A bed exit early warning and vital sign monitoring system for bedridden patients based on millimeter wave radar

Abstract

This application discloses a bedridden patient exit warning and vital sign monitoring system based on millimeter-wave radar. The millimeter-wave radar front-end is responsible for emitting electromagnetic waves and collecting the original intermediate-frequency echo signal of the bedridden area; the signal processing module performs time-frequency transformation and clutter filtering on the original signal; the bedding feature identification module senses the bedding status of the clutter-filtered signal and outputs compensation parameters; the vital sign adaptive extraction module restores the signal based on the compensation parameters and extracts vital signs; the exit intention prediction module integrates point cloud data and vital signs to extract multi-dimensional features, predicts the exit intention through an artificial intelligence model, and triggers warnings in stages. Belonging to the field of millimeter-wave radar early warning and monitoring technology, this structure not only achieves high-precision vital sign monitoring in all seasons and under all working conditions, but also issues early warnings as early as when the patient sits up and leans forward.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave radar monitoring technology, specifically a system for early warning and vital sign monitoring of bedridden patients. Background Technology

[0002] In the field of medical monitoring, millimeter-wave radar-based monitoring technology for bedridden patients has attracted widespread attention due to its non-contact characteristics; however, existing millimeter-wave radar monitoring systems have two significant technical shortcomings.

[0003] 1. During vital sign monitoring, existing technologies typically treat bedding as pure interference and statically filter it out. However, existing technologies often treat bedding as static interference and are difficult to adapt to signal attenuation problems under different bedding conditions. When the patient is covered with a thick blanket, the electromagnetic wave penetration loss increases sharply, causing useful human micro-motion signals to be submerged in background noise, resulting in serious deviations or even complete failure of the respiratory rate and heart rate extracted by the system.

[0004] 2. In the process of alerting patients to leave the bed, existing technologies are generally based on boundary triggering, which has a lag in determining whether a patient has left the bed. This mechanism only triggers an alarm when the human point cloud crosses a preset boundary, resulting in a serious warning lag. When the alarm is triggered, the patient has often already gotten out of bed or is even on the verge of falling. At the same time, when a patient sits up in bed or reaches for something, the point cloud height changes and is very likely to cross the boundary, causing the system to frequently generate false alarms.

[0005] Therefore, there is an urgent need to develop a bedridden patient early warning and vital sign monitoring system based on millimeter-wave radar to solve the problems in the existing technology. Summary of the Invention

[0006] The purpose of this invention is to provide a bedridden patient early warning and vital sign monitoring system based on millimeter-wave radar. This system not only achieves high-precision vital sign monitoring in all seasons and working conditions, but also issues early warnings when the patient sits up and leans forward. Furthermore, it has a simple structure and is easy to use, thus solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A system for early warning and vital sign monitoring of bedridden patients based on millimeter-wave radar includes a signal processing module, a bedding feature identification module, a vital sign adaptive extraction module, and a pre-judgment module for intention to leave the bed. The signal processing module is used to receive the original intermediate frequency echo signal, perform a two-dimensional fast Fourier transform to generate a range Doppler matrix, filter out static background clutter, and output the intermediate echo signal. The bedding feature identification module is used to locate the shortest distance unit from the intermediate echo signal, extract the mean echo amplitude and phase variance, and combine the bedding attenuation mapping database to match and output the real-time attenuation coefficient and phase delay compensation amount. The adaptive vital signs extraction module is used to perform amplitude amplification compensation and phase shift correction on the intermediate echo signal of the target distance unit using the real-time attenuation coefficient and the phase delay compensation amount, dynamically adjust the passband range of the bandpass filter according to the real-time attenuation coefficient, and output the real-time respiratory rate and the real-time respiratory rate change rate. The bed-leaving intention prediction module is used to acquire point cloud data and the real-time respiratory rate change rate, extract body dynamic energy features, local micro-Doppler energy features and vital sign coupling features, and splice them to generate a bed-leaving intention spatiotemporal map matrix. The local micro-Doppler energy features are obtained by dividing the point cloud into head region, trunk region and leg region and calculating energy integral values ​​respectively. The bed-leaving intention spatiotemporal map matrix is ​​input into the bed-leaving intention recognition model to output state probability values. When the state probability value exceeds a preset threshold, an early warning signal is triggered.

[0008] By adopting the above technical solution, continuous and accurate vital sign monitoring is achieved in a dynamic bedding-covered environment. At the same time, based on multi-dimensional spatiotemporal mapping analysis, the core system function of predicting the patient's intention to leave the bed and triggering graded early warning is realized in advance.

[0009] As a further aspect of the present invention, the signal processing module performs the following steps to filter out static background clutter: determining and filtering out range cells in the range Doppler matrix whose amplitude remains constant and changes slowly.

[0010] By adopting the above technical solution, it is possible to accurately distinguish and eliminate fixed environmental interference in complex electromagnetic echoes, providing a low-level data cleaning function for subsequent dynamic human body and bedding analysis with a clean signal-to-noise environment.

[0011] As a further aspect of the present invention, the execution method of constructing the bedding attenuation mapping database within the bedding feature identification module includes: collecting the mean echo amplitude and phase variance of bedding under known conditions, and binding and storing the above features with the electromagnetic wave attenuation coefficient and phase delay compensation amount corresponding to the bedding.

[0012] By adopting the above technical solution, the function of establishing a priori correspondence between the physical characteristics of bedding and electromagnetic parameters is realized, laying a data comparison foundation for the system to dynamically perceive the current occlusion state and provide a basis for compensation.

[0013] As a further aspect of the present invention: the execution method of the adaptive extraction module for vital signs dynamically adjusting the passband range of the bandpass filter according to the real-time attenuation coefficient includes: when the real-time attenuation coefficient is greater than a preset attenuation threshold, narrowing the passband bandwidth of the bandpass filter and increasing the signal accumulation time.

[0014] By adopting the above technical solution, the function of intelligently balancing signal-to-noise ratio and temporal resolution for high-attenuation occlusion scenarios is realized, ensuring the accuracy of vital sign extraction under extremely harsh penetration conditions.

[0015] As a further aspect of the present invention, the execution method of the bed-leaving intention prediction module generating the bed-leaving intention spatiotemporal map matrix includes the following sub-steps: S41: Calculate the displacement trajectory vector of the three-dimensional spatial coordinates of the centroid of the human point cloud within a continuous time frame as the body kinetic energy feature; S42: Perform short-time Fourier transform on the head region, torso region and leg region respectively, and calculate the energy integral value within the preset frequency range as the local micro-Doppler energy feature; S43: Use the real-time respiratory rate change rate as a vital sign coupling feature; S44: The body dynamic energy characteristics, the local micro-Doppler energy characteristics, and the body sign coupling characteristics are spliced ​​together in time sequence to generate a spatiotemporal map matrix of the intention to leave the bed.

[0016] By adopting the above technical solution, the isolated macroscopic human movements and microscopic physiological parameters are transformed into a structured, multidimensional spatiotemporal sequence expression function, which comprehensively depicts the behavioral evolution process before leaving the bed.

[0017] As a further aspect of the present invention: the intention to leave the bed recognition model is constructed using a spatiotemporal graph convolutional network combined with a long short-term memory network. The spatiotemporal graph convolutional network is used to extract the spatial energy transfer relationship between the head region, trunk region and leg region, and the long short-term memory network is used to extract the evolution law of body motion energy features and physical sign coupling features in the time dimension.

[0018] By adopting the above technical solution, an advanced feature extraction function is achieved, which can accurately capture the energy transfer patterns in local body space and the behavioral and physiological evolution patterns in the time dimension through a dual-network collaborative mechanism.

[0019] As a further aspect of the present invention, the execution method of the early warning signal triggered by the bed leaving intention prediction module includes: issuing a first-level early warning signal when the output probability value of the bed leaving intention state exceeds a first preset threshold; and issuing a second-level highest-level early warning signal when the output probability value of the actual bed leaving state exceeds a second preset threshold.

[0020] By adopting the above technical solution, a differentiated and tiered intervention response function that matches the degree of risk of leaving the bed is realized, avoiding information redundancy or insufficient response caused by a single alarm threshold.

[0021] As a further aspect of the present invention, it also includes a millimeter-wave radar front-end, which is used to transmit frequency-modulated continuous waves and receive the original intermediate frequency echo signal of the bedridden area and transmit it to the signal processing module.

[0022] By adopting the above technical solution, a physical source sensing function that provides stable, high-frequency electromagnetic wave transmission and raw echo acquisition for the entire back-end processing link is realized.

[0023] As a further aspect of the present invention, the training process of the bed-leaving intention recognition model includes: dividing the labeled bed-leaving intention spatiotemporal map matrix into a training set, a validation set, and a test set in a ratio of 7:2:1, wherein the training set includes 350,000 bed-leaving intention spatiotemporal map matrices.

[0024] By adopting the above technical solution, the intention recognition model can be ensured to have strong generalization ability and extremely high fitting accuracy by using sufficient sample data divided in a scientific proportion.

[0025] As a further aspect of the present invention, the training process of the bed leave intention recognition model further includes: setting the initial learning rate of the model to 0.05, using the adaptive moment estimation (Adam) optimizer to perform gradient descent iteration on the model parameters, and stopping training and saving the model weights when the loss function value of the model on the validation set no longer decreases.

[0026] By adopting the above technical solution, the function of ensuring efficient convergence of the model training process and locking in the optimal prediction performance can be achieved by setting specific optimization parameters and early stopping mechanism.

[0027] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention solves the problem of nonlinear signal attenuation caused by bedding of different thicknesses by dynamically sensing the occlusion state and back-calculating the attenuation coefficient for signal compensation and adaptive filtering. It breaks through the limitation of treating bedding as static noise and realizes high-precision continuous monitoring of vital signs under all working conditions.

[0028] 2. This invention extracts the spatiotemporal map of local micro-motion energy transfer and physiological characteristics of the human body, and uses an artificial intelligence model to identify the patient's intention to leave the bed before the patient has left the bed. It abandons the lagging early warning mechanism that relies on spatial virtual fences and moves the early warning node forward significantly.

[0029] 3. This invention uses respiratory rhythm change rate to perform causal correlation analysis, divides the point cloud into three regions—head, trunk, and legs—to independently track micro-motion energy, accurately distinguishes between ordinary turning over, sitting up and leaning forward, and substantial movements of getting out of bed, fundamentally eliminating the high false alarm rate in the bedside environment.

[0030] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall structure of a bedridden patient exit warning and vital sign monitoring system based on millimeter-wave radar, as described in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] In this embodiment of the invention, a system for early warning and vital sign monitoring of bedridden patients based on millimeter-wave radar is described, see [link to relevant documentation]. Figure 1 As shown, it includes a millimeter-wave radar front end and sequentially connected signal processing module, bedding feature identification module, vital sign adaptive extraction module, and bed leaving intention prediction module; The millimeter-wave radar front end is responsible for transmitting electromagnetic waves and acquiring the raw intermediate frequency echo signal of the bedridden area; The signal processing module performs time-frequency transformation and noise filtering on the original signal; The bedding feature identification module senses the bedding status of the signal after clutter filtering and outputs compensation parameters; The adaptive vital sign extraction module restores signals and extracts vital signs based on compensation parameters; The bed-leaving intention prediction module integrates point cloud data and vital signs to extract multi-dimensional features, and uses an artificial intelligence model to predict bed-leaving intentions and trigger warnings in a tiered manner.

[0034] Example 1 This embodiment includes the following: The complete data flow process from electromagnetic wave emission to final triggering of the early warning includes: The millimeter-wave radar front-end adopts a frequency-modulated continuous wave system, with a set starting frequency of 77 GHz, a frequency modulation bandwidth of 4 GHz, and a frequency modulation period of 40 milliseconds. The millimeter-wave radar front-end continuously transmits the above-mentioned frequency-modulated continuous wave and receives the reflected signal from the bed rest area. After down-conversion processing, the original intermediate frequency echo signal is output and transmitted to the signal processing module.

[0035] The signal processing module receives the original intermediate frequency echo signal and performs range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform on the original intermediate frequency echo signal in sequence to generate a range-Doppler matrix; the horizontal axis of the range-Doppler matrix represents the Doppler frequency, and the vertical axis represents the range unit.

[0036] The signal processing module calculates the amplitude variance of each distance cell in the range-Doppler matrix over multiple consecutive frames, and determines that distance cells with amplitude variance less than a preset variance threshold of 0.01 and whose amplitude mean remains constant are static background clutter and filters them out.

[0037] After initial filtering, the signal processing module outputs an intermediate echo signal containing dynamic micro-motion information of the human body and bedding.

[0038] The bedding feature identification module receives the intermediate echo signal; in the distance dimension, the bedding feature identification module locates the shortest distance unit closest to the radar.

[0039] The bedding feature identification module extracts the mean echo amplitude and phase variance of the shortest distance unit within a preset time window.

[0040] The mean echo amplitude is calculated by summing all amplitude values ​​of the shortest distance cell in the Doppler dimension and then averaging them; the phase variance is calculated by subtracting the mean phase value from all phase values ​​of the shortest distance cell in the Doppler dimension, summing the squares, and then averaging them.

[0041] The bedding feature identification module has a bedding attenuation mapping database pre-stored inside. This database is established by collecting radar echoes under bedding of known thickness and material before deployment. The mean amplitude and phase variance of the echoes are bound and stored with the electromagnetic wave attenuation coefficient and phase delay compensation amount corresponding to the bedding.

[0042] The bedding feature identification module matches the real-time extracted mean echo amplitude and phase variance with the bedding attenuation mapping database, and outputs the real-time attenuation coefficient and phase delay compensation amount of the current bedding.

[0043] The adaptive vital signs extraction module simultaneously receives the intermediate echo signal, as well as the real-time attenuation coefficient and phase delay compensation. For the target distance unit where the human body is located, the adaptive vital signs extraction module uses the real-time attenuation coefficient to amplify and compensate the amplitude of the intermediate echo signal, and uses the phase delay compensation to perform reverse translation correction on the phase sequence of the intermediate echo signal, thereby restoring the true phase information of the human body's chest cavity displacement after penetrating the bedding.

[0044] The adaptive vital signs extraction module is equipped with a bandpass filter. When the real-time attenuation coefficient is determined to be greater than the preset attenuation threshold of 0.6, it indicates that the current bedding is too thick, resulting in a low signal-to-noise ratio. The adaptive vital signs extraction module narrows the passband bandwidth of the bandpass filter to 0.15 Hz to 0.35 Hz and increases the signal accumulation time to 4 seconds.

[0045] After adaptive filtering, the vital signs adaptive extraction module performs a spectral peak search on the phase sequence and outputs the real-time respiratory rate and the real-time respiratory rate change rate.

[0046] The point cloud data generated by the signal processing module after initial filtering and the real-time respiratory rate change rate output by the vital sign adaptive extraction module are acquired by the bed-leaving intention prediction module. The following sub-steps are then performed to generate the bed-leaving intention spatiotemporal map matrix: (1) The intention to leave the bed prediction module calculates the displacement trajectory vector of the three-dimensional spatial coordinates of the human body point cloud centroid in a continuous time frame as the body kinetic energy feature; the three-dimensional spatial coordinates of the point cloud centroid are calculated by summing the coordinate values ​​of all points in the human body point cloud in the current frame on the horizontal axis, vertical axis and height axis respectively and then dividing by the total number of points in the point cloud. (2) The bed-leaving intention prediction module divides the point cloud longitudinally from head to toe into head region, torso region and leg region according to the normal lying posture of the human body; the bed-leaving intention prediction module performs short-time Fourier transform on the above three regions respectively, sets the window length to 256 sampling points and the step size to 128 sampling points, and calculates the energy integral value of the above three regions in the preset frequency range of 0.1 Hz to 0.5 Hz as the local micro-Doppler energy feature; (3) The intention to leave the bed prediction module directly uses the real-time respiratory rate change rate output by the vital sign adaptive extraction module as the vital sign coupling feature; (4) The bed leaving intention prediction module splices together the body motion energy characteristics, local micro Doppler energy characteristics and physical sign coupling characteristics in chronological order to generate a spatiotemporal map matrix of bed leaving intention.

[0047] The bed-leaving intention prediction module is equipped with a pre-trained bed-leaving intention recognition model; this bed-leaving intention recognition model is constructed using a spatiotemporal graph convolutional network combined with a long short-term memory network.

[0048] Spatiotemporal graph convolutional networks are used to extract the spatial energy transfer relationships between the head region, torso region, and leg region. Long Short-Term Memory (LSTM) networks are used to extract the evolution of body dynamic energy features and physical characteristic coupling features over time. The bed-leaving intention prediction module inputs the spatiotemporal map matrix of bed-leaving intention into the bed-leaving intention recognition model. The model outputs state probability values ​​for four categories: sound sleep state, sleeping posture adjustment state, bed-leaving intention state, and actual bed-leaving state.

[0049] When the probability value of the intention to leave the bed output by the model exceeds the first preset threshold of 0.75, the system issues a first-level warning signal.

[0050] When the probability value of the substantial bed-off state output by the model exceeds the second preset threshold of 0.90, the system issues a level-two maximum warning signal.

[0051] Example 2 The technical feature that distinguishes this embodiment from Embodiment 1 is that this embodiment also includes the specific training process of the bed-leaving intention recognition model in Embodiment 1.

[0052] In the data collection and annotation stage, radar signals of multiple subjects under different bedding conditions were collected in a real ward environment, and a spatiotemporal map matrix of intention to leave the bed was generated through four steps (1) to (4).

[0053] Medical staff manually labeled each frame of the spatiotemporal map matrix of intention to get out of bed through synchronous video monitoring. The labels were strictly limited to four categories: sound sleep state, adjusting sleeping position state, intention to get out of bed state, and actual getting out of bed state.

[0054] During the dataset partitioning phase, all labeled spatiotemporal map matrices of bed leaving intentions were divided into training, validation, and test sets in a ratio of 7:2:1.

[0055] The training set includes a spatiotemporal map matrix of 350,000 intentions to leave the bed, used for model training; the validation set includes a spatiotemporal map matrix of 100,000 intentions to leave the bed, used for adjusting model parameters; and the test set includes a spatiotemporal map matrix of 50,000 intentions to leave the bed, used for evaluating the final performance of the model.

[0056] During the model parameter configuration and iterative training phase, the initial learning rate of the bed leave intention recognition model was set to 0.05. The Adam optimizer was used to perform gradient descent iteration on the model parameters. The Adam optimizer generates an adaptive learning rate for different parameters by calculating the first moment estimate and second moment estimate of the gradient.

[0057] During model training, after each training round, the system inputs the validation set into the model to calculate the cross-entropy loss function value. When the cross-entropy loss function value on the validation set no longer decreases within 10 consecutive training rounds, the system triggers the early stop mechanism to stop training and saves the model weights of the current round as the final deployed bed leave intention recognition model.

[0058] Through the above training process, the model can accurately capture the temporal evolution of abnormal physical signs to local micro-motion energy transfer.

[0059] This invention provides a bedridden patient early warning and vital sign monitoring system based on millimeter-wave radar. It not only achieves high-precision vital sign monitoring in all seasons and working conditions, but also issues early warnings when the patient sits up and leans forward, with high reliability.

[0060] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0061] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A system for early warning and vital sign monitoring of bedridden patients based on millimeter-wave radar, characterized in that, It includes a signal processing module, a bedding feature recognition module, a vital sign adaptive extraction module, and an intention to get out of bed prediction module; The signal processing module is used to receive the original intermediate frequency echo signal, perform a two-dimensional fast Fourier transform to generate a range Doppler matrix, filter out static background clutter, and output the intermediate echo signal. The bedding feature identification module is used to locate the shortest distance unit from the intermediate echo signal, extract the mean echo amplitude and phase variance, and combine the bedding attenuation mapping database to match and output the real-time attenuation coefficient and phase delay compensation amount. The adaptive vital signs extraction module is used to perform amplitude amplification compensation and phase shift correction on the intermediate echo signal of the target distance unit using the real-time attenuation coefficient and the phase delay compensation amount, dynamically adjust the passband range of the bandpass filter according to the real-time attenuation coefficient, and output the real-time respiratory rate and the real-time respiratory rate change rate. The bed-leaving intention prediction module is used to acquire point cloud data and the real-time respiratory rate change rate, extract body dynamic energy features, local micro-Doppler energy features and vital sign coupling features, and splice them to generate a bed-leaving intention spatiotemporal map matrix. The local micro-Doppler energy features are obtained by dividing the point cloud into head region, trunk region and leg region and calculating energy integral values ​​respectively. The bed-leaving intention spatiotemporal map matrix is ​​input into the bed-leaving intention recognition model to output state probability values. When the state probability value exceeds a preset threshold, an early warning signal is triggered.

2. The bedridden patient exit warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The signal processing module filters out static background clutter by identifying and filtering out range cells in the range Doppler matrix whose amplitude remains constant and changes slowly.

3. The bedridden patient exit warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The execution method of constructing the bedding attenuation mapping database within the bedding feature identification module includes: collecting the mean echo amplitude and phase variance of bedding under known conditions, and binding and storing the above features with the electromagnetic wave attenuation coefficient and phase delay compensation amount corresponding to the bedding.

4. The bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The execution method of the adaptive extraction module for vital signs dynamically adjusting the passband range of the bandpass filter according to the real-time attenuation coefficient includes: when the real-time attenuation coefficient is greater than a preset attenuation threshold, narrowing the passband bandwidth of the bandpass filter and increasing the signal accumulation time.

5. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The execution method of generating the spatiotemporal map matrix of the intention to leave the bed includes the following sub-steps: S41: Calculate the displacement trajectory vector of the three-dimensional spatial coordinates of the centroid of the human point cloud within a continuous time frame as the body kinetic energy feature; S42: Perform short-time Fourier transform on the head region, torso region and leg region respectively, and calculate the energy integral value within the preset frequency range as the local micro-Doppler energy feature; S43: Use the real-time respiratory rate change rate as a vital sign coupling feature; S44: The body dynamic energy characteristics, the local micro-Doppler energy characteristics, and the body sign coupling characteristics are spliced ​​together in time sequence to generate a spatiotemporal map matrix of the intention to leave the bed.

6. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The intention to leave the bed model is constructed using a spatiotemporal graph convolutional network combined with a long short-term memory network. The spatiotemporal graph convolutional network is used to extract the spatial energy transfer relationship between the head region, torso region and leg region, and the long short-term memory network is used to extract the evolution law of body motion energy features and physical sign coupling features in the time dimension.

7. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, The execution mode of the warning signal triggered by the bed leaving intention prediction module includes: issuing a first-level warning signal when the output probability value of the bed leaving intention state exceeds a first preset threshold; and issuing a second-level highest warning signal when the output probability value of the actual bed leaving state exceeds a second preset threshold.

8. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 1, characterized in that, It also includes a millimeter-wave radar front-end, which is used to transmit frequency-modulated continuous waves and receive the original intermediate frequency echo signal from the bedridden area and transmit it to the signal processing module.

9. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 6, characterized in that, The training process of the bed-leaving intention recognition model includes: dividing the labeled bed-leaving intention spatiotemporal map matrix into a training set, a validation set, and a test set in a ratio of 7:2:1, wherein the training set includes 350,000 bed-leaving intention spatiotemporal map matrices.

10. A bedridden patient exit-of-bed early warning and vital sign monitoring system based on millimeter-wave radar according to claim 9, characterized in that, The training process of the bed leave intention recognition model also includes: setting the initial learning rate of the model to 0.05, using the Adam optimizer to perform gradient descent iteration on the model parameters, and stopping training and saving the model weights when the loss function value of the model on the validation set no longer decreases.

Images