Attention-based thoraco-abdominal motion signal segmentation method, system and storage medium

Abstract

The present application relates to the technical field of biomedical radar signal processing and artificial intelligence, in particular to a chest and abdominal motion signal segmentation method and system based on attention and a storage medium, the segmentation method comprising the following steps: obtaining physiological signals obtained by detecting a human body by a millimeter wave radar; performing feature extraction on the obtained physiological signals to obtain a multi-scale feature map; using a constructed attention fusion model to fuse time domain attention, space domain attention and channel attention of the multi-scale feature map to obtain a fusion signal based on attention enhancement; inputting the fusion signal based on attention enhancement into a segmentation network model to obtain a respiratory motion signal and a chest and abdominal motion signal. The present application uses a deep learning model to process FMCW millimeter wave radar signals, enhances relevant features of chest and abdominal motion through an attention mechanism, accurately segments the chest and abdominal motion signals, and provides reliable input for subsequent calculation of respiratory frequency and heart rate.

Description

Technical Field

[0001] This invention relates to the field of biomedical radar signal processing and artificial intelligence, specifically to an attention-based method, system, and storage medium for segmenting chest and abdominal motion signals. Background Technology

[0002] Respiratory rate and heart rate are vital signs in the human body, and their real-time monitoring is of great significance for health management and early warning of diseases. Traditional detection methods (such as chest straps and optical sensors) usually require direct contact with the human body, which not only affects the user's comfort but is also limited by environmental conditions.

[0003] FMCW millimeter-wave radar, as a non-contact, non-invasive monitoring method, can capture minute displacements on the body surface (such as chest and abdominal movements caused by breathing or heartbeat) by transmitting and receiving millimeter-wave signals, showing great potential in the field of vital sign monitoring. However, since radar signals are susceptible to background noise and multipath effects, accurately separating chest and abdominal movement signals from complex millimeter-wave signals has become a challenge for existing technologies. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide an attention-based method, system and storage medium for segmenting chest and abdominal motion signals, thereby solving the difficult problem of how to accurately separate chest and abdominal motion signals from millimeter-wave signals.

[0005] The technical solution to achieve the above objectives is:

[0006] This invention provides an attention-based method for segmenting chest and abdominal motion signals, comprising the following steps:

[0007] Acquire physiological signals obtained from millimeter-wave radar detection of the human body;

[0008] Feature extraction is performed on the acquired physiological signals to obtain multi-scale feature maps;

[0009] An attention fusion model is constructed, and the extracted multi-scale feature maps are fused with temporal attention, spatial attention, and channel attention to obtain an attention-enhanced fusion signal.

[0010] A segmentation network model is provided, and the attention-enhanced fusion signal is input into the provided segmentation network model to obtain the respiratory motion signal and chest and abdominal motion signal output by the segmentation network model.

[0011] A further improvement of the attention-based chest and abdominal motion signal segmentation method of this invention lies in the following steps in constructing the attention fusion model:

[0012] Construct a temporal attention computation model. GRU stands for Gated Loop Unit. W t Let F be the time-series feature weight matrix. t For multi-scale feature maps F l Temporal feature sequences after spatial dimension compression;

[0013] Construct a spatial attention computation model. W s F represents the spatial convolution kernel parameters. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension;

[0014] Construct a channel attention calculation model. W c1 W is the weight matrix of the global average pooling features. c2 F is the weight matrix of the global max pooling feature. c For multi-scale feature maps F l Channel features after global pooling in the spatiotemporal dimension;

[0015] Construct a spatiotemporal cross-attention computation model. ,

[0016] Construct a spatiotemporal correlation matrix. , where d k Q is the attention dimension normalization factor. st For the spatiotemporal query matrix, K st Let W be the spacetime bond matrix. q To query the projection weight matrix, W k H×W is the key projection weight matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames.

[0017] Construct an attention fusion computation model. ,in, This is element-wise multiplication. For tensor outer product, The attention weight coefficients are used to fuse temporal attention, spatial attention, and channel attention in the extracted multi-scale feature maps using the constructed attention fusion calculation model, so as to obtain a fusion signal based on attention enhancement.

[0018] A further improvement of the attention-based chest and abdominal motion signal segmentation method of this invention lies in providing a segmentation network model comprising the following steps:

[0019] Based on the U-Net network architecture, the encoder part is defined as follows: DownBlock l Includes convolutional layers and downsampling: ;

[0020] The decoder part for skip connections and feature fusion is configured as follows: UpBlock l Includes upsampling and feature concatenation: , ;

[0021] Set the split output layer as follows: ,

[0022] Generate a 3D probability map: .

[0023] A further improvement of the attention-based chest and abdominal motion signal segmentation method of this invention lies in the following steps for feature extraction of the acquired physiological signals:

[0024] Construct a multi-scale feature extraction module: ,in, ResBlock provides the original 3D feature maps of distance, velocity, and angle. l For residual blocks, MaxPool l-1 For max pooling;

[0025] The original three-dimensional feature maps of distance, velocity, and angle are obtained by calculation from the acquired physiological signals, using the following formula:

[0026] The formula for calculating distance dimension data is as follows: , The frequency of the difference frequency signal. ,in, To adjust the frequency, T c For coherent processing time, distance resolution c is the speed of light, and B is the bandwidth. The acquired physiological signal is expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let n be the radial velocity of the nth target point. For the initial phase, It is Gaussian white noise. Wavelength;

[0027] The formula for calculating velocity dimension data is as follows: ,in Let K be the range spectrum of the k-th frame, with velocity resolution of . T s f is the frame interval. d The frequency is the Doppler frequency.

[0028] The formula for calculating angular dimension data is as follows: , where the angular resolution is M is the number of antennas, d is the antenna spacing, and S m Let m be the range Doppler spectrum value output by the m-th antenna channel. Let m be the azimuth angle of the target relative to the normal direction of the radar antenna array, and m be the antenna index.

[0029] A further improvement of the attention-based chest and abdominal motion signal segmentation method of the present invention is that it also includes calculating the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signal and chest and abdominal motion signal, so as to quantify the complexity of the breathing pattern.

[0030] The formula for calculating the root mean square is: ,in The duration of the i-th respiratory cycle;

[0031] The formula for calculating approximate entropy is: ,in This represents the template matching probability.

[0032] The present invention also provides a storage medium storing a program for an attention-based chest and abdominal motion signal segmentation method.

[0033] When the program of the attention-based chest and abdominal motion signal segmentation method is executed by the processor, it implements the steps of the attention-based chest and abdominal motion signal segmentation method.

[0034] This invention also provides an attention-based chest and abdominal motion signal segmentation system, comprising:

[0035] The acquisition unit is used to acquire physiological signals obtained by millimeter-wave radar from human detection.

[0036] A feature extraction unit, connected to the acquisition unit, is used to extract features from the acquired physiological signals to obtain a multi-scale feature map.

[0037] An attention fusion model, connected to the feature extraction unit, is used to fuse temporal attention, spatial attention, and channel attention on the extracted multi-scale feature maps to obtain an attention-enhanced fusion signal.

[0038] A segmentation network model, connected to the attention fusion model, is used to segment the enhanced fusion signal to obtain respiratory motion signals and chest and abdominal motion signals.

[0039] A further improvement of the attention-based chest and abdominal motion signal segmentation system of the present invention is that the attention fusion model includes:

[0040] Temporal attention computation model Where GRU is a gated recurrent unit, W t Let F be the time-series feature weight matrix. t For multi-scale feature maps F l Temporal feature sequences after spatial dimension compression;

[0041] Spatial attention computation model, W s F represents the spatial convolution kernel parameters. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension;

[0042] Channel attention calculation model, W c1 W is the weight matrix of the global average pooling features. c2 F is the weight matrix of the global max pooling feature. c For multi-scale feature maps F l Channel features after global pooling in the spatiotemporal dimension;

[0043] Spatiotemporal cross-attention computation model, ,

[0044] Spatiotemporal correlation matrix , where d k Q is the attention dimension normalization factor. st For the spatiotemporal query matrix, K st Let W be the spacetime bond matrix. q To query the projection weight matrix, W k H×W is the key projection weight matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames.

[0045] Attention fusion computation model ,in, This is element-wise multiplication. For tensor outer product, The attention weight coefficients are used to fuse temporal attention, spatial attention, and channel attention into the extracted multi-scale feature maps to obtain an attention-enhanced fusion signal.

[0046] A further improvement of the attention-based chest and abdominal motion signal segmentation system of the present invention is that the feature extraction unit includes a multi-scale feature extraction module, wherein the multi-scale feature extraction module is as follows: ,in, ResBlock provides the original 3D feature maps of distance, velocity, and angle. l For residual blocks, MaxPool l-1 For max pooling;

[0047] The original three-dimensional feature maps of distance, velocity, and angle are obtained by calculation from the acquired physiological signals, using the following formula:

[0048] The formula for calculating distance dimension data is as follows: , The frequency of the difference frequency signal. ,in, For frequency modulation, Tc is the coherent processing time, and distance resolution. c is the speed of light, and B is the bandwidth. The acquired physiological signal is expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let n be the radial velocity of the nth target point. For the initial phase, It is Gaussian white noise. Wavelength;

[0049] The formula for calculating velocity dimension data is as follows: ,in Let k be the range spectrum, and let velocity resolution be... T s f is the frame interval. d The frequency is the Doppler frequency.

[0050] The formula for calculating angular dimension data is as follows: , where the angular resolution is M is the number of antennas, d is the antenna spacing, and S m Let m be the range Doppler spectrum value output by the m-th antenna channel. Let m be the azimuth angle of the target relative to the normal direction of the radar antenna array, and m be the antenna index.

[0051] A further improvement of the attention-based chest and abdominal motion signal segmentation system of the present invention is that it also includes a quantization calculation unit connected to the segmentation network model, which is used to calculate the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signal and chest and abdominal motion signal, so as to quantify the complexity of the respiratory pattern.

[0052] The formula for calculating the root mean square is: ,in The duration of the i-th respiratory cycle;

[0053] The formula for calculating approximate entropy is: ,in This represents the template matching probability.

[0054] The beneficial effects of the attention-based chest and abdominal motion signal segmentation method, system, and storage medium of this invention are as follows:

[0055] The present invention relates to an attention-based method, system, and storage medium for segmenting chest and abdominal motion signals. This method utilizes a deep learning model to process FMCW millimeter-wave radar signals and enhances the relevant features of chest and abdominal motion through an attention mechanism, thereby accurately segmenting chest and abdominal motion signals and providing reliable input for subsequent calculations of respiratory rate and heart rate.

[0056] The attention-based chest and abdominal motion signal segmentation method, system, and storage medium of the present invention are particularly suitable for scenarios such as sleep apnea monitoring, chronic obstructive pulmonary disease (COPD) assessment, and postoperative rehabilitation exercise analysis, providing high spatiotemporal resolution quantitative biomarker data for clinical decision-making. Attached Figure Description

[0057] Figure 1 This is a flowchart of the attention-based chest and abdominal motion signal segmentation method of the present invention.

[0058] Figure 2 This is a system diagram of the attention-based chest and abdominal motion signal segmentation system of the present invention. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0060] See Figure 1 This invention provides an attention-based method, system, and storage medium for chest and abdominal motion signal segmentation, addressing the technical challenges of chest and abdominal motion segmentation in millimeter-wave radar signals. It aims to achieve efficient and accurate chest and abdominal motion segmentation of physiological signals (such as respiratory signals and body surface motion signals) through deep learning technology combined with attention mechanisms, providing precise data and feature support for respiratory monitoring, disease diagnosis, and rehabilitation assessment models. The attention-based chest and abdominal motion signal segmentation method, system, and storage medium of this invention are described below with reference to the accompanying drawings.

[0061] See Figure 2 The diagram shows a system diagram of the attention-based chest and abdominal motion signal segmentation system of the present invention. The following is in conjunction with... Figure 2 The attention-based chest and abdominal motion signal segmentation system of the present invention will be described.

[0062] like Figure 2As shown, the attention-based chest and abdominal motion signal segmentation system of the present invention includes an acquisition unit 21, a feature extraction unit 22, an attention fusion model 23, and a segmentation network model 24. The acquisition unit 21 is connected to the feature extraction unit 22, the feature extraction unit 22 is connected to the attention fusion model 23, and the attention fusion model 23 is connected to the segmentation network model 24. The acquisition unit 21 is used to acquire physiological signals obtained by millimeter-wave radar from human body detection. These physiological signals include respiratory signals, body surface motion signals, etc. The millimeter-wave radar can transmit and receive millimeter-wave signals to capture small displacements on the human body surface, such as chest and abdominal movements caused by breathing or heartbeat.

[0063] FMCW millimeter-wave radar acquires data by transmitting linear frequency modulated signals, which can be represented as: ,in, The starting frequency, To adjust the frequency, B represents bandwidth, T represents scan period, A represents amplitude, and j represents imaginary unit. This is used to construct the FMCW millimeter-wave radar difference frequency signal in the complex domain. Its core function is to completely preserve the signal change information caused by minute movements of the chest and abdomen, such as breathing, heartbeat fluctuations, and other micro-body movements. Compared with the problem of losing details in traditional real-domain signal processing, this complex representation provides key phase dimension data support for subsequent accurate segmentation of chest and abdominal motion signals. It is the basic mathematical carrier for achieving high-resolution extraction of chest and abdominal motion.

[0064] The difference frequency signal after the signal is mixed is: ,

[0065] Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let n be the radial velocity of the nth target point. For the initial phase, It is Gaussian white noise. The wavelength is determined by the center frequency of the FMCW millimeter-wave radar used. It is certain that the center frequency of a radar is an inherent parameter, determined during radar design and manufacturing, based on the relationship between wavelength and frequency. It can be calculated that the difference frequency signal is the physiological signal to be segmented by the attention-based chest and abdominal motion signal segmentation system of the present invention, and the difference frequency signal is a type of millimeter-wave signal received by millimeter-wave radar.

[0066] The feature extraction unit 22 is used to extract features from the acquired physiological signals to obtain multi-scale feature maps;

[0067] Attention fusion model 23 is used to fuse temporal attention, spatial attention and channel attention on the extracted multi-scale feature maps to obtain a fusion signal based on attention enhancement;

[0068] The segmentation network model 24 is used to segment the enhanced fused signal to obtain respiratory motion signals and chest and abdominal motion signals.

[0069] The attention-based chest and abdominal motion signal segmentation system of the present invention further includes a processing unit, which is connected to the aforementioned acquisition unit 21, feature extraction unit 22, attention fusion model 23, and segmentation network model 24. The processing unit is used to send the physiological signals acquired by the acquisition unit 21 to the feature extraction unit 22 for feature extraction. The processing unit is also used to send the extracted features to the attention fusion model 23 for the fusion of temporal attention, spatial attention, and channel attention to obtain an attention-enhanced fusion signal. The processing unit is also used to send the enhanced fusion signal to the segmentation network model 24 for segmentation, and then receive the respiratory motion signal and chest and abdominal motion signal output by the segmentation network model 24.

[0070] In one specific embodiment of the present invention, the feature extraction unit 22 includes a multi-scale feature extraction module, which is as follows: ,in, ResBlock provides the original 3D feature maps of distance, velocity, and angle. l For residual blocks, MaxPool l-1 For max pooling;

[0071] The original distance, velocity, and angle 3D feature maps are obtained by calculating the acquired physiological signals, using the following formula:

[0072] The formula for calculating distance dimension data is as follows: , The frequency of the difference frequency signal. ,in, For frequency modulation, Tc is the coherent processing time, and distance resolution. c is the speed of light, and B is the bandwidth. The acquired physiological signal is expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let n be the radial velocity of the nth target point. For the initial phase, It is Gaussian white noise. The wavelength is determined by the center frequency of the FMCW millimeter-wave radar used. It is certain that the center frequency of a radar is an inherent parameter, determined during radar design and manufacturing, based on the relationship between wavelength and frequency. It can be calculated;

[0073] The formula for calculating velocity dimension data is as follows: ,in Let k be the range spectrum, and let velocity resolution be... T s f is the frame interval. d The Doppler frequency (unit: Hz) is used to characterize the frequency shift caused by the radial motion of the target.

[0074] The formula for calculating angular dimension data is as follows: , where the angular resolution is M is the number of antennas, d is the antenna spacing, and S m for , where is the range-Doppler spectrum value output by the m-th antenna channel, i.e., the complex spectral components of the received signal after processing in the range and velocity dimensions. Let m be the azimuth angle of the target relative to the normal direction of the radar antenna array, and m be the antenna index.

[0075] ResBlock l The method includes convolutional layers, batch normalization, and the ReLU function, as follows:

[0076] ReLU activation: ,

[0077] Residual connection: .in, This is a learnable weight matrix for the i-th convolutional layer in the l-th residual block (ResBlock). Its dimension matches the number of channels in the input feature map and the size of the convolutional kernel. During training, it can adaptively learn key features of chest and abdominal movement, such as the local signal differences in chest cavity rise and fall and abdominal cavity expansion during the respiratory cycle. This overcomes the limitation of traditional fixed-weight filtering in capturing dynamic features of chest and abdominal movement, and provides an optimizable parameter basis for the accurate extraction of multi-scale chest and abdominal movement features. This weight matrix and the attention weight coefficients (α, β, γ, δ) in the model are iteratively updated through an adaptive gradient algorithm with the goal of minimizing the cross-entropy loss function, that is, minimizing the deviation between the chest and abdominal movement segmentation results and the labeled data, and finally converge to the optimal value that is suitable for the chest and abdominal movement segmentation task. The output feature map of the i-th layer in the l-th residual block (ResBlock) is obtained by " The multidimensional tensor, processed by "convolutional feature extraction, batch normalization (BN) to eliminate gradient offset, and ReLU activation to enhance nonlinear expression," has dimensions [B, H, W, C], where B is the batch frame number, H / W is the spatial dimension, and C is the number of feature channels. Its core function is to preserve and enhance chest and abdominal motion-related features layer by layer, such as the chest and abdominal cavity boundary features in the distance dimension and the respiratory motion rate features in the velocity dimension. This lays a high-quality feature foundation for the subsequent "three-way attention mechanism" to focus on key regions and features. The value of i is the index of the convolutional layer inside the l-th residual block (ResBlock), satisfying 1≤i≤K (K is the total number of convolutional layers in a single ResBlock, and the value of K is related to the actual device parameters). This layered design can achieve gradual enhancement of chest and abdominal motion features. The shallow layer can capture basic distance / velocity features, and the deep layer can extract subtle differences in chest and abdominal cavity motion features, adapting to the multi-scale and non-stationary signal characteristics of chest and abdominal motion and avoiding the problem of insufficient feature extraction at a single level.

[0078] In one specific embodiment of the present invention, the attention fusion model 23 includes:

[0079] Temporal attention computation model In this context, GRU stands for Gated Cyclic Unit, which captures the temporal dependence of respiratory movements and outputs temporal weights At to enhance feature differentiation between the inspiratory and expiratory phases. Wt is the temporal feature weight matrix. C represents the number of feature channels, used to perform linear transformation on the temporal features output by the GRU; The data is obtained automatically through the model training process. During the training phase, the cross-entropy loss function of chest and abdominal motion segmentation is used as the optimization objective, and the Adam optimizer is employed, with iterative updates via backpropagation. The parameter values ​​are calculated until the model converges, and Ft is the multi-scale feature map F. l Temporal feature sequences after spatial dimension compression T represents the number of time frames, and C represents the number of feature channels;

[0080] Spatial attention computation model, Where Ws is the spatial convolution kernel parameter, (where k is the kernel size and C is the number of feature channels). The method of obtaining F: It is automatically learned through the model training process. During the training phase, with the segmentation loss function as the objective, it is iteratively updated through gradient descent until the model's segmentation accuracy on the validation set no longer improves. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension. H×W represents the spatial dimensions, and C represents the number of feature channels. By aggregating temporal information, the convolutional layer can more accurately locate the spatial boundaries of the thoracic and abdominal regions. When performing spatial attention calculations, the convolutional kernel can be adjusted according to the sampling rate, or a kernel can be manually set. While meeting computational accuracy requirements, the kernel should be set as small as possible to reduce computational load. For example, if the sampling rate is 40Hz, a 3×3 convolutional kernel can be used to generate a spatial weight map, highlighting the features of the thoracic and abdominal regions and accurately locating the boundary between the ribs and the abdominal cavity.

[0081] Channel attention calculation model, W c1 The weight matrix for the global average pooling features. W c2 The weight matrix for the global max-pooling features. r is the channel compression ratio; and The acquisition method is as follows: obtained through end-to-end self-learning of the model. During training, the cross-entropy loss combined with the chest-abdomen phase difference consistency loss is used as the overall objective. Parameters are updated through backpropagation until the model converges. Fc is the channel feature of the multi-scale feature map Fl after global pooling in the spatiotemporal dimension. C represents the number of feature channels, which is an attention mechanism used to enhance respiratory-related features.

[0082] Spatiotemporal cross-attention computation model, ,

[0083] Spatiotemporal correlation matrix , where d k Q is the attention dimension normalization factor. st For spatiotemporal query matrix, K is used to characterize the query features of each time frame. st For the space-time bond matrix, W is used to characterize the bond features at each spatial location. q To query the projection weight matrix, W is used to map input features to a query matrix. k The key projection weight matrix, , is used to map input features to a key matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames. The acquisition method: obtained through self-learning during the model training process. During the training phase, these values ​​are compared with those in the model. The optimization process involves multiple components, with the overall objective being a combination of the cross-entropy loss function for chest and abdominal motion segmentation and a physiological constraint loss. The Adam optimizer is employed, iteratively updating parameter values ​​via backpropagation until the model's segmentation accuracy on the validation set no longer improves, at which point convergence is achieved. . It is through input features and respectively Obtained through matrix operations. Specifically, the processed features containing spatiotemporal information are respectively compared with... Multiplying, and after adjusting the dimensions accordingly, yields... This information is then used to calculate the subsequent spatiotemporal correlation matrix to capture the dynamic correlation features between time frames and spatial locations. By calculating the feature similarity between each time frame and spatial location using the spatiotemporal correlation matrix, dynamic correlations such as "the change of the anterior convexity area of ​​the chest cavity over time during inhalation" can be captured.

[0084] Attention fusion computation model This is used to fuse temporal attention, spatial attention, and channel attention into the extracted multi-scale feature maps to obtain an attention-enhanced fused signal. This is element-wise multiplication. For tensor outer product, These are the attention weight coefficients; spatiotemporal-channel attention weight fusion is achieved through tensor outer product, and the weight coefficients are optimized using an adaptive gradient algorithm.

[0085] ,

[0086] in, This is element-wise division, EMA is the exponential moving average, and η is the learning rate. It is the minimum constant. This is the cross-entropy loss function.

[0087] In one specific embodiment of the present invention, the segmentation network model 24 adopts a variant of the U-Net network architecture, that is, based on the U-Net network architecture, the encoder part is set as follows: DownBlock l Includes convolutional layers and downsampling: ;

[0088] The decoder part for skip connections and feature fusion is configured as follows: UpBlock l Includes upsampling and feature concatenation: , ;

[0089] Set the split output layer as follows: ,

[0090] Generate a 3D probability map: , representing the segmentation probability of the three regions: background, thoracic cavity, and abdominal cavity.

[0091] When segmenting the attention-enhanced fusion signal, the attention-enhanced fusion signal is input into the segmentation network model 24. The segmentation network model 24 generates a three-dimensional probability map of the attention-enhanced fusion signal, and then performs spatiotemporal integration on the three-dimensional probability map to extract the respiratory motion signal.

[0092]

[0093] Where, r i For spatial coordinates, Let be the segmentation probability.

[0094] The chest-abdomen phase difference can be obtained by calculating the phase using Hilbert transform and then smoothing it using Kalman filtering.

[0095]

[0096] Where F is the state transition matrix and H is the observation matrix. This is the Kalman gain.

[0097] In one specific embodiment of the present invention, the attention-based chest and abdominal motion signal segmentation system of the present invention further includes a quantization calculation unit connected to the segmentation network model, which is used to calculate the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signal and chest and abdominal motion signal, so as to quantify the complexity of the respiratory pattern.

[0098] The formula for calculating the root mean square is: ,in The duration of the i-th respiratory cycle;

[0099] The formula for calculating approximate entropy is: ,in This represents the template matching probability.

[0100] The segmentation system of this invention achieves accurate segmentation of chest and abdominal movements based on spatiotemporal attention segmentation, providing reliable technical support for respiratory monitoring and disease diagnosis.

[0101] This invention also provides an attention-based method for segmenting chest and abdominal motion signals, comprising the following steps:

[0102] Acquire physiological signals obtained from millimeter-wave radar detection of the human body;

[0103] Feature extraction is performed on the acquired physiological signals to obtain multi-scale feature maps;

[0104] An attention fusion model is constructed, and the extracted multi-scale feature maps are fused with temporal attention, spatial attention, and channel attention to obtain an attention-enhanced fusion signal.

[0105] A segmentation network model is provided, and the attention-enhanced fusion signal is input into the provided segmentation network model to obtain the respiratory motion signal and chest and abdominal motion signal output by the segmentation network model.

[0106] In one specific embodiment of the present invention, constructing an attention fusion model includes the following steps:

[0107] Construct a temporal attention computation model. Where GRU is a gated recurrent unit, W t Let F be the time-series feature weight matrix. t For multi-scale feature maps F l Temporal feature sequences after spatial dimension compression;

[0108] Construct a spatial attention computation model. W s F represents the spatial convolution kernel parameters. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension;

[0109] Construct a channel attention calculation model. W c1 W is the weight matrix of the global average pooling features. c2 F is the weight matrix of the global max pooling feature. c For multi-scale feature maps F l Channel features after global pooling in the spatiotemporal dimension;

[0110] Construct a spatiotemporal cross-attention computation model. ,

[0111] Construct a spatiotemporal correlation matrix. , where d k Q is the attention dimension normalization factor. st For the spatiotemporal query matrix, K st Let W be the spacetime bond matrix. q To query the projection weight matrix, W k H×W is the key projection weight matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames.

[0112] Construct an attention fusion computation model. ,in, This is element-wise multiplication. For tensor outer product, The attention weight coefficients are used to fuse temporal attention, spatial attention, and channel attention in the extracted multi-scale feature maps using the constructed attention fusion calculation model, so as to obtain a fusion signal based on attention enhancement.

[0113] In one specific embodiment of the present invention, providing a segmentation network model includes the following steps:

[0114] Based on the U-Net network architecture, the encoder part is defined as follows: DownBlock l Includes convolutional layers and downsampling: ;

[0115] The decoder part for skip connections and feature fusion is configured as follows: UpBlock l Includes upsampling and feature concatenation: , ;

[0116] Set the split output layer as follows: ,

[0117] Generate a 3D probability map: .

[0118] When segmenting the attention-enhanced fusion signal, the signal is input into a segmentation network model. The model generates a 3D probability map from the signal, and then performs spatiotemporal integration on the map to extract respiratory motion signals.

[0119]

[0120] Where, r i For spatial coordinates, Let be the segmentation probability.

[0121] The chest-abdomen phase difference can be obtained by calculating the phase using Hilbert transform and then smoothing it using Kalman filtering.

[0122]

[0123] Where F is the state transition matrix and H is the observation matrix. This is the Kalman gain.

[0124] In one specific embodiment of the present invention, feature extraction of the acquired physiological signals includes the following steps:

[0125] Construct a multi-scale feature extraction module: ,in, ResBlock provides the original 3D feature maps of distance, velocity, and angle. l For residual blocks, MaxPool l-1 For max pooling;

[0126] The original three-dimensional feature maps of distance, velocity, and angle were obtained by calculating the acquired physiological signals, using the following formula:

[0127] The formula for calculating distance dimension data is as follows: , The frequency of the difference frequency signal. ,in, To adjust the frequency, T c For coherent processing time, distance resolution c is the speed of light, and B is the bandwidth. The acquired physiological signal is expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let n be the radial velocity of the nth target point. For the initial phase, It is Gaussian white noise. Wavelength;

[0128] The formula for calculating velocity dimension data is as follows: ,in Let K be the range spectrum of the k-th frame, with velocity resolution of . T s f is the frame interval. d The frequency is the Doppler frequency.

[0129] The formula for calculating angular dimension data is as follows: , where the angular resolution is M represents the number of antennas, d represents the antenna spacing, and Sm represents the range Doppler spectrum value output by the m-th antenna channel. Let m be the azimuth angle of the target relative to the normal direction of the radar antenna array, and m be the antenna index.

[0130] In one specific embodiment of the present invention, the method further includes calculating the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signal and chest and abdominal motion signal, so as to quantify the complexity of the respiratory pattern.

[0131] The formula for calculating the root mean square is: ,in The duration of the i-th respiratory cycle;

[0132] The formula for calculating approximate entropy is:

[0133] ,in This represents the template matching probability.

[0134] The present invention also provides a storage medium storing a program for an attention-based chest and abdominal motion signal segmentation method, wherein the program for the attention-based chest and abdominal motion signal segmentation method is executed by a processor and implements the steps of the attention-based chest and abdominal motion signal segmentation method.

[0135] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. Those skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention, and the scope of protection of the present invention shall be defined by the appended claims.

Claims

1. A method for segmenting chest and abdominal motion signals based on attention, characterized in that, Includes the following steps: Acquire physiological signals obtained from millimeter-wave radar detection of the human body; Feature extraction is performed on the acquired physiological signals to obtain multi-scale feature maps; An attention fusion model is constructed, and the extracted multi-scale feature maps are fused with temporal attention, spatial attention, and channel attention to obtain an attention-enhanced fusion signal. A segmentation network model is provided, and the attention-enhanced fusion signal is input into the provided segmentation network model to obtain the respiratory motion signal and chest and abdominal motion signal output by the segmentation network model. Building an attention fusion model involves the following steps: Construct a temporal attention computation model, A t =Sigmoid(W t ·GRU(F t ), where GRU is a gated loop unit, W t Let F be the time-series feature weight matrix. t For multi-scale feature maps F l Temporal feature sequences after spatial dimension compression; Construct a spatial attention computation model, A s =Sigmoid(W s ·Conv2D(F s )), where W s F represents the spatial convolution kernel parameters. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension; Construct a channel attention calculation model. A c =Sigmaoid(W c1 ·GobalAvgPool(F c )+W c2 ·GlobalMaxPool(F c )), where W c1 W is the weight matrix of the global average pooling features. c2 F is the weight matrix of the global max pooling feature. c For multi-scale feature maps F l Channel features after global pooling in the spatiotemporal dimension; Construct a spatiotemporal cross-attention computation model. Construct a spatiotemporal correlation matrix, A st ∈R T×(H×W) , where d k Q is the attention dimension normalization factor. st For the spatiotemporal query matrix, K st Let W be the spacetime bond matrix. q To query the projection weight matrix, W k H×W is the key projection weight matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames. Construct an attention fusion computation model. Where ⊙ represents element-wise multiplication. For tensor outer product, α, β, γ, δ are attention weight coefficients. The constructed attention fusion calculation model is used to fuse temporal attention, spatial attention, and channel attention on the extracted multi-scale feature maps to obtain an attention-enhanced fused signal.

2. The attention-based chest and abdominal motion signal segmentation method as described in claim 1, characterized in that, Providing a segmentation network model involves the following steps: Based on the U-Net network architecture, the encoder part is defined as follows: E l =DownBlock l (E l-1 ), E0 = EE att DownBlock l Includes convolutional layers and downsampling: X l，i =Conv(BatchNorm(LeakyReLU(X l，,i-1 ))))(E l =MaxPool(X l,N ); The decoder part for skip connections and feature fusion is configured as follows: D l =UpBlock l (D l+1 E l UpBlock l Includes upsampling and feature concatenation: U l =ConvTransposee((D l+1 ), D l =Conv(BatchNorm(LeakyReLU(Concatenate(U l ,E l ))); Set the split output layer as: S = Softmax(Conv(D1)), Generate a 3D probability map:

3. The attention-based chest and abdominal motion signal segmentation method as described in claim 1, characterized in that, Feature extraction from the acquired physiological signals includes the following steps: Construct a multi-scale feature extraction module: Where F0 is the original three-dimensional feature map of distance, velocity, and angle, and ResBlock l For residual blocks, MaxPool l-1 For max pooling; The original three-dimensional feature maps of distance, velocity, and angle are obtained by calculation from the acquired physiological signals, using the following formula: The formula for calculating distance dimension data is as follows: f r The frequency of the difference frequency signal. Where β is the modulation frequency, T c For coherent processing time, distance resolution c is the speed of light, B is the bandwidth, and s is the speed of light. IF (t) represents the acquired physiological signal, expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let φ be the radial velocity of the nth target point. n Let n(t) be the initial phase, n(t) be Gaussian white noise, and λ be the wavelength. The formula for calculating velocity dimension data is as follows: Where S k (f r ) represents the range spectrum of the k-th frame, with a velocity resolution of . T s f is the frame interval. d The frequency is the Doppler frequency. The formula for calculating angular dimension data is as follows: Wherein, the angular resolution is M is the number of antennas, d is the antenna spacing, and S m θ is the range Doppler spectrum value output by the m-th antenna channel, θ is the azimuth angle of the target relative to the normal direction of the radar antenna array, and m is the antenna index.

4. The attention-based chest and abdominal motion signal segmentation method as described in claim 1, characterized in that, It also includes calculating the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signals and chest and abdominal motion signals, in order to quantify the complexity of the respiratory pattern; The formula for calculating the root mean square is: Where RR i The duration of the i-th respiratory cycle; The formula for calculating approximate entropy is: ApEn(m,r)=lim N→∞ [φ m (N,r)-φ m+1 [(N,r)], where φ m (N, r) represents the template matching probability.

5. A storage medium, characterized in that, The storage medium stores a program for an attention-based chest and abdominal motion signal segmentation method. When the program of the attention-based chest and abdominal motion signal segmentation method is executed by a processor, it implements the steps of the attention-based chest and abdominal motion signal segmentation method as described in any one of claims 1 to 4.

6. An attention-based chest and abdominal motion signal segmentation system, characterized in that, include: The acquisition unit is used to acquire physiological signals obtained by millimeter-wave radar from human detection. A feature extraction unit, connected to the acquisition unit, is used to extract features from the acquired physiological signals to obtain a multi-scale feature map. An attention fusion model, connected to the feature extraction unit, is used to fuse temporal attention, spatial attention, and channel attention on the extracted multi-scale feature maps to obtain an attention-enhanced fusion signal. A segmentation network model, connected to the attention fusion model, is used to segment the enhanced fusion signal to obtain respiratory motion signals and chest and abdominal motion signals; The attention fusion model includes: Temporal attention computation model, A t =Sigmoid(W t ·GRU(F t ), where GRU is a gated loop unit, W t Let F be the time-series feature weight matrix. t For multi-scale feature maps F l Temporal feature sequences after spatial dimension compression; Spatial attention computation model, A s =Sigmoid(W s ·Conv2D(F s )), where W s F represents the spatial convolution kernel parameters. s For multi-scale feature maps F l Spatial feature map aggregated along the time dimension; Channel attention calculation model, A c =Sigmoid(W c1 ·GlobalAvgPool(F c )+W c2 ·GlobalMaxPool(F c )), where W c1 W is the weight matrix of the global average pooling features. c2 F is the weight matrix of the global max pooling feature. c For multi-scale feature maps F l Channel features after global pooling in the spatiotemporal dimension; Spatiotemporal cross-attention computation model, Spatiotemporal correlation matrix, A st ∈RT×(H×W), where d k Q is the attention dimension normalization factor. st For the spatiotemporal query matrix, K st Let W be the spacetime bond matrix. q To query the projection weight matrix, W k H×W is the key projection weight matrix, where H×W is the spatial size, C is the number of feature channels, and T is the number of time frames. Attention fusion computation model Where ⊙ represents element-wise multiplication. For tensor outer product, α, β, γ, and δ are attention weight coefficients used to fuse temporal attention, spatial attention, and channel attention on the extracted multi-scale feature maps to obtain an attention-enhanced fused signal.

7. The attention-based chest and abdominal motion signal segmentation system as described in claim 6, characterized in that, The feature extraction unit includes a multi-scale feature extraction module, which is as follows: F l =ResBlock l (MaxPool l-1 (F l-1 ), l=1,2,...,L, where F0 is the original three-dimensional feature map of distance, velocity and angle, ResBlock l For residual blocks, MaxPool l-1 For max pooling; The original three-dimensional feature maps of distance, velocity, and angle are obtained by calculation from the acquired physiological signals, using the following formula: The formula for calculating distance dimension data is as follows: f r The frequency of the difference frequency signal. Where β is the modulation frequency, T c For coherent processing time, distance resolution c is the speed of light, B is the bandwidth, and s is the speed of light. IF (t) represents the acquired physiological signal, expressed as: Where N is the number of target points, A n R is the magnitude of the nth target point. n v is the target distance for the nth target point. n Let φ be the radial velocity of the nth target point. n Let n(t) be the initial phase, n(t) be Gaussian white noise, and λ be the wavelength. The formula for calculating velocity dimension data is as follows: Where S k (f r ) represents the range spectrum of the k-th frame, with a velocity resolution of . T s f is the frame interval. d The frequency is the Doppler frequency. The formula for calculating angular dimension data is as follows: Wherein, the angular resolution is M is the number of antennas, d is the antenna spacing, and S m θ is the range Doppler spectrum value output by the m-th antenna channel, θ is the azimuth angle of the target relative to the normal direction of the radar antenna array, and m is the antenna index.

8. The attention-based chest and abdominal motion signal segmentation system as described in claim 6, characterized in that, It also includes a quantization calculation unit, which is connected to the segmentation network model, and is used to calculate the root mean square and approximate entropy of the difference between adjacent respiratory cycles on the obtained respiratory motion signal and chest and abdominal motion signal, so as to quantify the complexity of the respiratory pattern. The formula for calculating the root mean square is: Where RR i The duration of the i-th respiratory cycle; The formula for calculating approximate entropy is: ApEn(m, r) = lim N→∞ [φ m (N, r)-φ m+1 [(N, r)], where φ m (N, r) represents the template matching probability.

Images