CN-120899237-B - Attention-based chest and abdomen movement signal segmentation method, system and storage medium

CN120899237BCN 120899237 BCN120899237 BCN 120899237BCN-120899237-B

Abstract

The invention relates to the technical field of biomedical radar signal processing and artificial intelligence intersection, in particular to a chest and abdomen movement signal segmentation method, a system and a storage medium based on attention, wherein the segmentation method comprises the following steps of obtaining physiological signals obtained by detecting a human body by a millimeter wave radar; the method comprises the steps of obtaining a physiological signal, carrying out feature extraction on the obtained physiological signal to obtain a multi-scale feature map, carrying out fusion of time domain attention, space domain attention and channel attention on the multi-scale feature map by using a constructed attention fusion model to obtain a fusion signal based on attention enhancement, inputting the fusion signal based on attention enhancement into a segmentation network model, and further obtaining a respiratory motion signal and a chest-abdomen motion signal. According to the invention, the FMCW millimeter wave radar signal is processed by using a deep learning model, the relevant characteristics of chest and abdomen movement are enhanced by a attention mechanism, the chest and abdomen movement signal is accurately segmented, and reliable input is provided for subsequent respiratory frequency and heart rate calculation.

Inventors

ZHU NAN
LI WEN
Tu Caigang
FANG TAO

Assignees

上海淘景立画信息技术有限公司
上海镭达晶元智能科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251009

Claims (8)

1. The chest and abdomen movement signal segmentation method based on the attention is characterized by comprising the following steps of: acquiring physiological signals obtained by detecting a human body by a millimeter wave radar; extracting features of the acquired physiological signals to obtain a multi-scale feature map; Constructing an attention fusion model, and fusing the time domain attention, the space domain attention and the channel attention of the extracted multi-scale feature map by utilizing the constructed attention fusion model so as to obtain a fusion signal based on attention enhancement; providing a segmentation network model, and inputting the fusion signal based on attention enhancement into the provided segmentation network model so as to obtain a respiratory motion signal and a chest and abdomen motion signal output by the segmentation network model; the construction of the attention fusion model comprises the following steps: Constructing a time domain attention calculation model, A t ＝Sigmoid(W t ·GRU(F t )), wherein GRU is a gating circulation unit, W t is a time sequence feature weight matrix, and F t is a time sequence feature sequence of the multi-scale feature map F l after space dimension compression; Constructing a airspace attention calculation model, A s ＝Sigmoid(W s ·Conv2D(F s )), wherein W s is a space convolution kernel parameter, F s is a space feature map of the multi-scale feature map F l after time dimension aggregation; A channel attention computation model is constructed, A c ＝Sigmaoid(W c1 ·GobalAvgPool(F c )+W c2 ·GlobalMaxPool(F c )), wherein W c1 is a weight matrix of global average pooling features, W c2 is a weight matrix of global maximum pooling features, and F c is channel features of the multi-scale feature map F l after space-time dimension global pooling; a space-time cross-attention calculation model is constructed, Constructing a space-time correlation matrix, A st ∈R T×(H×W) , wherein d k is a attention dimension normalization factor, Q st is a space-time query matrix, K st is a space-time key matrix, W q is a query projection weight matrix, W k is a key projection weight matrix, H×W is a space size, C is a characteristic channel number, and T is a time frame number; an attention fusion calculation model is constructed, Wherein, as the element level multiplication, And (3) taking tensor outer products, alpha, beta, gamma and delta as attention weight coefficients, and fusing the time domain attention, the space domain attention and the channel attention of the extracted multi-scale feature map by using the constructed attention fusion calculation model to obtain a fusion signal based on attention enhancement.
2. The attention-based chest and abdomen motion signal segmentation method according to claim 1, wherein providing the segmentation network model comprises the steps of: Based on the U-Net network architecture, the encoder part is set as follows: e l ＝DownBlock l (E l-1 ),E 0 ＝EE att , wherein DownBlock l comprises a convolutional layer and downsampling: X l,i ＝Conv(BatchNorm(LeakyReLU(X l,,i-1 ))))(E l ＝MaxPool(X l,N ); the decoder part setting the jump connection and feature fusion is: d l ＝UpBlock l (D l+1 ,E l ), wherein UpBlock l comprises upsampling and feature stitching: U l ＝ConvTransposee((D l+1 ), D l ＝Conv(BatchNorm(LeakyReLU(Concatenate(U l ,E l ))); The split output layer is set to s=softmax (Conv (D 1 )), Generating a three-dimensional probability map:
3. The attention-based chest and abdomen movement signal segmentation method according to claim 1, wherein the feature extraction of the acquired physiological signal comprises the steps of: constructing a multi-scale feature extraction module: Wherein, F 0 is the original three-dimensional characteristic diagram of distance, speed and angle, resBlock l is the residual block, maxPool l-1 is the maximum pooling; the original distance, speed and angle three-dimensional feature map is obtained by calculating the acquired physiological signals, and the calculation formula is as follows: the calculation formula of the distance dimension data is as follows F r is the frequency of the difference frequency signal, Wherein, beta is the modulation frequency, T c is the coherent processing time and the distance resolution C is the speed of light, B is the bandwidth, s IF (t) is the acquired physiological signal, and the expression is: Wherein N is the number of target points, A n is the amplitude of the nth target point, R n is the target distance of the nth target point, v n is the radial speed of the nth target point, phi n is the initial phase, N (t) is Gaussian white noise, and lambda is the wavelength; the calculation formula of the speed dimension data is as follows Wherein S k (f r ) is the kth frame distance spectrum, the speed resolution is T s is the frame interval, f d is the doppler frequency; the calculation formula of the angle dimension data is as follows Wherein the angular resolution is M is the number of antennas, d is the antenna spacing, S m is the range Doppler spectrum value output by the mth 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 abdomen motion signal segmentation method according to claim 1, further comprising calculating root mean square and approximate entropy of adjacent respiratory cycle differences for the obtained respiratory motion signal and chest and abdomen motion signal to quantify the respiratory pattern complexity; the root mean square is calculated as: Wherein RR i is the duration of the ith respiratory cycle; the calculation formula of the approximate entropy is as follows: ApEn (m, r) =lim N→∞ [φ m (N,r)-φ m+1 (N, r) ], where Φ m (N, r) is the template matching probability.
5. A storage medium, characterized in that the storage medium stores a program of a chest and abdomen movement signal segmentation method based on attention, The program of the attention-based chest and abdomen motion signal segmentation method, when executed by a processor, implements the steps of the attention-based chest and abdomen motion signal segmentation method according to any one of claims 1 to 4.
6. An attention-based chest and abdomen movement signal segmentation system, comprising: the acquisition unit is used for acquiring physiological signals obtained by detecting the human body by the millimeter wave radar; The feature extraction unit is connected with the acquisition unit and is used for carrying out feature extraction on the acquired physiological signals so as to obtain a multi-scale feature map; The attention fusion model is connected with the feature extraction unit and is used for fusing the time domain attention, the space domain attention and the channel attention of the extracted multi-scale feature map so as to obtain a fusion signal based on attention enhancement; The segmentation network model is connected with the attention fusion model and is used for segmenting the enhanced fusion signal to obtain a respiratory motion signal and a chest-abdomen motion signal; The attention fusion model includes: A time domain attention calculation model, a t ＝Sigmoid(W t ·GRU(F t )), wherein the GRU is a gating cyclic unit, W t is a time sequence feature weight matrix, and F t is a time sequence feature sequence of the multi-scale feature map F l after space dimension compression; Airspace attention calculation model, a s ＝Sigmoid(W s ·Conv2D(F s )), wherein W s is a spatial convolution kernel parameter, and F s is a spatial feature map of the multi-scale feature map F l after time dimension aggregation; a channel attention calculation model is provided, A c ＝Sigmoid(W c1 ·GlobalAvgPool(F c )+W c2 ·GlobalMaxPool(F c )), wherein W c1 is a weight matrix of global average pooling features, W c2 is a weight matrix of global maximum pooling features, and F c is channel features of the multi-scale feature map F l after space-time dimension global pooling; A space-time cross-attention computation model, A space-time correlation matrix, A st epsilon RT× (H×W), wherein d k is a attention dimension normalization factor, Q st is a space-time query matrix, K st is a space-time key matrix, W q is a query projection weight matrix, W k is a key projection weight matrix, H×W is a space size, C is a feature channel number, and T is a time frame number; the attention is fused to the computational model, Wherein, as the element level multiplication, The tensor outer product, alpha, beta, gamma and delta are attention weight coefficients, and are used for fusing the time domain attention, the space domain attention and the channel attention of the extracted multi-scale feature map to obtain a fused signal based on attention enhancement.
7. The attention-based chest and abdomen motion signal segmentation system according to claim 6, wherein the feature extraction unit comprises a multi-scale feature extraction module, the multi-scale feature extraction module being: F l ＝ResBlock l (MaxPool l-1 (F l-1 ), l=1, 2, a.c., L, wherein, F 0 is an original distance, velocity and angle three-dimensional feature map, resBlock l is a residual block, maxPool l-1 is max pooling; the original distance, speed and angle three-dimensional feature map is obtained by calculating the acquired physiological signals, and the calculation formula is as follows: the calculation formula of the distance dimension data is as follows F r is the frequency of the difference frequency signal, Wherein, beta is the modulation frequency, T c is the coherent processing time and the distance resolution C is the speed of light, B is the bandwidth, s IF (t) is the acquired physiological signal, and the expression is: Wherein N is the number of target points, A n is the amplitude of the nth target point, R n is the target distance of the nth target point, v n is the radial speed of the nth target point, phi n is the initial phase, N (t) is Gaussian white noise, and lambda is the wavelength; the calculation formula of the speed dimension data is as follows Wherein S k (f r ) is the kth frame distance spectrum, the speed resolution is T s is the frame interval, f d is the doppler frequency; the calculation formula of the angle dimension data is as follows Wherein the angular resolution is M is the number of antennas, d is the antenna spacing, S m is the range Doppler spectrum value output by the mth 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 abdomen motion signal segmentation system according to claim 6, further comprising a quantization calculation unit coupled to the segmentation network model for calculating a root mean square and an approximate entropy of the adjacent respiratory cycle differences for the resulting respiratory motion signal and chest and abdomen motion signal to quantify respiratory pattern complexity; the root mean square is calculated as: Wherein RR i is the duration of the ith respiratory cycle; the calculation formula of the approximate entropy is as follows: ApEn (m, r) =lim N→∞ [φ m (N,r)-φ m+1 (N, r) ], where Φ m (N, r) is the template matching probability.

Description

Attention-based chest and abdomen movement signal segmentation method, system and storage medium Technical Field The invention relates to the technical field of biomedical radar signal processing and artificial intelligence intersection, in particular to a chest and abdomen movement signal segmentation method and system based on attention and a storage medium. Background Respiratory rate and heart rate are vital signs of the human body, and real-time monitoring is of great importance for health management and early warning of diseases. Conventional detection methods (e.g., chest straps, optical sensors, etc.) typically require direct contact with the human body, which not only affects the comfort of the user, but is also limited by environmental conditions. FMCW millimeter wave radar, as a contactless, non-invasive monitoring means, is capable of capturing small displacements of the body surface (such as chest and abdomen movements caused by respiration or heartbeat) by transmitting and receiving millimeter wave signals, and has great potential in the field of vital sign monitoring. However, since radar signals are susceptible to background noise and multipath effects, how to accurately separate chest and abdomen motion signals from complex millimeter wave signals becomes a difficulty in the prior art. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides a chest and abdomen movement signal segmentation method, a system and a storage medium based on attention, which solve the problem of how to accurately separate chest and abdomen movement signals from millimeter wave signals. The technical scheme for achieving the purpose is as follows: The invention provides a chest and abdomen movement signal segmentation method based on attention, which comprises the following steps: acquiring physiological signals obtained by detecting a human body by a millimeter wave radar; extracting features of the acquired physiological signals to obtain a multi-scale feature map; Constructing an attention fusion model, and fusing the time domain attention, the space domain attention and the channel attention of the extracted multi-scale feature map by utilizing the constructed attention fusion model so as to obtain a fusion signal based on attention enhancement; providing a segmentation network model, inputting the fusion signal based on attention enhancement into the provided segmentation network model, and further obtaining a respiratory motion signal and a chest and abdomen motion signal output by the segmentation network model. The chest and abdomen movement signal segmentation method based on attention is further improved in that the construction of the attention fusion model comprises the following steps: a time-domain attention computation model is constructed, Wherein the GRU is a gating circulating unit, W t is a time sequence feature weight matrix, and F t is a time sequence feature sequence of the multi-scale feature map F l after space dimension compression; a spatial domain attention calculation model is constructed, Wherein W s is a space convolution kernel parameter, F s is a space feature map of the multi-scale feature map F l after time dimension aggregation; a channel attention computation model is constructed, Wherein W c1 is the weight matrix of the global average pooling feature, W c2 is the weight matrix of the global maximum pooling feature, and F c is the channel feature of the multi-scale feature map F l after the space-time dimension global pooling; a space-time cross-attention calculation model is constructed, , A space-time correlation matrix is constructed,Wherein d k is the attention dimension normalization factor, Q st is the space-time query matrix, K st is the space-time key matrix, W q is the query projection weight matrix, W k is the key projection weight matrix, H×W is the spatial dimension, C is the number of characteristic channels, and T is the number of time frames; an attention fusion calculation model is constructed, , wherein,For the element-level multiplication to be performed,In order to be the tensor outer product,And for attention weight coefficients, performing fusion of time domain attention, space domain attention and channel attention on the extracted multi-scale feature map by using the constructed attention fusion calculation model so as to obtain a fusion signal based on attention enhancement. The chest and abdomen motion signal segmentation method based on the attention is further improved in that the segmentation network model comprises the following steps: Based on the U-Net network architecture, the encoder part is set as follows: DownBlock l of which contains convolutional layers and downsampling: ; the decoder part setting the jump connection and feature fusion is: wherein UpBlock l comprises upsampling and feature stitching: , ; Setting the split output layer as: , Generating a three-dimensional probability map: 。 The c