Remote non-contact heart rate monitoring system used in severe environment of railway

Abstract

The invention relates to a remote non-contact heart rate monitoring system used in a severe railway environment, and the system comprises a microwave radar sensing module which is used for transmitting and receiving a microwave signal, measuring the heart micro-motion information of a worker in a non-contact manner, and transmitting the heart micro-motion information to a signal fusion module in the form of time sequence data; the inertial measurement unit is used for monitoring the body movement state of a worker to obtain body movement data and transmitting the body movement data to the signal fusion module; the signal fusion module is used for carrying out data fusion on the heart micro-motion information and the body motion data to generate a fused heart rate time sequence signal; the feature extraction module is used for receiving the fused heart rate time sequence signals, extracting heart rate feature data and removing pseudo signals to obtain pseudo-removed heart rate feature data; and the classification module is used for processing the de-falsification heart rate feature data by adopting a pre-trained BiLSTM network structure to obtain a predicted label corresponding to the de-falsification heart rate feature data.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical signal processing, and particularly to a remote non-contact heart rate monitoring system for harsh railway environments. Background Art

[0002] In the railway industry, staff members (such as train drivers and railway inspectors) are often in high-intensity and high-risk working environments. Long-term high-load work may lead to health problems such as fatigue and abnormal heart rates, which in turn affect work safety. Therefore, real-time monitoring of the heart rate status of staff members is of great significance for improving the safety and work efficiency of railway operations. Currently, common heart rate monitoring technologies mainly include the following categories: Using optical sensors to detect blood flow changes, which are widely used in wearable devices such as smart bracelets and smart watches. The signals are easily interfered with under conditions such as strong light, sweat, and poor skin contact, and wearing the device is required, which is not suitable for remote non-contact monitoring in special environments such as railways. By adhering electrodes to the skin to collect cardiac bioelectrical signals for high-precision heart rate monitoring. Direct skin contact is required, which is not suitable for remote monitoring, and long-term wearing of electrodes may cause discomfort. Using microwave radar technology (such as FMCW radar) to detect the minute vibrations of the human heart to achieve remote non-contact heart rate monitoring. Radar measurements are easily affected by body movement interference, such as vibrations in the railway environment and the walking of staff members, resulting in increased signal noise and affecting measurement accuracy.

[0003] The railway operating environment has the following characteristics, which pose challenges to traditional heart rate monitoring methods: During train operation or railway inspection, the environmental noise and vibrations are relatively large, affecting the stability of the radar micro-motion signals. There are significant body movements during the inspection or operation of staff members, resulting in a large number of pseudo-signals mixed in the heart rate monitoring signals. Railway operations require remote monitoring of the health status of staff members, and traditional contact heart rate monitoring methods are difficult to meet the requirements of remote and unperceived detection.

[0004] For remote heart rate monitoring in harsh railway environments, currently mainly microwave radar or inertial measurement units (IMUs) are used alone for monitoring, but there are the following defects: Using microwave radar alone may result in large heart rate measurement errors due to body movement pseudo-signals, while inertial measurement units (IMUs) can only reflect large-scale body movements and cannot effectively distinguish subtle cardiac micro-motion signals. Traditional methods are difficult to effectively fuse microwave radar signals and inertial measurement data, thus unable to achieve accurate heart rate detection. Traditional time-frequency analysis methods are difficult to effectively remove body movement pseudo-signals, resulting in unstable heart rate measurements, especially a decrease in accuracy in high-noise environments. Most existing heart rate detection algorithms are based on traditional signal processing methods such as FFT and STFT, and lack optimization strategies for non-linear feature extraction and classification using deep learning models. Summary of the Invention

[0005] In view of the above-mentioned disadvantages and deficiencies of the prior art, the present invention provides a remote non-contact heart rate monitoring system for harsh railway environments.

[0006] To achieve the above object, the main technical solutions adopted by the present invention include:

[0007] An embodiment of the present invention provides a remote non-contact heart rate monitoring system for harsh railway environments, including:

[0008] A microwave radar sensing module, configured to transmit and receive microwave signals, measure the cardiac micro-motion information of the staff in a non-contact manner, and transmit the cardiac micro-motion information to the signal fusion module in the form of time series data;

[0009] An inertial measurement unit, including an accelerometer and a gyroscope, configured to monitor the body motion state of the staff to obtain body motion data, and transmit the body motion data to the signal fusion module;

[0010] A signal fusion module, configured to receive the data of the microwave radar sensing module and the inertial measurement unit, and perform data fusion on the cardiac micro-motion information and the body motion data to generate a fused heart rate time series signal;

[0011] A feature extraction module, including a time-frequency analysis unit and a variational autoencoder, configured to receive the fused heart rate time series signal, extract heart rate feature data and remove pseudo signals to obtain de-pseudo heart rate feature data;

[0012] A classification module, which processes the de-pseudo heart rate feature data using a pre-trained BiLSTM network structure to obtain a predicted label corresponding to the de-pseudo heart rate feature data.

[0013] Preferably, the microwave radar sensing module adopts FMCW radar technology to measure the cardiac micro-motion information of the staff in a non-contact manner;

[0014] The cardiac micro-motion information includes: heart rate, amplitude of heartbeat;

[0015] The body motion data includes: acceleration data related to the body motion of the staff collected by the accelerometer, and angular velocity data related to the body motion of the staff collected by the gyroscope.

[0016] Preferably, the signal fusion module receives the data of the microwave radar sensing module and the inertial measurement unit, and performs data fusion on the cardiac micro-motion information and the body motion data to generate a fused heart rate time series signal, specifically including:

[0017] Obtain the cardiac micro-motion information from the microwave radar sensing module and the body motion data from the inertial measurement unit

[0018] For each time point t, calculate the weighted coefficients w radar (t, i) and w IMU (t, i) from the microwave radar signal and the body movement signal based on the locally weighted regression method, where the weighted coefficients are adaptively adjusted by the Gaussian kernel function and the signal noise variance;

[0019] Use the calculated weighted coefficients to perform weighted summation on the microwave radar signal and the body movement signal to obtain the fused heart rate time series signal

[0020]

[0021] N(t) is a set containing the current time point t and its neighborhood points.

[0022] Preferably, among them, the weighted coefficient w radar (t, i) is calculated by formula (1);

[0023] The formula (1) is:

[0024]

[0025] The weighted coefficient w IMU (t, i) is calculated by formula (2);

[0026] The formula (2) is:

[0027]

[0028] Among them, is the Gaussian kernel function, which is used to adjust the weighted coefficient according to the distance between time points, and are the noise variances of the microwave radar signal and the body movement signal.

[0029] Preferably, the neighborhood point set N(t) is several time points before and after the time point t, and the size of the neighborhood is controlled by the bandwidth parameter h.

[0030] Preferably, the feature extraction module includes a time-frequency analysis unit and a variational autoencoder, which are used to receive the fused heart rate time series signal, extract heart rate feature data and remove pseudo signals, specifically including:

[0031] Receive the fused heart rate time series signal from the signal fusion module

[0032] Perform time-frequency transformation on the fused heart rate time series signal based on the time-frequency analysis unit to obtain the spectrogram S(f, t);

[0033] Adaptive learning of spectrograms is performed using a variational autoencoder to obtain a low-dimensional heart rate feature representation z t ;

[0034] Based on the time-frequency analysis results of the spectrogram, formula (3) is applied to obtain de-pseudo heart rate feature data;

[0035] The formula (3) is:

[0036]

[0037] VAε(z t ) is the output of the variational autoencoder; γ is the regularization coefficient;

[0038] is the de-pseudo heart rate feature data.

[0039] Preferably,

[0040] Among them, the structure of the variational autoencoder includes an encoder, a decoder, and a latent space, and is trained by maximizing the variational lower bound to generate a low-dimensional heart rate feature representation z t ;

[0041] Among them, the spectrogram S(f, t) obtained by time-frequency transformation is acquired based on the short-time Fourier transform or wavelet transform method to represent the distribution characteristics of the heart rate signal in the time-frequency domain.

[0042] Preferably, the system further includes:

[0043] A data storage and transmission module for recording historical heart rate data and transmitting the data to a remote monitoring center via wireless communication;

[0044] An early warning module for detecting abnormal conditions based on the heart rate change trend and sending an alarm to the staff or the monitoring center;

[0045] The data storage and transmission module adopts an edge computing architecture to perform preliminary signal processing locally and only uploads key feature data to reduce the network burden;

[0046] The early warning module constructs a heart rate prediction model based on a long short-term memory network and issues an early warning in advance by detecting abnormal change trends;

[0047] The system synchronizes data with the intelligent terminal devices of railway operating personnel and provides real-time health status assessment and remote medical advice.

[0048] Preferably,

[0049] The BiLSTM network structure sequentially includes: an input layer, a BiLSTM layer, a fully connected layer, and an output layer;

[0050] The input layer is used to receive the de - pseudonymized heart rate feature data;

[0051] The BiLSTM layer is used to simultaneously capture the complete temporal features of the de - pseudonymized heart rate feature data through forward and backward propagation;

[0052] The fully - connected layer is used to integrate the complete temporal features output by the BiLSTM layer to obtain the final feature vector;

[0053] The output layer is used to determine the predicted label corresponding to the de - pseudonymized heart rate feature data according to the final feature vector.

[0054] Preferably, the BiLSTM network structure is pre - trained using a training data set to obtain a trained BiLSTM network structure;

[0055] The training data set includes multiple samples;

[0056] Each sample includes a de - pseudonymized heart rate feature data in a historical time period and the true label corresponding to the de - pseudonymized heart rate feature data.

[0057] The beneficial effects of the present invention are:

[0058] A remote non - contact heart rate monitoring system for railway harsh environments according to the present invention, by combining a microwave radar sensing module and an inertial measurement unit, and fusing cardiac micro - motion information and body motion data through a signal fusion module, compared with the prior art, it can effectively reduce the interference of vibration and noise in the railway operation environment on heart rate monitoring, improve the stability and accuracy of heart rate detection, achieving the technical effect of realizing high - precision heart rate monitoring under remote and non - contact conditions and adapting to railway harsh environments.

[0059] In addition, by using time - frequency analysis and variational auto - encoder (VAE) for feature extraction and combining with a pre - trained BiLSTM network for classification, compared with the prior art, it can effectively remove body motion pseudo - signals, improve the purity of heart rate signals, enhance the ability to identify abnormal heart rates, achieving the improvement of the robustness and real - time performance of heart rate detection in complex environments, and providing a more reliable health monitoring means for railway workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic structural diagram of a remote non - contact heart rate monitoring system for railway harsh environments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] To better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the accompanying drawings through specific embodiments.

[0062] To better understand the above technical solution, the exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and the scope of the present invention can be fully communicated to those skilled in the art.

[0063] Embodiment 1

[0064] Refer to Figure 1 , this embodiment provides a remote non-contact heart rate monitoring system for harsh railway environments, including:

[0065] A microwave radar sensing module, which is used to transmit and receive microwave signals, measure the cardiac micro-motion information of the staff in a non-contact manner, and transmit the cardiac micro-motion information to the signal fusion module in the form of time-series data;

[0066] The microwave radar sensing module adopts FMCW radar technology to measure the cardiac micro-motion information of the staff in a non-contact manner;

[0067] In this embodiment, by adopting FMCW radar technology, the cardiac micro-motion information can be measured non-contact, which is suitable for remote monitoring in harsh railway environments. Through time-series data transmission, the real-time nature of the data is ensured and the signal transmission delay is reduced.

[0068] The cardiac micro-motion information includes: heart rate, amplitude of heartbeat;

[0069] The body motion data includes: acceleration data related to the body motion of the staff collected by the accelerometer, and angular velocity data related to the body motion of the staff collected by the gyroscope.

[0070] An inertial measurement unit, including an accelerometer and a gyroscope, is used to monitor the body motion state of the staff to obtain body motion data, and transmit the body motion data to the signal fusion module;

[0071] Combining the data of the accelerometer and the gyroscope can separate the influence of cardiac micro-motion and body motion, improving the accuracy of heart rate measurement. It can not only obtain heart rate information, but also evaluate the body motion state of the operator, which is helpful for comprehensive health monitoring.

[0072] In this embodiment, the inertial measurement unit can detect the body motion state of the operator, avoid mismeasurement caused by external motion, and improve the data quality. Combining with the microwave radar signal enables the system to still maintain a high measurement accuracy in a high-noise environment (such as a railway construction site).

[0073] A signal fusion module, configured to receive data from a microwave radar sensing module and an inertial measurement unit, and perform data fusion on the cardiac micro-motion information and the body motion data to generate a fused heart rate time series signal;

[0074] In this embodiment, by fusing microwave radar signals and body motion signals, the anti-interference ability of heart rate measurement is improved, and the reliability of measurement is enhanced.

[0075] Specifically, the signal fusion module receives data from the microwave radar sensing module and the inertial measurement unit, and performs data fusion on the cardiac micro-motion information and the body motion data to generate a fused heart rate time series signal, which specifically includes:

[0076] Obtain cardiac micro-motion information from the microwave radar sensing module and body motion data from the inertial measurement unit

[0077] For each time point t, calculate the weighting coefficients w radar (t, i) and w IMU (t, i) of the microwave radar signal and the body motion signal based on the locally weighted regression method, where the weighting coefficients are adaptively adjusted through a Gaussian kernel function and signal noise variance;

[0078] Perform weighted summation on the microwave radar signal and the body motion signal using the calculated weighting coefficients to obtain a fused heart rate time series signal

[0079]

[0080] N(t) is a set containing the current time point t and its neighborhood points.

[0081] A feature extraction module, including a time-frequency analysis unit and a variational autoencoder, configured to receive the fused heart rate time series signal, extract heart rate feature data and remove false signals to obtain de-falsified heart rate feature data;

[0082] A classification module processes the de-falsified heart rate feature data using a pre-trained BiLSTM network structure to obtain a predicted label corresponding to the de-falsified heart rate feature data.

[0083] The BiLSTM network structure sequentially includes: an input layer, a BiLSTM layer, a fully connected layer, and an output layer;

[0084] Adopt a bidirectional long short-term memory (BiLSTM) network to improve the time series modeling ability of heart rate data and reduce the misclassification rate. The fully connected layer integrates time series features to improve the effectiveness of feature expression, enabling the system to more accurately identify the health status of the operator.

[0085] The input layer is used to receive the de - pseudo heart rate feature data;

[0086] The BiLSTM layer is used to simultaneously capture the complete temporal features of the de - pseudo heart rate feature data through forward and backward propagation;

[0087] The fully - connected layer is used to integrate the complete temporal features output by the BiLSTM layer to obtain the final feature vector;

[0088] The output layer is used to determine the predicted label corresponding to the de - pseudo heart rate feature data according to the final feature vector.

[0089] In the practical application of this embodiment, among them, the weighting coefficient w radar (t, i) is calculated by formula (1);

[0090] The formula (1) is:

[0091]

[0092] The weighting coefficient w IMU (t, i) is calculated by formula (2);

[0093] The formula (2) is:

[0094]

[0095] Among them, is the Gaussian kernel function, which is used to adjust the weighting coefficient according to the distance between time points, and are the noise variances of the microwave radar signal and the body movement signal.

[0096] Among them, the neighborhood point set N(t) is several time points before and after the time point t, and the size of the neighborhood is controlled by the bandwidth parameter h.

[0097] The feature extraction module includes a time - frequency analysis unit and a variational auto - encoder, and is used to receive the fused heart rate time - series signal, extract the heart rate feature data and remove the pseudo - signals, specifically including:

[0098] Receiving the fused heart rate time - series signal from the signal fusion module

[0099] Performing time - frequency transformation on the fused heart rate time - series signal based on the time - frequency analysis unit to obtain the spectrogram S(f, t);

[0100] Using the variational auto - encoder to perform adaptive learning on the spectrogram to obtain the low - dimensional heart rate feature representation z t ;

[0101] Based on the time-frequency analysis results of the spectrogram, the formula (3) is applied to obtain the de-pseudo heart rate feature data;

[0102] The formula (3) is as follows:

[0103]

[0104] VAε(z t ) is the output of the variational autoencoder; γ is the regularization coefficient;

[0105] is the de-pseudo heart rate feature data.

[0106] In this embodiment, the time-frequency analysis unit can transform the heart rate signal into time-frequency domain features through short-time Fourier transform or wavelet transform to improve the separability of the signal. The variational autoencoder (VAE) generates a low-dimensional heart rate feature representation through adaptive learning, effectively removing pseudo signals and improving the reliability of heart rate data. The regularization method (based on formula (3)) ensures the stability of the de-pseudo heart rate data and avoids the influence of outliers on the classification results.

[0107] Among them, the structure of the variational autoencoder includes an encoder, a decoder, and a latent space, and is trained by maximizing the variational lower bound to generate a low-dimensional heart rate feature representation z t ;

[0108] Among them, the spectrogram S(f,t) obtained through time-frequency transformation is acquired based on the short-time Fourier transform or wavelet transform method to represent the distribution characteristics of the heart rate signal in the time-frequency domain.

[0109] In this embodiment, the system further includes:

[0110] A data storage and transmission module for recording historical heart rate data and transmitting the data to a remote monitoring center through wireless communication;

[0111] An early warning module for detecting abnormal conditions based on the heart rate change trend and sending an alarm to the staff or the monitoring center;

[0112] The data storage and transmission module adopts an edge computing architecture to perform preliminary signal processing locally and only uploads key feature data to reduce the network burden;

[0113] The early warning module constructs a heart rate prediction model based on a long short-term memory network and issues an early warning in advance by detecting abnormal change trends;

[0114] The system synchronizes data with the intelligent terminal device of railway operating personnel and provides real-time health status assessment and remote medical advice.

[0115] Among them, the BiLSTM network structure is pre-trained using a training dataset to obtain a trained BiLSTM network structure;

[0116] The training dataset includes multiple samples;

[0117] Each sample includes a de-noised heart rate feature data in a historical time period and a true label corresponding to the de-noised heart rate feature data.

[0118] In this embodiment, BiLSTM (Bidirectional Long Short-Term Memory network) is used to process the de-noised heart rate feature data, which has the following advantages: bidirectional propagation (forward + backward), which can capture complete temporal information and improve the temporal dependence of heart rate features. The fully connected layer integrates the output of the BiLSTM layer to further optimize the feature representation and improve the prediction accuracy. The output layer provides the final prediction result to achieve intelligent classification of the heart rate feature data.

[0119] Embodiment 2

[0120] This embodiment provides a remote non-contact heart rate monitoring system for harsh railway environments. The system consists of multiple modules, including a microwave radar sensing module, an inertial measurement unit, a signal fusion module, a feature extraction module, a classification module, a data storage and transmission module, and an early warning module. Each module works in coordination to achieve high-precision, remote, and non-contact heart rate monitoring.

[0121] First, the microwave radar sensing module uses FMCW radar technology to transmit and receive microwave signals in a non-contact manner, thereby measuring the cardiac micro-motion information of the staff and transmitting the acquired temporal data to the signal fusion module. This microwave radar sensing module can effectively sense heart rate and heartbeat amplitude information and adapt to complex working conditions in the railway environment.

[0122] At the same time, the inertial measurement unit consists of an accelerometer and a gyroscope, which are used to monitor the body movement state of the staff in real time and generate corresponding acceleration data and angular velocity data. These body movement data are synchronously transmitted to the signal fusion module to provide the ability to correct body movement pseudo-signals.

[0123] The core function of the signal fusion module lies in the fusion processing of data. Specifically, this module receives the cardiac micro-motion information from the microwave radar sensing module and the body movement data from the inertial measurement unit, and calculates the weighting coefficient based on the locally weighted regression method. The calculation formula of the weighting coefficient is as follows:

[0124] w(t) represents the weighting coefficient at time t;

[0125] K h (t - t i) is the Gaussian kernel function, representing the weighted influence degree at different time points, where t i represents the i-th sampling time point, and t is the current time;;

[0126] represents the noise variance of the microwave radar signal;

[0127] represents the noise variance of the body movement signal provided by the inertial measurement unit.

[0128] Then, the signal fusion module uses the calculated weighting coefficients to perform weighted summation on the microwave radar signal and the body movement signal to generate the fused heart rate time series signal.

[0129] The feature extraction module includes a time-frequency analysis unit and a variational autoencoder. First, the time-frequency analysis unit uses the short-time Fourier transform or wavelet transform to perform time-frequency transformation on the fused heart rate time series signal to obtain a spectrogram. Subsequently, the variational autoencoder performs adaptive learning on the spectrogram, extracts the low-dimensional heart rate feature representation, and calculates the pseudo-heart rate feature data based on the regularization formula:

[0130] The regularization formula is:

[0131]

[0132] where is the pseudo-heart rate feature data;

[0133] VAE(HR raw ) represents the feature data output by the variational autoencoder;

[0134] L is the regularization coefficient, controlling the strength of the pseudo-signal;

[0135] R is the noise suppression term, used to reduce the influence of environmental interference on heart rate detection.

[0136] The classification module is processed by a pre-trained BiLSTM network, which includes an input layer, a BiLSTM layer, a fully connected layer, and an output layer. The BiLSTM layer uses bidirectional long short-term memory units to capture both the forward and backward temporal features of the pseudo-heart rate feature data simultaneously, improving the accuracy of heart rate classification. The fully connected layer further integrates the temporal features and generates a prediction label at the output layer.

[0137] In addition, the system is also equipped with a data storage and transmission module and an early warning module. The data storage and transmission module adopts an edge computing architecture, performs preliminary signal processing locally, and uploads key feature data to the remote monitoring center via wireless communication to reduce the network burden. The early warning module uses a long short-term memory (LSTM) network to predict the heart rate trend and sends an alarm to the staff or the monitoring center based on abnormal changes.

[0138] The remote non-contact heart rate monitoring system in this embodiment can synchronize data with the intelligent terminal devices of railway workers, provide real-time health status assessment and remote medical advice, thereby enhancing the operational safety in the railway environment and improving the health monitoring level of the staff.

[0139] Embodiment III

[0140] Railway workers work in extreme environments (such as high temperature, low temperature, strong noise, high altitude, etc.) for a long time and are prone to be affected by fatigue, hypoxia or cardiovascular diseases. Therefore, developing a remote non-contact heart rate monitoring system for harsh railway environments to achieve real-time monitoring, health assessment and safety warning of railway workers' heart rates has important practical value.

[0141] A remote non-contact heart rate monitoring system for harsh railway environments in this embodiment mainly consists of the following modules:

[0142] The signal acquisition module is used to collect the ballistocardiogram signals of railway workers by using a millimeter-wave radar or an infrared sensor. In a non-contact manner, it improves comfort and reduces the discomfort caused by wearing devices.

[0143] The signal processing module is used to standardize the collected original signals and perform dimensional expansion to meet the model input requirements. It uses a BiLSTM-GAN (Bidirectional Long Short-Term Memory Generative Adversarial Network) model to generate high-quality arrhythmia signals to supplement the training data and improve data balance. It uses a BiLSTM-FCN (Bidirectional Long Short-Term Memory and Fully Convolutional Network) model to extract signal features, including: LSTM channels: capturing dynamic changes in time series. FCN channels: extracting global features through global average pooling and finally classifying and outputting six different ballistocardiogram signal categories.

[0144] The noise filtering module is used to use a third-order Butterworth band-pass filter of 1 - 10 Hz to remove high-frequency noise and low-frequency interference to ensure the accuracy of the signals.

[0145] The data analysis module is used to analyze the processed signals, calculate the heart rate value, generate a trend curve, and provide long-term health data support.

[0146] The display and alarm module is used to display the monitoring results in real time through an LCD screen or a mobile APP. When the heart rate is abnormal, the system automatically triggers an audible and visual alarm and pushes the alarm information to the remote monitoring platform.

[0147] This system realizes remote heart rate monitoring by the following steps:

[0148] Signal acquisition: The heart impact signals of railway workers are acquired through millimeter-wave radar or infrared sensors.

[0149] Signal preprocessing, including: normalizing the signal data to analyze it on a unified scale; dimension expansion to improve the neural network's learning ability for features.

[0150] Signal enhancement, including using the BiLSTM-GAN model to generate arrhythmia signals to improve data balance:

[0151] Generator (G): Based on the BiLSTM architecture, it generates high-quality heart rhythm signals.

[0152] Discriminator (D): Adopting CNN (Convolutional Neural Network) to discriminate the differences between the generated signals and the real signals.

[0153] Feature extraction and classification, including: using the BiLSTM-FCN model for classification: BiLSTM layer: extracting dynamic features in the time series; FCN layer: extracting global features and classifying. Six types of heart impact signal categories are output through the Softmax function: normal signal, large body movement, small body movement, deep breath, pulse body movement, invalid segment.

[0154] Noise filtering, including using a third-order Butterworth band-pass filter of 1 - 10 Hz to remove environmental noise and improve signal purity.

[0155] Result output, including calculating the heart rate value and generating a visualized heart rate trend graph. Providing a remote monitoring data interface to achieve long-term storage and analysis of health data.

[0156] Experimental results show that the heart rate monitoring effect of this system is remarkable in the harsh railway environment, and the specific data is as follows:

[0157] BiLSTM-GAN model:

[0158] Both MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) are increased by 50%.

[0159] R 2 The index is increased by 21.25%.

[0160] BiLSTM-FCN model: The classification accuracy of six types of heart impact signals reaches over 95%.

[0161] Non-contact monitoring improves the comfort of workers and reduces the impact brought by wearing equipment.

[0162] This system is specially designed for the harsh railway environment and has the following characteristics: waterproof, dustproof, resistant to high and low temperatures, shock-resistant, and resistant to environmental interference.

[0163] In this embodiment, BiLSTM-GAN (Bidirectional Long Short-Term Memory-Generative Adversarial Network) is a method that combines BiLSTM (Bidirectional Long Short-Term Memory Network) and GAN (Generative Adversarial Network). It is mainly used to enhance and generate high-quality time series data and is used for non-contact heart rate monitoring in harsh railway environments in this invention. BiLSTM-GAN consists of a generator (Generator, G) and a discriminator (Discriminator, D). The two are continuously optimized through adversarial training to improve the authenticity of the generated data.

[0164] Generator (G): Adopts the BiLSTM structure and utilizes the ability of the bidirectional long short-term memory network to capture temporal information to generate realistic heart rhythm signal data.

[0165] The input is random noise. After being processed by BiLSTM, it outputs data close to real ballistocardiogram signals.

[0166] Discriminator (D): Adopts the structure of a convolutional neural network (CNN) to classify the input signal data and determine whether it is real data or generated data. The training objective is to improve the ability to distinguish between real signals and synthetic signals.

[0167] Adversarial training mechanism: The generator continuously optimizes itself, making the generated ballistocardiogram signals more and more realistic, making it difficult for the discriminator to distinguish. The discriminator continuously optimizes itself to improve its discrimination ability, thus prompting the generator to generate higher-quality data.

[0168] In the description of this invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this invention, "a plurality" means two or more unless otherwise specifically defined.

[0169] In this invention, unless otherwise clearly specified and limited, terms such as "installed", "connected", "connected to", "fixed" and other terms should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in this invention can be understood according to specific circumstances.

[0170] In the present invention, unless otherwise clearly specified and defined, a first feature being "on" or "under" a second feature may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact via an intermediate medium. Moreover, a first feature being "above", "over" and "on top of" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply that the horizontal height of the first feature is higher than that of the second feature. A first feature being "under", "below" and "beneath" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply that the horizontal height of the first feature is lower than that of the second feature.

[0171] In the description of this specification, the descriptions of terms such as "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0172] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A remote non-contact heart rate monitoring system for harsh railway environments, characterized in that, Including: A microwave radar sensing module, which is used to transmit and receive microwave signals, measure the micro-movement information of the staff's heart in a non-contact manner, and transmit the heart micro-movement information to the signal fusion module in the form of time-series data; An inertial measurement unit, including an accelerometer and a gyroscope, which is used to monitor the body movement state of the staff to obtain body movement data and transmit the body movement data to the signal fusion module; A signal fusion module, which is used to receive the data of the microwave radar sensing module and the inertial measurement unit, and perform data fusion on the heart micro-movement information and the body movement data to generate a fused heart rate time-series signal; A feature extraction module, including a time-frequency analysis unit and a variational autoencoder, which is used to receive the fused heart rate time-series signal, extract heart rate feature data and remove pseudo signals to obtain pseudo-signal-free heart rate feature data; A classification module, which processes the pseudo-signal-free heart rate feature data using a pre-trained BiLSTM network structure to obtain a predicted label corresponding to the pseudo-signal-free heart rate feature data.

2. The remote non-contact heart rate monitoring system according to claim 1, wherein The microwave radar sensing module adopts FMCW radar technology to measure the micro-movement information of the staff's heart in a non-contact manner; The heart micro-movement information includes: heart rate, amplitude of heartbeat; The body movement data includes: acceleration data related to the body movement of the staff collected by the accelerometer, and angular velocity data related to the body movement of the staff collected by the gyroscope.

3. The remote non-contact heart rate monitoring system according to claim 1, characterized in that The signal fusion module receives the data of the microwave radar sensing module and the inertial measurement unit, and performs data fusion on the heart micro-movement information and the body movement data to generate a fused heart rate time-series signal, specifically including: Obtain cardiac micro-motion information from the microwave radar sensing module and body motion data from the inertial measurement unit For each time point t, calculate the weighted coefficients w radar (t, i) and w IMU (t, i) from the microwave radar signal and the body movement signal based on the locally weighted regression method, where the weighted coefficients are adaptively adjusted by the Gaussian kernel function and the signal noise variance; radar (t, i) and w IMU (t, i), where the weighted coefficients are adaptively adjusted by the Gaussian kernel function and the signal noise variance; The calculated weighting coefficients are used to perform weighted summation on the microwave radar signal and the body movement signal to obtain the fused heart rate time series signal N(t) is a set containing the current time point t and its neighborhood points.

4. The remote non-contact heart rate monitoring system according to claim 3, characterized in that, Wherein, The weighted coefficient w radar (t, i) is calculated by formula (1); The formula (1) is: The weighted coefficient w IMU (t, i) is calculated by formula (2); The formula (2) is: Among them, is the Gaussian kernel function, which is used to adjust the weighting coefficient according to the distance between time points, and are the noise variances of the microwave radar signal and the body movement signal.

5. The remote non-contact heart rate monitoring system according to claim 4, wherein The neighborhood point set N(t) is several time points before and after the time point t, and the size of the neighborhood is controlled by the bandwidth parameter h.

6. The remote non-contact heart rate monitoring system according to claim 1, wherein The feature extraction module, including a time-frequency analysis unit and a variational autoencoder, is used to receive the fused heart rate time-series signal, extract heart rate feature data and remove pseudo signals, specifically including: Receive the fused heart rate time series signal from the signal fusion module Performing time-frequency transformation on the fused heart rate time-series signal based on the time-frequency analysis unit to obtain a spectrogram S(f, t); Use a variational autoencoder to adaptively learn the spectrogram and obtain a low-dimensional heart rate feature representation z t ; Based on the time-frequency analysis result of the spectrogram, applying formula (3) to obtain pseudo-signal-free heart rate feature data; The formula (3) is: VAε(z t ) is the output of the variational autoencoder; γ is the regularization coefficient; To eliminate false heart rate characteristic data.

7. The remote non-contact heart rate monitoring system according to claim 6, wherein Among them, The structure of the variational autoencoder includes an encoder, a decoder, and a latent space, and is trained by maximizing the variational lower bound to generate a low-dimensional heart rate feature representation z t ; The spectrogram S(f, t) obtained through time-frequency transformation is obtained based on the short-time Fourier transform or wavelet transform method to represent the distribution characteristics of the heart rate signal in the time-frequency domain.

8. The remote non-contact heart rate monitoring system according to claim 1, characterized in that, The system further includes: A data storage and transmission module, which is used to record historical heart rate data and transmit the data to a remote monitoring center through wireless communication; An early warning module, which is used to detect abnormal situations based on the heart rate change trend and send an alarm to the staff or the monitoring center; The data storage and transmission module adopts an edge computing architecture, performs preliminary signal processing locally, and only uploads key feature data to reduce the network burden; The warning module constructs a heart rate prediction model based on a long short-term memory network and issues a warning in advance by detecting abnormal change trends; The system synchronizes data with the intelligent terminal device of railway operators and provides real-time health status assessment and remote medical advice.

9. The remote non-contact heart rate monitoring system according to claim 1, characterized in that, It is characterized in that The BiLSTM network structure successively includes: an input layer, a BiLSTM layer, a fully connected layer, and an output layer; The input layer is used to receive the de-falsified heart rate feature data; The BiLSTM layer is used to simultaneously capture the complete temporal features of the de-falsified heart rate feature data through forward and backward propagation; The fully connected layer is used to integrate the complete temporal features output by the BiLSTM layer to obtain the final feature vector; The output layer is used to determine the predicted label corresponding to the de-falsified heart rate feature data according to the final feature vector.

10. The remote non-contact heart rate monitoring system according to claim 9, wherein It is characterized in that Among them, the BiLSTM network structure is pre-trained using a training data set to obtain a trained BiLSTM network structure; The training data set includes multiple samples; Each of the samples includes a de-falsified heart rate feature data in a historical time period and the true label corresponding to the de-falsified heart rate feature data.

Images