wavelet transform and Fourier transform on the respiratory signal to extract corresponding features such as wavelet transform coefficients, LFn (Normalized Low-Frequency Power), HFn (Normalized High-Frequency Power), and LF / HF, and performs spectral analysis on the respiratory signal to calculate the spectral parameters of the window signal, including the spectral centroid and centroid frequency.

[0066] The short-time Fourier transform is defined as follows:

[0067] Ω n (e jw )=∑x(m)w(nm)e (-jωm)

[0068] Where x(m) is the input respiratory signal, and w(nm) is the window function. The features obtained using the short-time Fourier transform can be expressed as:

[0069] C STFT =[Ω1(e jw ),Ω2(e jw ),Ω3(e jw ),…,Ω N (e jw )]

[0070] Where jw is the operator of the short-time Fourier transform.

[0071] III. Using a deep learning network (CNN) for data modeling and prediction, an active prediction model for respiratory changes is obtained.

[0072] 3. Improved CNN Algorithm Based on Feature Fusion

[0073] This invention employs a CNN (Convolutional Neural Network) model, a deep learning model used to process data with a grid structure (such as images, audio, and text). Through convolution and pooling operations, it extracts features and reduces dimensionality from the input data, and finally achieves regression prediction through fully connected layers. Therefore, a deep learning model composed of convolution, pooling, fully connected layers, and activation functions can automatically learn the feature representation of the input data (respiratory parameters), and is trained and optimized through backpropagation to ultimately obtain an active prediction model for respiratory changes. (Reference) Figure 6 As shown, this invention fuses the respiratory characteristic parameters and deep learning features extracted from each respiratory signal and performs regression prediction.

[0074] In this feature fusion method, 2-second segments extracted from the respiratory signal are individually numbered. This facilitates the later fusion of various features generated from each signal, preventing the network from scrambling the signal order during training, which would otherwise prevent the generated features from being fused in sequence. For example, the nth signal undergoes manual feature extraction, and the resulting 2D image is then fed into a convolutional neural network to obtain deep learning features. Furthermore, this method primarily employs noise removal and inversion correction methods for signal preprocessing. These preprocessing steps reduce noise interference in the signal and correct signal inversion issues, providing more accurate and reliable data for subsequent feature extraction and fusion.

[0075] (1) Extraction of manual respiratory features

[0076] In the manual feature extraction section, respiratory characteristic change parameters were used as features for manual extraction, and the features were filtered based on the final results, ultimately retaining 29 manual features.

[0077] (2) Deep learning feature extraction

[0078] In the deep learning feature extraction part, the time-domain image of the respiratory signal's STFT transform, which performed well in deep convolutional neural network classification experiments, was adopted as the model for deep learning feature extraction. The time-domain image of the respiratory signal's STFT transform was processed through a deep convolutional neural network and exported before the final Softmax classification operation, resulting in 128-dimensional deep learning features of the respiratory signal. To extract deep learning features, the last layer, the softmax layer, was removed from the original network. Therefore, the penultimate layer—the fully connected layer—was directly used as the vector output. The vector obtained from the modified network is the deep learning feature of the signal.

[0079] (3) Feature fusion algorithm implementation

[0080] In this study, the primary fusion method considered was pre-classification fusion. Pre-classification fusion refers to first merging and summarizing various features, then using the fused features as a feature matrix, and finally classifying them using a machine learning algorithm to obtain the experimental results of the fusion model.

[0081] For the pre-classification fusion method, after the features from both manually extracted and STFT-transformed deep learning features are extracted sequentially, the manually extracted features and deep learning features of each respiratory signal are then summarized and arranged by number to ensure that the features correspond to the signals. This results in a data matrix consisting of 157 features. This matrix is ​​then fed into the model for machine learning classification.

[0082] (4) Training and experimental results of CNN convolutional neural network model

[0083] When 70% of the respiratory data was randomly used as the training set and 30% as the test set, the model's MSE was 92.40%. The parameter settings for each layer during model training are shown in Table 1.

[0084] Table 1: Model Training Parameter Settings

[0085]

[0086]

[0087] Training results as follows Figure 2 , 3 As shown, Figure 2 and Figure 3 The diagram illustrates the model training process and results. The horizontal axis of both curves represents the number of iterations, while the vertical axes represent the root mean square error (RMSE) and mean squared error (MSE), respectively. MSE serves as the loss function during model training. The loss function curves show that the model converges quickly and has a good fit. At the end of training, the final value of the loss function is 8.9184e-05.

[0088] The comparison chart of predicted values ​​and true values, and the error chart are shown below. Figure 4 . Figure 4 The chart is divided into two parts: the top chart shows the horizontal axis representing over 40,000 test data sets, and the vertical axis representing the oxygen supply outlet pressure. This chart compares the model's predicted values ​​with the corresponding ground truth values, with blue representing the model's predictions and red representing the ground truth values. The bottom chart has the same meaning for both axes, but represents the difference between the model's predictions and the ground truth values. It is evident that the model performs well on the test set. A magnified view is provided below. Figure 5 . Figure 4 and Figure 5 The results are similar, but the local area has been magnified, allowing for a more intuitive observation of how close the predicted and true values ​​are. As shown in the figure, the predicted results match the true values ​​very well, demonstrating the model's good performance.

[0089] IV. Establish a respiratory index prediction model based on the gated circulation unit (GRU) and use iterative prediction to obtain longer-term prediction fitting results.

[0090] This invention innovatively improves the connection weights between neurons to reduce the error between the actual and expected output of the system. It also integrates traditional threshold rules into high-dimensional features extracted by deep learning, establishing a respiratory prediction model that synergistically combines rules and deep learning. This approach yields interpretable, accurate, and predictable experimental data. Therefore, by applying BP neural networks, CNNs, and attention-based GRU neural network models to analyze the impact of respiratory inlet pressure, displacement, and cabin pressure on outlet pressure, as well as the correlation and differences in the time dimension of different respiratory states in time series data, this invention fully explores the information contained in the time series and considers the dependencies between different indicator data. Finally, by predicting outlet pressure, it predicts future respiratory trends.

[0091] The GRU (Gate Recurrent Unit) neural network model can better uncover the relationships between raw respiratory data, ensuring that important information is not lost, improving prediction accuracy, and making it more suitable for monitoring and regulating pilots' respiratory signals during high-altitude flight. The training data for this GRU neural network involves setting several hyperparameters, including the number of neurons (m), time step (T), and batch size. The number of neurons determines the degree of fit of the neural network, while the time step and batch size directly affect the model training results.

[0092] By comparing prediction indicators such as normalized oxygen inlet pressure, normalized displacement data, predicted oxygen outlet pressure, and oxygen outlet pressure label, a relatively accurate prediction of respiratory rhythm was obtained in the model. The indicator parameters obtained from training the model are shown in the figure below: the left side of the figure is the training set, and the right side is the test set; the first and second rows are the data input, the third row is the prediction result, and the fourth row is the true label. x-axis: sampling point. y-axis: 1) normalized oxygen inlet pressure; 2) normalized displacement data; 3) predicted oxygen outlet pressure; 4) oxygen outlet pressure label (original data without unit label). According to the experimental results, the respiratory prediction accuracy within 5 seconds was 87.65%, 92.40%, and 89.23%, respectively, which met the experimental requirements.

[0093] This invention continuously monitors and automatically analyzes respiratory signals to understand the changing trends of human respiration. By proactively predicting subsequent respiratory processes based on these trends, it achieves active regulation of the human respiratory system. This not only improves the oxygen utilization rate of positive pressure oxygen masks but is also crucial for respiratory protection. Therefore, this invention aligns with the nation's major strategic needs in aerospace, utilizing artificial intelligence algorithms to analyze respiratory signals and accurately and effectively predict respiratory trends. This provides vital data support and standardized reference for ensuring the physiological respiratory signals of pilots during high-altitude flights.

[0094] Example 2: This example proposes a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method described in this invention, which will not be repeated here.

[0095] Example 3: This example also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in this embodiment of the invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this example can be found in the method provided in this embodiment of the invention.

[0096] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0097] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0098] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for analyzing and predicting the changing trends of respiratory characteristic parameters in high-altitude flight based on deep learning, characterized in that, include: Step S1: Collect the pilot's breathing data during high-altitude flight and extract relevant time series waveform parameter features; The system collects respiratory data based on pressure and displacement sensors. The pressure sensor is used to collect the airway ventilation pressure signal, and the differential pressure signal during the patient's breathing is collected through a differential pressure tube, which serves as the front-end signal for respiratory detection. Step S2: Perform canonical correlation analysis on the parameter features extracted in step S1, and then perform feature fusion under two modes, including manual feature extraction and STFT transform deep learning feature extraction, and obtain the human respiratory characteristic transformation curve through predictive analysis. Step S3: Convert each feature vector of the respiratory parameter sequence into the input and output of a deep learning network through a sliding window, and then use a deep learning network CNN for data modeling and prediction to obtain an active prediction model for respiratory changes. Step S4: Based on the active prediction model of respiratory changes, a respiratory index prediction model based on the gated circulation unit (GRU) is further established, and long-term prediction fitting results are obtained by using iterative prediction. The influence of respiratory data such as oxygen inlet pressure, displacement, and cabin pressure on oxygen outlet pressure is analyzed. By comparing normalized oxygen inlet pressure and normalized displacement data, oxygen outlet pressure and oxygen outlet pressure label prediction indicators are predicted to obtain accurate predicted respiratory rhythm in the respiratory index prediction model, which is used for respiratory signal monitoring and regulation of pilots during high-altitude flight.

2. The method according to claim 1, characterized in that, In step S2, when performing canonical correlation analysis, the breathing pattern is first studied using time-domain analysis. Then, wavelet transform and Fourier transform are performed on the breathing signal using frequency-domain analysis to extract the corresponding features and perform spectral analysis on the breathing signal to calculate the spectral parameters of the window signal. Then, the manually extracted features and deep learning features of each breathing signal are summarized and arranged according to their corresponding numbers to obtain a data matrix composed of several features, which is then fed into the breathing index prediction model for regression prediction to obtain the human breathing characteristic transformation curve.

3. The method according to claim 1 or 2, characterized in that, In step S2, during feature fusion, the segments extracted from the respiratory signal are numbered one by one, and noise removal and inversion correction methods are used to preprocess the signal.

4. The method according to claim 1 or 2, characterized in that, Step S2 involves feature fusion before classification. First, all types of features are merged and summarized. Then, the fused features are treated as a feature matrix and classified using a machine learning algorithm classification model to obtain the experimental results of the fusion model.

5. The method according to claim 1 or 2, characterized in that, The deep learning network used in step S3 is a deep convolutional neural network. In the deep learning feature part, the time-domain map of the respiratory signal STFT transformation is used as the model for deep learning feature extraction. The time-domain map of the respiratory signal STFT transformation is processed by the deep convolutional neural network and exported before the final Softmax classification operation to obtain 128-dimensional deep learning features of the respiratory signal. The last layer in the original network, i.e., the softmax layer, is removed, and the fully connected layer is directly used as the vector output. The vector obtained by the modified network is the deep learning feature of the signal.

6. The method according to claim 5, characterized in that, In step S4, the establishment of a respiratory index prediction model based on a gated recurrent unit (GRU) involves the setting of multiple hyperparameters in the training data of the GRU neural network, including: the number of neurons m, the time step T, and the batch size.

7. A system for analyzing and predicting the changing trends of breathing characteristic parameters in high-altitude flight based on deep learning, characterized in that, include: The feature extraction module is used to collect the pilot's breathing data during high-altitude flight and extract relevant time series waveform parameter features; Respiratory data is collected using pressure and displacement sensors. The pressure sensor is used to collect the airway ventilation pressure signal, and the differential pressure signal during the patient's breathing is collected through a differential pressure tube, which serves as the front-end signal for respiratory detection. The analysis and fusion module is used to perform canonical correlation analysis on the extracted parameter features, and then perform feature fusion in two modes, including manual feature extraction and STFT transform deep learning feature extraction, to obtain the human respiratory characteristic transformation curve through predictive analysis. The deep learning module is used to transform the feature vectors of the respiratory parameter sequence into the input and output of a deep learning network through a sliding window. Then, a deep learning network (CNN) is used for data modeling and prediction to obtain an active prediction model of respiratory changes. The predictive regulation module is used to further establish a respiratory index prediction model based on the active respiratory change prediction model and the gated respiration unit (GRU). It uses iterative prediction to obtain long-term prediction fitting results, analyzes the impact of respiratory data such as oxygen inlet pressure, displacement, and cabin pressure on oxygen outlet pressure, and predicts oxygen outlet pressure and oxygen outlet pressure label prediction indicators by comparing normalized oxygen inlet pressure and normalized displacement data. This results in an accurate predicted respiratory rhythm in the respiratory index prediction model, which is used for respiratory signal monitoring and regulation by pilots during high-altitude flight.

8. An electronic system comprising: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, characterized in that the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-6.

9. A computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

Images