Non-contact blood pressure detection method and system based on deep learning

Abstract

The invention discloses a non-contact blood pressure detection method and system based on deep learning, and relates to the technical field of computer vision, and the method comprises the following steps: 1, shooting a face video, and collecting a data set; 2, extracting a frame-by-frame picture through the face video, and calculating an average pixel value to obtain an IPPG signal; 3, an IPPG-Mama network model is constructed, the extracted IPPG signal data set is utilized to train the IPPG-Mama network model, and a blood pressure detection model is obtained; 4, collecting a to-be-detected face video file, and extracting an IPPG signal; and 5, inputting the IPPG signal to be detected into the blood pressure detection model to obtain a blood pressure detection result. The network model used in the invention combines HRV feature extraction and improved CNN-Mamba Block, through multi-feature fusion and novel network structure design, the pulse wave features of the IPPG signals can be better extracted, and the accuracy of blood pressure prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and more specifically, to a non-contact blood pressure detection method and system based on deep learning. Background Art

[0002] Currently, the measurement of blood pressure mainly relies on traditional cuff-type or wrist-type sphygmomanometers, which require manual operation or specific hardware devices. Although these methods are mature in clinical applications and have a certain measurement accuracy, they have obvious limitations in daily use. They are not only cumbersome to operate, but also prone to causing discomfort to users in high-frequency or continuous measurements, especially difficult to achieve efficient and convenient monitoring at night or in working scenarios. In addition, for some special populations, such as children, the elderly, or those with sensitive skin, this contact-type measurement method is not friendly enough and difficult to meet the daily health needs of a wide range of people.

[0003] With the development of biosignal acquisition and intelligent sensing technologies, photoplethysmography (PPG) has become a widely used non-invasive physiological signal acquisition method, which can infer vital signs such as an individual's heart rate, blood oxygen, and pulse rate by detecting minute changes in subcutaneous blood volume. However, traditional PPG signal acquisition relies on sensor devices that directly contact the skin, such as finger clips and bracelets, which to a certain extent limits the flexibility of its application scenarios. Especially in the context of the increasing prominence of non-contact requirements such as remote monitoring and mobile healthcare, the contact problem of existing PPG devices has gradually emerged.

[0004] To overcome the inconvenience of contact sensors in practical applications, researchers have proposed image-based photoplethysmography (IPPG) technology, that is, using ordinary camera devices to collect facial videos, and analyzing the pulse signals reflected by the minute changes in skin color with blood flow through image processing and signal extraction technologies to achieve non-contact monitoring of vital signs. Compared with traditional PPG, IPPG has significant comfort and scalability. Because its acquisition process does not require physical contact, it is more suitable for special groups such as children and the elderly, and is also convenient for realizing intelligent applications such as telemedicine and home health management. Currently, IPPG technology has achieved high accuracy in non-contact heart rate and respiratory rate detection and has been verified in multiple experimental environments.

[0005] However, compared with vital signs such as heart rate and respiratory rate, blood pressure estimation is more challenging. Blood pressure is not only affected by a variety of physiological parameters such as pulse wave morphology, wave velocity, and amplitude, but also involves complex hemodynamic mechanisms, making the mapping relationship between it and the IPPG signal more non-linear and unstable. Traditional shallow learning models and feature engineering methods are difficult to effectively establish a high-dimensional mapping relationship between blood pressure and IPPG. Therefore, how to improve the accuracy of blood pressure estimation by means of more powerful modeling capabilities has become a key issue in current research.

[0006] In recent years, with the rapid development of deep learning technology, researchers have begun to attempt to use models such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and Transformers to model IPPG signals in order to achieve more accurate blood pressure estimation. Although such models have certain advantages in feature extraction and sequence modeling, there are still many limitations in dealing with long time series, low latency, weak signal interference, etc. Especially in complex environments, such as under different lighting conditions, skin colors, facial expressions, or pose changes, the problems of insufficient model robustness and poor generalization ability are still widespread, limiting their popularization in practical applications.

[0007] Therefore, how to improve the accuracy and applicability of blood pressure estimation using IPPG signals is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides a non-contact blood pressure detection method and system based on deep learning, which solves the problems of inconvenient contact, poor environmental adaptability, and insufficient accuracy existing in traditional blood pressure detection methods.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A non-contact blood pressure detection method based on deep learning, comprising the following steps:

[0011] Step 1: Collect face video samples, including face videos and blood pressure data;

[0012] Step 2: Extract frame-by-frame images from the face video, and calculate the pixel mean value in the images to generate IPPG signals;

[0013] Step 3: Construct an IPPG-Mamba network model, and train the IPPG-Mamba network model according to the IPPG signals and blood pressure data to obtain a blood pressure detection model; the IPPG-Mamba network model includes an HRV feature extraction unit, a CNN-MambaBlock, a regression layer, and a feed-forward neural network; the HRV feature extraction unit extracts HRV feature sequences; the CNN-Mamba Block is an improved network module that combines a CNN network and a wavelet transform convolution Mamba Block to extract fused features; the regression layer performs feature integration on the fused features to obtain deep features; the feed-forward neural network fuses the HRV feature sequences and the deep features to predict the blood pressure detection result;

[0014] Step 4: Collect the face video to be detected, extract frame-by-frame images, and calculate the pixel mean value in the images to generate the IPPG signal to be detected;

[0015] Step 5: Input the IPPG signal to be detected into the blood pressure detection model for detection to obtain the blood pressure detection result.

[0016] Preferably, the feedforward neural network includes a linearly connected Linear 64, activation function ELU, Linear 32, activation function ELU, and Linear 1 in sequence.

[0017] Preferably, the CNN-Mamba Block includes a Root Mean Square Normalization (RMS Norm) layer, two groups of LinearLayer, a CNN network module, activation function SILU, a Selective State Space Model (SSM), and two groups of residual modules; the CNN network module includes two groups of transformation modules and a Dropout layer, and each group of transformation modules includes a One-Dimensional Wavelet Transform Convolution (WTConv1d), Batch Normalization layer, and One-Dimensional Max Pooling (MaxPool1d) layer; the first group of residual modules includes a linear layer and an activation function.

[0018] Preferably, the regression layer includes a Root Mean Square Normalization layer, a Dropout layer, a linear layer, and a Global Average Pooling layer.

[0019] Preferably, the one-dimensional wavelet transform convolutional layer WTConv1d includes a wavelet transform module, a one-dimensional convolutional module, and an inverse wavelet transform module connected in sequence.

[0020] Preferably, the process of the blood pressure detection model for detection is as follows:

[0021] Step 31: The HRV feature extraction unit extracts the RR interval from the IPPG signal through peak detection and calculates the HRV feature sequence based on the RR interval;

[0022] Step 32: The Root Mean Square Normalization layer normalizes the IPPG signal to obtain a normalized input signal;

[0023] Step 33: The first group of linear layers performs a linear transformation on the normalized input signal to obtain linear features; the linear layer of the first group of residual modules performs a linear transformation on the normalized input signal to obtain linear features, and then introduces non-linearity to the linear features through the activation function to obtain the first residual features;

[0024] Step 34: The linear features output by the first group of linear layers sequentially undergo feature extraction by two groups of transformation modules and partial feature discarding by the Dropout layer to obtain multi-scale convolutional features; the feature extraction process of each group of transformation modules is as follows:

[0025] In the one-dimensional wavelet transform convolutional layer, the wavelet transform module performs wavelet transform on linear features, decomposes the linear features into different frequency sub-bands to obtain multi-scale features; the one-dimensional convolutional module performs one-dimensional convolutional operations on the multi-scale features to extract local features of each frequency sub-band; the inverse wavelet transform module reconstructs the original signal space according to the local features of different frequency sub-bands, realizes the extraction of multi-scale deep features of linear features, and obtains multi-scale depth features;

[0026] The batch normalization layer normalizes the multi-scale depth features to obtain multi-scale high-dimensional normalized features, reduces internal covariate shift, and improves the training stability of the model;

[0027] The one-dimensional max pooling layer reduces the dimension of the multi-scale high-dimensional normalized features with a set pooling window and stride, retains key features while reducing the amount of data, reduces the computational complexity and overfitting risk of the model, and obtains multi-scale convolutional features;

[0028] Step 35: Introduce non-linearity to the multi-scale convolutional features through an activation function to obtain non-linear features;

[0029] Step 36: Select a state space model to selectively process information recursively according to the non-linear features, so as to focus on relevant data and discard unimportant information, and obtain selected features;

[0030] Step 37: Perform a multiplication operation on the selected features and the first residual features, and the multiplication result;

[0031] Step 38: The second group of linear layers performs a linear transformation on the multiplication result to obtain multiplication linear features, and performs an addition operation in combination with the IPPG signal transmitted by the second group of residual modules to obtain fused features;

[0032] Step 39: The fused features are sequentially passed through the root mean square normalization layer, dropout layer, linear layer and global average pooling layer of the regression layer for feature integration to obtain depth features;

[0033] Step 310: The feed-forward neural network efficiently fuses the HRV feature sequence and the depth features to predict the systolic blood pressure SBP and diastolic blood pressure DBP, and obtains the blood pressure detection result.

[0034] Preferably, a camera with a resolution not lower than 1080p and a frame rate not lower than 30fps is used to capture the facial dynamic image data of the person to be measured with a duration of 10 seconds and a distance of 0.1-0.8 meters in an environment with uniform and soft light and quiet and stable as a face video sample or a face video to be detected.

[0035] Preferably, the HRV feature sequence includes the time-domain features and frequency-domain features of heart rate variability (HRV), including the number of adjacent normal heartbeat interval differences greater than 50 ms (NN50), the standard deviation of all normal sinus rhythm RR intervals (SDNN), the root mean square of adjacent RR interval differences (RMSSD), the proportion of adjacent normal heartbeat interval differences greater than 50 ms (PNN50), low-frequency power (LF), high-frequency power (HF), and the ratio of low-frequency to high-frequency power (LF / HF). The state of the cardiovascular system is reflected by the HRV features.

[0036] Preferably, the specific process of step 31 is as follows:

[0037] Step 311: Perform a first-order difference operation on an IPPG signal s = {s1, s2,..., s w} of a time series to obtain a difference sequence , where w represents the length of the IPPG signal; mark the i-th position in the difference sequence that satisfies and as a peak P i , retain the peaks to construct a peak index sequence {P1, P2,..., P n}, where n represents the sequence length of the peak index sequence, and calculate the RR interval in combination with the sampling frequency ; The RR interval is expressed as:

[0038] ;

[0039] represents the RR interval at the i-th position;

[0040] Step 312: Calculate the time-domain features of HRV. Let be the extracted RR interval sequence, and n be the sequence length. Calculate:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] Step 313: Calculate the frequency-domain features of HRV. Perform a Fourier transform on the RR interval sequence to obtain a frequency sequence frequencies = [f1, f2,..., f n , and calculate the power spectral density of each frequency in the frequency sequence to obtain the corresponding PSD sequence power = [p1, p2,..., p n ;

[0046] Filter out low-frequency band frequencies from the frequency sequence to construct a low-frequency index set , where \(f\) represents frequency and \(m\) represents the number of low-frequency band frequencies; the frequency range of the low-frequency band is , calculate the low-frequency power according to the PSD sequence positions corresponding to the low-frequency index set ; represents the power spectral density; represents the power spectral density at the \(j\)-th position in the PSD sequence, represents the power spectral density at the \((j + 1)\)-th position in the PSD sequence; represents the frequency at the \((j + 1)\)-th position in the low-frequency index set; represents the frequency at the \(j\)-th position in the low-frequency index set;

[0047] Filter out high-frequency band frequencies from the frequency sequence to construct a high-frequency index set , where \(v\) represents the number of high-frequency band frequencies, and the frequency range of the high-frequency band is , calculate the high-frequency power according to the PSD sequence positions corresponding to the high-frequency index set ;

[0048] Calculate the ratio of low-frequency to high-frequency power LF / HF: .

[0049] A non-contact blood pressure detection system based on deep learning, comprising:

[0050] A face video acquisition module that acquires face video samples and the face video to be detected;

[0051] An IPPG signal extraction module that extracts frame-by-frame images from the face video samples and the face video to be detected and calculates the corresponding image-based photoplethysmogram (IPPG) signals;

[0052] A model construction module that constructs an IPPG-Mamba network model and trains it according to the IPPG signals corresponding to the face video samples to obtain a blood pressure detection model;

[0053] A blood pressure detection module that loads the blood pressure detection model to detect the IPPG signals corresponding to the face video to be detected and obtains blood pressure detection results.

[0054] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a non-contact blood pressure detection method and system based on deep learning. By combining traditional HRV feature sequences (such as RR interval, LF / HF, etc.) and deep features extracted by deep learning, making full use of the time-frequency information of IPPG signals, effectively mining the key information related to blood pressure in IPPG signals, compared with traditional non-contact detection models, significantly improving the accuracy of blood pressure detection. Specifically, a CNN-Mamba Block with powerful feature extraction capabilities is designed, which combines the characteristics of CNN and Mamba networks, introduces WTConv1d wavelet transform convolution, can finely analyze IPPG signals at different scales, effectively capture subtle and key feature changes in the signals, enhance the multi-scale feature extraction ability of the model for IPPG signals, and break through the limitation of the limited receptive field of traditional convolution. At the same time, various operations such as normalization, dimensionality reduction, and overfitting prevention within the module ensure the stability and efficiency of feature extraction; improve non-contact convenience, realize non-contact detection by shooting face videos with a camera, get rid of the dependence on professional operations of traditional contact detection, reduce patient discomfort, greatly improve the detection convenience, and facilitate wide application in scenarios such as home and mobile medical care; reasonably standardize the environment in the data acquisition link, and adopt various advanced technologies such as HRV feature sequence extraction and CNN-MambaBlock to process signals in model construction, which can effectively cope with complex environmental conditions such as different lighting and postures, reduce the interference of environmental factors on the detection results, ensure stable and reliable blood pressure detection in diverse environments, and improve the environmental adaptability of non-contact blood pressure detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0056] Figure 1 Schematic flow chart of the face video blood pressure detection method based on deep learning provided by the present invention;

[0057] Figure 2 Schematic structural diagram of the IPPG-Mamba network model provided by the present invention;

[0058] Figure 3 Schematic structural diagram of the CNN-Mamba Block provided by the present invention;

[0059] Figure 4 Schematic structural diagram of the one-dimensional wavelet transform convolution layer provided by the present invention;

[0060] Figure 5 Schematic diagram of the feedforward neural network structure provided by the present invention. Detailed implementation manners

[0061] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0062] An embodiment of the present invention discloses a non-contact blood pressure detection method based on deep learning, as Figure 1 shown, including the following steps:

[0063] S1: Collect face video samples, including face videos and blood pressure data;

[0064] S2: Extract frame-by-frame images from the face video, and calculate the pixel mean value in the images to generate an IPPG signal;

[0065] S3: Construct an IPPG-Mamba network model, and train the IPPG-Mamba network model according to the IPPG signal and blood pressure data to obtain a blood pressure detection model; the IPPG-Mamba network model includes an HRV feature extraction unit, a CNN-MambaBlock, a regression layer (Regression Layer), and a feedforward neural network; the HRV feature extraction unit extracts an HRV (Heart Rate Variability) feature sequence; the CNN-Mamba Block is a network module improved by combining a CNN network and a wavelet transform convolution Mamba Block to extract deep features; the feedforward neural network fuses the HRV feature sequence and the deep features to predict the blood pressure detection result;

[0066] S4: Collect the face video to be detected, extract frame-by-frame images, and calculate the pixel mean value in the images to generate an IPPG signal to be detected;

[0067] S5: Input the IPPG signal to be detected into the blood pressure detection model for detection to obtain a blood pressure detection result.

[0068] In a specific embodiment, the specific process of performing blood pressure detection using the non-contact blood pressure detection method based on deep learning is as follows:

[0069] S1: Use a camera to shoot a 10-second face video and collect the blood pressure data of the person being photographed;

[0070] A camera with a resolution of at least 1080p and a frame rate of at least 30fps should be used to capture a 10-second face video at a distance of 0.1 - 0.8 meters in an environment with uniform, soft light and quiet stability. During the shooting, a cuff-type sphygmomanometer should be used to measure the blood pressure of the person being photographed. After the measurement, record and save the systolic blood pressure SBP and diastolic blood pressure DBP values.

[0071] S2: Extract frame-by-frame images from the captured face video, calculate the pixel mean of each pixel point in the image, and the series of consecutive values obtained will be used as the IPPG signal.

[0072] S3: Build the IPPG-Mamba network model; the constructed IPPG-Mamba network model is as Figure 2 shown, and the IPPG-Mamba network model takes the IPPG signal as the input.

[0073] S4: Train the constructed IPPG-Mamba network model:

[0074] Specifically, divide the extracted IPPG signal and the corresponding blood pressure data of the person being photographed, with 80% as the training set and 20% as the validation set. When training, select the Adam optimizer, fix the learning rate at 0.001, the batch size at 32, and the number of training epochs at 300.

[0075] The HRV feature extraction unit of the IPPG-Mamba network model serves as an input branch. It extracts the RR interval (the time interval between adjacent peaks) from the IPPG signal through peak detection, and combines the sampling frequency to further calculate a series of HRV features, such as the number of NN50 with a difference in adjacent normal heartbeats greater than 50ms, the standard deviation SDNN of all normal sinus rhythm RR intervals, the root mean square RMSSD of the differences between adjacent RR intervals, the proportion PNN50 of adjacent normal heartbeats with a difference greater than 50ms, the low-frequency power LF, the high-frequency power HF, and the ratio LF / HF of low-frequency to high-frequency power.

[0076] The core part of the other input branch is the CNN-Mamba Block, which is improved from the Mamba network, as Figure 3As shown, it includes a Root Mean Square Normalization layer (RMS Norm), two groups of Linear Layers, a CNN network module, an activation function (SILU), a Selective State Space Model (SSM), and two groups of Residual Modules; the first group of Residual Modules includes a Linear Layer and an activation function (SILU); the CNN network module contains units composed of two groups of one-dimensional Wavelet Transform Convolution Layers (WTConv1d), Batch Normalization Layers (BatchNorm), and one-dimensional Max Pooling Layers (MaxPool1d), as well as a Dropout layer with a rate of 0.3, where the WTConv1d model module is as Figure 4 shown, including a Wavelet Transform module (WT), a one-dimensional Convolution module (Conv1d), and an Inverse Wavelet Transform module (IWT); after the IPPG signal is processed by the CNN-Mamba Block, the features will enter the regression layer, and the regression layer sequentially includes operations such as RMS Norm (Root Mean Square Normalization), Dropout (random inactivation and discard to prevent overfitting), and Linear (Linear Layer), and finally, feature integration is performed through Global Average Pooling to obtain deep features;

[0077] The features extracted from the two input branches, namely the HRV features and the deep features, are finally input into a Feed-Forward Neural Network (FNN). The structure of the Feed-Forward Neural Network is as Figure 5 shown. Through its internal Linear Layers (such as Linear 64, Linear 32, Linear 1) and activation function ELU (Exponential Linear Unit), these features with different properties are fused. After being processed by the FNN, the model finally outputs the values of Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP);

[0078] During training, the hyperparameters are adjusted based on the SBP and DBP predicted by the validation set and the network model, and the weights of the finally trained model are saved as the blood pressure detection model;

[0079] S5: Use a camera to collect a 10-second face video file of the person to be detected, then extract the IPPG signal from the face video of the person to be detected in the face video file, and input the extracted IPPG signal into the blood pressure detection model to output the blood pressure detection results including SBP and DBP.

[0080] In a specific embodiment, the process of extracting the RR interval from the IPPG signal through peak detection and calculating the HRV features in combination with the sampling frequency is as follows:

[0081] S41: Perform a first-order difference operation on the IPPG signal s = {s1, s2,..., s w} to obtain the difference sequence , where w represents the length of the IPPG signal; the difference sequence that satisfies and mark the i-th position of i as peak P i , retain the peaks to construct a peak index sequence {P1, P2,..., P n}; n represents the sequence length of the peak index sequence. Combine with the sampling frequency to calculate the RR interval; the RR interval is expressed as:

[0082] ;

[0083] represents the RR interval at the i-th position;

[0084] S42: Calculate the time-domain features of HRV. Let be the extracted RR interval sequence, and n be the sequence length. Calculate:

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] S43: Calculate the frequency-domain features of HRV. Perform Fourier transform on the RR interval sequence to obtain a frequency sequence frequencies = [f1, f2,..., f n , and calculate the power spectral density to obtain the corresponding PSD sequence power = [p1, p2,..., p n ;

[0090] Filter out the low-frequency band frequencies from the frequency sequence to construct a low-frequency index set , f represents the frequency, and m represents the number of low-frequency band frequencies; the frequency range of the low-frequency band is . Calculate the low-frequency power according to the positions of the PSD sequence corresponding to the low-frequency index set ; represents the power spectral density; represents the power spectral density at the j-th position in the PSD sequence, represents the power spectral density at the (j + 1)-th position in the PSD sequence; represents the frequency at the (j + 1)-th position in the low-frequency index set; represents the frequency at the j-th position in the low-frequency index set;

[0091] Filter out the high-frequency band frequencies from the frequency sequence to construct a high-frequency index set , where \(v\) represents the number of high - frequency band frequencies, and the frequency range of the high - frequency band is , calculate the high - frequency power according to the PSD sequence position corresponding to the high - frequency index set ;

[0092] Calculate the ratio of low - frequency to high - frequency power LF / HF: .

[0093] On the other hand, in a specific embodiment, a non - contact blood pressure detection system based on deep learning includes:

[0094] A face video acquisition module that acquires face video samples and the face video to be detected;

[0095] An IPPG signal extraction module that extracts frame - by - frame images from the face video samples and the face video to be detected and calculates the corresponding IPPG signals;

[0096] A model construction module that constructs an IPPG - Mamba network model and trains it according to the IPPG signals corresponding to the face video samples to obtain a blood pressure detection model;

[0097] A blood pressure detection module that loads the blood pressure detection model to detect the IPPG signals corresponding to the face video to be detected and obtains a blood pressure detection result.

[0098] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0099] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A non-contact blood pressure detection method based on deep learning, characterized in that, It includes the following steps: Step 1: Collect face video samples, including face videos and blood pressure data; Step 2: Extract frame-by-frame images from the face video, calculate the pixel mean in the images, and generate IPPG signals; Step 3: Construct an IPPG-Mamba network model, and train the IPPG-Mamba network model according to the IPPG signals and blood pressure data to obtain a blood pressure detection model; the IPPG-Mamba network model includes an HRV feature extraction unit, a CNN-MambaBlock, a regression layer, and a feedforward neural network; the HRV feature extraction unit extracts HRV feature sequences; the CNN-Mamba Block is a network module improved by combining the CNN network and the wavelet transform convolution Mamba Block, which extracts fused features; the regression layer integrates the fused features to obtain deep features; the feedforward neural network fuses the HRV feature sequences and the deep features to predict the blood pressure detection result; Step 4: Collect the face video to be detected, extract frame-by-frame images, and calculate the pixel mean in the images to generate the IPPG signal to be detected; Step 5: Input the IPPG signal to be detected into the blood pressure detection model for detection to obtain the blood pressure detection result.

2. The non-contact blood pressure detection method based on deep learning according to claim 1, wherein The feedforward neural network includes a linear layer and an activation function.

3. The non-contact blood pressure detection method based on deep learning according to claim 1, characterized in that, The CNN-Mamba Block includes a root mean square normalization layer, two groups of linear layers, a CNN network module, an activation function, a selective state space model, and two groups of residual modules; the CNN network module includes two groups of transformation modules and a dropout layer, and each group of transformation modules includes a one-dimensional wavelet transform convolutional layer, a batch normalization layer, and a one-dimensional max pooling layer; the first group of residual modules includes a linear layer and an activation function.

4. The non-contact blood pressure detection method based on deep learning according to claim 1, characterized in that, The regression layer includes a root mean square normalization layer, a dropout layer, a linear layer, and a global average pooling layer.

5. The non-contact blood pressure detection method based on deep learning according to claim 3, characterized in that The one-dimensional wavelet transform convolutional layer includes a wavelet transform module, a one-dimensional convolutional module, and an inverse wavelet transform module connected in sequence.

6. The non-contact blood pressure detection method based on deep learning according to claim 3, wherein, The process of the blood pressure detection model for detection is as follows: Step 31: The HRV feature extraction unit extracts the RR interval from the IPPG signal through peak detection, and calculates the HRV feature sequence according to the RR interval; Step 32: The root mean square normalization layer normalizes the IPPG signal to obtain a normalized input signal; Step 33: The first group of linear layers performs a linear transformation on the normalized input signal to obtain linear features; The linear layer of the first group of residual modules performs a linear transformation and introduces non-linearity to the normalized input signal to obtain the first residual feature; Step 34: The linear features output by the first group of linear layers obtain multi-scale convolutional features through the CNN network module, and introduce non-linearity through the activation function to obtain non-linear features; Step 35: The selective state space model selectively processes information according to the non-linear features through recursion to obtain selective features, and performs a multiplication operation on the selective features and the first residual features, and the multiplication result; Step 36: The second group of linear layers performs a linear transformation on the multiplication result to obtain multiplication linear features, and performs an addition operation in combination with the IPPG signal transmitted by the second group of residual modules to obtain fused features, and then performs feature integration through the regression layer to obtain deep features; Step 37: The feedforward neural network fuses the HRV feature sequence and the depth features to predict the systolic blood pressure and diastolic blood pressure, obtaining the blood pressure detection result.

7. The non-contact blood pressure detection method based on deep learning according to claim 1, characterized in that Use a camera with a resolution of not less than 1080p and a frame rate of not less than 30fps to capture the facial dynamic image data of the person to be measured with a duration of 10 seconds and a distance of 0.1 - 0.8 meters as the face video sample or the face video to be detected.

8. The non-contact blood pressure detection method based on deep learning according to claim 6, characterized in that, The HRV feature sequence includes the time-domain features and frequency-domain features of HRV. The time-domain features include the number of adjacent normal heartbeat interval differences greater than 50ms, the standard deviation of all normal sinusoidal heartbeat RR intervals, the root mean square of adjacent RR interval differences, and the proportion of adjacent normal heartbeat interval differences greater than 50ms. The frequency-domain features include low-frequency power, high-frequency power, and the ratio of low-frequency to high-frequency power.

9. The non-contact blood pressure detection method based on deep learning according to claim 8, characterized in that, The specific process of Step 31 is as follows: Step 311: Perform a first-order difference operation on the IPPG signal s = {s1, s2,..., s w}, to obtain a difference sequence , where w represents the length of the IPPG signal; mark the i-th position in the difference sequence that satisfies and as the peak P i , retain the peaks to construct a peak index sequence {P1, P2,..., P n}, where n represents the sequence length of the peak index sequence, and calculate the RR interval in combination with the sampling frequency ; the RR interval is expressed as: ； denote the RR interval at the i-th position; Step 312: Calculate the time-domain features and frequency-domain features of HRV.

10. A non-contact blood pressure detection system based on deep learning, characterized in that, Apply the non-contact blood pressure detection method based on deep learning according to any one of claims 1 - 9, including: A face video acquisition module that acquires face video samples and face videos to be detected; An IPPG signal extraction module that extracts frame-by-frame images from the face video samples and face videos to be detected and calculates the corresponding IPPG signals; A model construction module that constructs an IPPG-Mamba network model and trains it according to the IPPG signals corresponding to the face video samples to obtain a blood pressure detection model; A blood pressure detection module that loads the blood pressure detection model to detect the IPPG signals corresponding to the face videos to be detected, obtaining the blood pressure detection result.

Images