Mattress type non-inductive sleep detection method integrating millimeter wave radar and pressure sensing

Abstract

The invention relates to the technical field of sleep detection and analysis, in particular to a millimeter-wave radar and pressure sensing integrated mattress type non-inductive sleep detection method, which comprises the following steps of: synchronously acquiring radar echoes and body pressure distribution signals of a user through the combination of a millimeter-wave radar and a pressure sensor, and providing a data source for subsequent analysis; radar and pressure signals are respectively processed, initial vital sign signals, micro-motion tracks, pressure center tracks and pressure fluctuation signals are effectively extracted, and the analysis capability of physiological and motion characteristics is enhanced; a signal processing strategy is dynamically selected according to the body motion fusion confidence coefficient, and optimal data processing in different physiological states is ensured; through track coding, clustering and body movement grammar library matching, macroscopic body movement such as turning over and sitting up is recognized, and the sleep behavior analysis dimension is enriched; by constructing a sleep depth potential energy field and combining physiological parameters and body movement types, sleep state evolution is simulated, and dynamic judgment of the sleep stage is achieved.

Description

Technical Field

[0001] This invention relates to the field of sleep detection and analysis technology, and in particular to a mattress-based, non-invasive sleep detection method that integrates millimeter-wave radar and pressure sensing. Background Technology

[0002] Sleep quality assessment is crucial for health management, and accurate sleep monitoring is the core foundation. However, the current field of sleep monitoring still faces several challenges, including: traditional sleep monitoring devices often rely on a single type of sensor, such as pressure sensors or accelerometers, resulting in incomplete data collection and an inability to continuously monitor the user's sleep state, especially in capturing subtle physiological changes and obvious body movement behaviors simultaneously; existing technologies often employ separate analysis methods when processing body movement and physiological signals, lacking effective multimodal signal fusion mechanisms, which limits the system's accurate interpretation of complex sleep behaviors, especially the comprehensive judgment when body movement and physiological changes interact; when faced with frequent body movements or rapid transitions between sleep stages, traditional systems struggle to adjust signal processing parameters and discrimination strategies in real time, leading to unstable monitoring data and affecting the accuracy of sleep quality assessment; most sleep monitoring solutions focus on monitoring microscopic physiological indicators, neglecting the identification and analysis of body movement behaviors, limiting the depth and breadth of sleep behavior pattern mining; and traditional methods may use static or simplified sleep stage division standards, failing to accurately simulate the continuous evolution of sleep states over time. To address this, the present invention proposes a mattress-based, non-invasive sleep detection method that integrates millimeter-wave radar and pressure sensing. Summary of the Invention

[0003] The purpose of this invention is to solve the problems in the background art by proposing a mattress-type non-invasive sleep detection method that integrates millimeter-wave radar and pressure sensing.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A mattress-based, non-invasive sleep detection method integrating millimeter-wave radar and pressure sensing includes: S1. Simultaneously collect the user's radar echo signal and body pressure distribution signal through a millimeter-wave radar sensor array and a distributed thin-film pressure sensor array laid inside the mattress; S2. Process the radar echo signal to extract the initial vital signs signal and micro-motion trajectory; at the same time, process the body pressure distribution signal to extract the pressure center trajectory and pressure fluctuation signal. S3. Based on the micro-motion trajectory and the pressure center trajectory, establish a trajectory consistency fusion processing mechanism and calculate the body motion fusion confidence level; S4. Based on the confidence level of body movement fusion, adaptively select and execute different signal processing pathways, and obtain fused vital sign signals and macroscopic body movement types; among which, different signal processing pathways include frequency domain signal fusion pathways and body movement analysis pathways; S5. Based on the fusion of vital signs signals and macroscopic body movement types, a sleep depth potential energy field is established; by simulating the evolution process of sleep state driven by dynamic equations in this potential energy field, the user's sleep stage is dynamically determined.

[0005] Furthermore, in step S2, the process of processing the radar echo signal to extract the initial vital signs signal and micro-motion trajectory includes: S21a. Preprocess the multi-channel radar echo signals synchronously acquired by the millimeter-wave radar sensor array; S22a. Perform joint spatial-temporal processing on the preprocessed multi-channel radar echo signal and output the signal after spatial enhancement. S23a. For the signal after spatial enhancement, perform a fast Fourier transform in the slow time dimension to generate a high signal-to-noise ratio range-Doppler spectrum and construct a range-Doppler-time three-dimensional data cube. S24a. Combining the range-Doppler-time three-dimensional data cube, using an energy entropy-based region of interest automatic detection algorithm, the range-Doppler cell corresponding to the human chest cavity is located, and the phase history sequence of the signal within the cell is extracted; the phase history sequence is unwrapped, and the unwrapped phase signal is bandpass filtered to separate the initial vital signs signal. S25a. Perform dynamic target detection on the three-dimensional data cube to identify dynamic point clouds other than the main target of the human torso; use the trajectory association method to find the nearest point cloud in the dynamic point cloud set of adjacent time moments for each detected dynamic point cloud and associate them to form a trajectory segment; and form a micro-motion trajectory by splicing and smoothing the trajectory segments of multiple time moments.

[0006] Furthermore, in step S2, the process of processing the body pressure distribution signal and extracting the pressure center trajectory and pressure fluctuation signals includes: S21b. For each frame of body pressure distribution signal acquired by the distributed thin-film pressure sensor array, construct a two-dimensional pressure matrix, where each element of the matrix corresponds to the pressure value of a sensor unit; calculate the centroid coordinates of the two-dimensional pressure matrix. ; S22b. Calculate the centroid coordinates of each frame of body pressure distribution signal in chronological order. Connect them sequentially to form a continuous pressure center trajectory; S23b. In each frame of the body pressure distribution signal, using the centroid coordinates of the current frame... Based on this, a rectangular region of interest corresponding to the user's chest cavity is automatically defined. The pressure values ​​of all sensor units within this rectangular region of interest are summed to obtain the total pressure sequence for that region. The total pressure sequence is then bandpass filtered to extract the pressure fluctuation signals caused by respiratory activity.

[0007] Furthermore, in step S3, based on the micro-motion trajectory and the pressure center trajectory, a trajectory consistency fusion processing mechanism is established, and the process of calculating the body motion fusion confidence score includes: S31. Calculate the quality indices of the micro-motion trajectory and the pressure center trajectory respectively: the quality index of the micro-motion trajectory is its average signal-to-noise ratio within the sliding time window; the quality index of the pressure center trajectory is its trajectory smoothness within the sliding time window. S32. Set the first threshold T1 as the lowest reliable value of the average signal-to-noise ratio of the micro-motion trajectory, the second threshold T2 as the highest reliable value of the trajectory smoothness of the pressure center trajectory, the third threshold T3 as the lowest reliable value of the trajectory smoothness of the pressure center trajectory, and the fourth threshold T4 as the highest reliable value of the average signal-to-noise ratio of the micro-motion trajectory. S33. Determine the dominant reliable trajectory based on the quality indicators of the micro-motion trajectory and the pressure center trajectory: If the average signal-to-noise ratio of the micro-motion trajectory is lower than the first threshold and the trajectory smoothness of the pressure center trajectory is higher than the second threshold, then the pressure center trajectory is determined to be the dominant reliable trajectory; if the trajectory smoothness of the pressure center trajectory is lower than the third threshold and the average signal-to-noise ratio of the micro-motion trajectory is higher than the fourth threshold, then the micro-motion trajectory is determined to be the dominant reliable trajectory. S34. Based on the determination results of the dominant reliable trajectory, different strategies are used to generate the confidence level of body motion fusion.

[0008] Furthermore, in step S34, the process of generating body motion fusion confidence scores using different strategies based on the determination result of the dominant reliable trajectory includes: When the pressure center trajectory is determined to be dominant, the body motion fusion confidence is obtained based on the overall displacement of the trajectory within the determination time window. The overall displacement of the trajectory is the weighted sum of the Euclidean distance between the trajectory start and end points and the sum of the magnitudes of all displacement vectors within the time window. When the micro-motion trajectory is determined to be dominant, the confidence level of body motion fusion is calculated based on the product of the spatial distribution dispersion of all dynamic points of the trajectory within the determination time window and the average motion velocity.

[0009] When a single dominant trajectory cannot be determined, the confidence level of body-motion fusion is obtained by calculating the geometric mean of the cosine similarity between the micro-motion trajectory and the pressure center trajectory in the displacement direction and the Pearson correlation coefficient in the motion velocity.

[0010] Furthermore, step S4 also includes: if the confidence level of body movement fusion is lower than the preset pathway switching threshold, the user is determined to be in the static-dominant stage, and the frequency domain signal fusion pathway is activated to perform frequency domain fusion of the initial vital sign signal and the pressure fluctuation signal to generate a fused vital sign signal; if the confidence level of body movement fusion is higher than or equal to the preset pathway switching threshold, the user is determined to be in the body movement-dominant stage, and the body movement analysis pathway is activated to identify the macroscopic body movement type based on the micro-motion trajectory and the pressure center trajectory.

[0011] Furthermore, the process of frequency domain fusion of the initial vital signs signal and the pressure fluctuation signal to generate a fused vital signs signal includes: Short-time Fourier transforms were performed on the initial vital signs signal and the pressure fluctuation signal respectively to obtain the time spectrum of the initial vital signs signal, i.e., the radar source time spectrum, and the time spectrum of the pressure fluctuation signal, i.e., the pressure source time spectrum. Define adaptive fusion weights and use them to perform frequency-point weighted fusion of the radar source time spectrum and the pressure source time spectrum to generate a fused time spectrum; The fused time-domain fusion vital signs signal is reconstructed by performing an inverse short-time Fourier transform on the fused time-domain spectrum.

[0012] Furthermore, the process of identifying macroscopic body movement types includes: In the body motion analysis pathway, the micro-motion trajectory and the pressure center trajectory are synchronously divided into multiple consecutive short-time windows on the time axis with the same starting point and duration; The trajectory data within each short time window is encoded to generate a fused trajectory segment descriptor, which includes the direction of movement of the pressure center within the segment, the spatial distribution entropy of the micro-movement points, and the mutual information of the velocity changes of the two trajectories. An online clustering algorithm is used to dynamically cluster trajectory segment descriptors generated in all time windows. The online clustering algorithm adopts the CluStream streaming data clustering algorithm, which maintains data summaries online through micro-cluster structures and periodically generates macro-clusters to group trajectory segments with the same motion mode into the same cluster. The cluster center of each macro-cluster is defined as a behavioral primitive to represent a cross-modal motion mode. The chronological sequence of action primitives is matched with a predefined body movement grammar library; the body movement grammar library consists of a set of automata, each of which corresponds to a macroscopic body movement type, and its state transitions are triggered by a predefined sequence of action primitives. The currently generated sequence of behavioral primitives is used as the input string and fed in parallel into the automata corresponding to all macroscopic body movement types. When an automaton reaches its terminal state after undergoing a state transition from the initial state, the macroscopic body movement type corresponding to that automaton is determined and the recognition result is output.

[0013] Furthermore, in step S5, the process of dynamically determining the user's sleep stage by establishing a sleep depth potential field includes: Define a one-dimensional potential field for sleep depth. ; At each processing time Based on the physiological parameters parsed from fused vital sign signals and the macroscopic body movement types identified by the body movement analysis pathway, the net force driving the evolution of sleep state in the sleep depth potential field is calculated and analyzed. ; Based on net force The sleep depth state is obtained by evolving the dynamic equations of simulated inertia and damping. ; The sleep depth state is obtained by solving the dynamic equations through numerical integration. The evolutionary trajectory; The sleep stage determination is based on the currently locked stable potential well; changes in sleep stage are modeled as sleep depth states. The dynamic process of transitions between different stable potential wells.

[0014] Furthermore, in the defined sleep depth potential field In the diagram, DS represents the sleep depth coordinates; potential energy field It consists of multiple predefined stable potential wells stacked together, each potential well corresponding to a sleep stage and defined by its center position and width parameters: for a wakefulness potential well, the center is located at... Width is Its effective range of action is For the potential well during rapid eye movement (REM) sleep, the center is located at... Width is Its effective range of action is For the potential well during deep sleep, the center is located at... Width is Its effective range of action is Among them, when the sleep depth state When a potential well falls into its effective range, it is determined to enter the corresponding sleep stage. As the potential well center during rapid eye movement, For the potential well half-width during rapid eye movement (REM) It is the potential trap center during deep sleep. This represents the upper limit of the sleep depth coordinates. The potential trap width is half-width during deep sleep.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: By simultaneously acquiring the user's radar echo signal and body pressure distribution signal through a millimeter-wave radar sensor array and a distributed thin-film pressure sensor array laid within the mattress, the advantages of both sensors are fully utilized. The millimeter-wave radar can capture subtle movements and vital signs, while the pressure sensors reflect the distribution of body pressure, providing a comprehensive and complementary data foundation for subsequent analysis. Through in-depth processing of the acquired signals, radar signal processing extracts initial vital sign signals and micro-motion trajectories, while pressure signal processing yields the pressure center trajectory and pressure fluctuation signals, providing strong support for accurate analysis of the user's condition. Based on the micro-motion trajectory... A fusion processing mechanism is established with the pressure center trajectory to calculate the body movement fusion confidence score. By evaluating the quality of the two trajectories, the dominant reliable trajectory is determined, thereby generating the body movement fusion confidence score and improving the accuracy of judging the user's body movement. Based on the body movement fusion confidence score, the signal processing path is adaptively selected. In static conditions, the frequency domain signal fusion path is activated to generate fused vital sign signals, and in the case of body movement, the body movement analysis path is activated to identify the macroscopic body movement type, ensuring optimal processing of the system under different states. By establishing a sleep depth potential energy field, simulating the evolution of sleep state, and dynamically distinguishing sleep stages, physiological parameters and body movement types can be combined to achieve accurate analysis of sleep stages through dynamic equations. Attached Figure Description

[0016] Figure 1 This is a flowchart of the mattress-type imperceptible sleep detection method that integrates millimeter-wave radar and pressure sensing proposed in this invention. Detailed Implementation

[0017] 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.

[0018] Reference Figure 1 A mattress-based, non-invasive sleep detection method integrating millimeter-wave radar and pressure sensing includes: S1. Simultaneously collect the user's radar echo signal and body pressure distribution signal through a millimeter-wave radar sensor array and a distributed thin-film pressure sensor array laid inside the mattress; S2. Process the radar echo signal to extract the initial vital signs signal and micro-motion trajectory; at the same time, process the body pressure distribution signal to extract the pressure center trajectory and pressure fluctuation signal. S3. Based on the micro-motion trajectory and the pressure center trajectory, establish a trajectory consistency fusion processing mechanism and calculate the body motion fusion confidence level; S4. Based on the confidence level of body movement fusion, different signal processing pathways are adaptively selected and executed, including frequency domain signal fusion pathway and body movement analysis pathway. If the confidence level of body movement fusion is lower than the preset pathway switching threshold, the user is determined to be in the static-dominant stage, and the frequency domain signal fusion pathway is activated to perform frequency domain fusion of the initial vital sign signal and the pressure fluctuation signal to generate a fused vital sign signal. If the confidence level of body movement fusion is higher than or equal to the preset pathway switching threshold, the user is determined to be in the body movement-dominant stage, and the body movement analysis pathway is activated to identify the macroscopic body movement type based on the micro-motion trajectory and the pressure center trajectory. The pathway switching threshold is a preset critical value used to achieve adaptive selection of signal processing pathways. This threshold dynamically determines the system working mode by comparing the quantitative results of body movement fusion confidence level, ensuring that the system can adopt the optimal data processing strategy under different physiological states. S5. Based on the fusion of vital signs signals and macroscopic body movement types, a sleep depth potential energy field is established; by simulating the evolution process of sleep state driven by dynamic equations in this potential energy field, the user's sleep stage is dynamically determined.

[0019] It should be further explained that, in the specific implementation process, step S2 involves processing the radar echo signal to extract the initial vital signs signal and micro-motion trajectory; simultaneously, processing the body pressure distribution signal to extract the pressure center trajectory and pressure fluctuation signal includes: S211. Preprocess the multi-channel radar echo signal synchronously acquired by the millimeter-wave radar sensor array, including DC component removal and background clutter cancellation, to enhance the signal-to-noise ratio of dynamic targets. S212. Perform joint spatial-temporal processing on the preprocessed multi-channel radar echo signals: Perform a fast Fourier transform on the radar echo signal of each channel in the fast time dimension (i.e., the sampling sequence within a single radar frequency modulation period) to form the range profile of each channel; Utilize the geometric configuration of the millimeter-wave radar sensor array to perform spatial filtering or adaptive beamforming on the multi-channel radar echo signals in the same range cell, locate and enhance the echo signals from the master bedroom area above the mattress by means of spatial spectrum estimation, while suppressing static clutter and multipath reflection interference from the side or non-target areas, and output the signal after spatial enhancement. S213. For the signal after spatial enhancement, perform a fast Fourier transform in the slow time dimension (i.e., for the sequence spanning multiple consecutive radar frequency modulation cycles) to generate a high signal-to-noise ratio range-Doppler spectrum and construct a range-Doppler-time three-dimensional data cube. S214. Combining the range-Doppler-time three-dimensional data cube, using an energy entropy-based region of interest automatic detection algorithm, the range-Doppler cell corresponding to the human chest cavity is located, and the phase history sequence of the signal within the cell is extracted; the phase history sequence is unwrapped, and the unwrapped phase signal is bandpass filtered with a passband range of 0.1Hz to 3Hz to separate the initial vital signs signal. Understandably, by using an energy entropy-based region of interest automatic detection algorithm, the distance-Doppler cell corresponding to the human thoracic cavity is located, specifically including: Z1. For the distance-Doppler spectrum at the current moment (i.e., a slice of the three-dimensional data cube), it is denoted as a matrix. Its size is ,in The total number of distance cells. Given the total number of Doppler elements; calculate the energy matrix of the spectrum. Each element That is, the square of the signal amplitude of each unit, which is used to characterize the energy of that unit; For row index, indicating the row number. One distance unit, ; For column index, it means the first... One Doppler unit, ; Indicates the first The distance unit, the first Complex signal values ​​on each Doppler unit; Z2, Define an energy matrix Upward sliding rectangular window Its size is , This indicates the number of cells covered by the window in the distance dimension (row direction). This indicates the number of elements covered by the window in the Doppler (column direction); the window traverses the energy matrix. All possible positions; for a rectangular window In position The local area covered by time is labeled with the set of energy values ​​it contains. ;in, This indicates the row index (corresponding to the distance cell index) where the top left corner of the window is located. This indicates the column index (corresponding to the Doppler cell index) where the top left corner of the window is located. This represents the first energy value. This indicates the second energy value; This indicates the last energy value; Z3, for each window position Calculate the energy entropy of this local region. Normalize the energy values ​​of each cell within the window to a probability distribution; calculate the total energy within the window. (in The total energy within the window. (This involves iterating through the energy values ​​within the window using indices) and simultaneously calculating the energy percentage of each unit. ( (This represents the energy percentage); according to the information entropy formula, calculate the energy entropy of this local region: Understandably, energy entropy The entropy value quantifies the degree of order or concentration of energy distribution in the local area. A lower entropy value indicates that the energy is highly concentrated in a few units (e.g., a sharp spectral peak), which may correspond to a real target (e.g., chest cavity micromovement); while a higher entropy value indicates that the energy distribution is very dispersed (e.g., uniform background noise or dispersed interference), and does not have obvious target characteristics. Z4. After traversal, an energy entropy map with the same size as the distance-Doppler spectrum is obtained. In the energy entropy map, all local minima with entropy values ​​lower than the preset entropy threshold are searched. Combined with the prior distance or Doppler range of vital signs (for example, the human chest cavity is usually located within a specific distance range above the mattress, and its micro-motion velocity (i.e., Doppler frequency, within a specific low-frequency range)), the units that best meet the expected distance and Doppler range are selected in these low-entropy regions. Finally, the distance-Doppler unit with the lowest entropy value that meets the spatial and Doppler constraints is locked as the target unit corresponding to the human chest cavity. S215. Perform dynamic target detection on the 3D data cube to identify dynamic point clouds other than the main target of the human torso; adopt a trajectory association method based on the nearest neighbor criterion, for the dynamic point cloud detected at each time moment, find the nearest point cloud in the dynamic point cloud set of adjacent time moments, and associate them to form a trajectory segment; by splicing and smoothing the trajectory segments at multiple time moments, eliminate the trajectory discontinuity caused by detection errors or brief interference, and form a micro-motion trajectory representing the user's limbs. S216. For each frame of body pressure distribution signal acquired by the distributed thin-film pressure sensor array, construct a two-dimensional pressure matrix, where each element of the matrix corresponds to the pressure value of a sensor unit; calculate the centroid coordinates of the two-dimensional pressure matrix. The calculation formula is as follows: ; In the formula, For the first Line 1 Coordinates of the column sensor unit This is the pressure value of the unit. These represent the number of rows and columns of the pressure matrix, respectively. S217. According to the time sequence, calculate the centroid coordinates of each frame of body pressure distribution signal. Perform sequential concatenation: for the current moment in a continuous time series centroid coordinates Find its centroid coordinates at the next moment. And connect them with straight line segments to form a continuous pressure center trajectory; S218. In each frame of the body pressure distribution signal, using the centroid coordinates of the current frame... Based on this, a rectangular region of interest corresponding to the user's chest cavity is automatically defined. The pressure values ​​of all sensor units within this rectangular region of interest are summed to obtain the total pressure sequence for that region. The total pressure sequence is then subjected to bandpass filtering with a passband range of 0.1Hz to 0.8Hz to extract the pressure fluctuation signals mainly caused by respiratory activity.

[0020] It should be further explained that, in the specific implementation process, the process of establishing a trajectory consistency fusion processing mechanism based on the micro-motion trajectory and the pressure center trajectory in step S3, and calculating the body motion fusion confidence level, includes: S31. Calculate the quality indices of the micro-motion trajectory and the pressure center trajectory respectively: The quality index of the micro-motion trajectory is its average signal-to-noise ratio within the sliding time window, which is obtained by averaging the ratio of the radar cross-section of all points in the trajectory to the background noise power; The quality index of the pressure center trajectory is its trajectory smoothness within the sliding time window, which is quantified by calculating the norm of the second derivative of the trajectory displacement vector. The smaller the norm, the higher the smoothness. S32. Set the first threshold T1 as the lowest reliable value of the average signal-to-noise ratio of the micro-motion trajectory, the second threshold T2 as the highest reliable value of the trajectory smoothness of the pressure center trajectory, the third threshold T3 as the lowest reliable value of the trajectory smoothness of the pressure center trajectory, and the fourth threshold T4 as the highest reliable value of the average signal-to-noise ratio of the micro-motion trajectory. Specifically, the first threshold T1 (the lowest reliable value of the average signal-to-noise ratio of the micro-motion trajectory): This threshold is used to determine whether the radar signal is severely degraded. The basis for setting it is that when the signal-to-noise ratio of the micro-motion trajectory is lower than T1, it indicates that the radar is severely blocked (such as being covered by a thick blanket) or there is strong interference. At this time, the micro-motion trajectory it outputs is fragmented and full of noise, and can no longer reliably reflect the real body movement. The value of T1 is determined by statistically analyzing the signal-to-noise ratio data of a large number of radar failure scenarios (such as a user turning over in a blanket) (for example, taking the 5th percentile). The second threshold T2 (the highest reliable value for the smoothness of the pressure center trajectory): This threshold is used to identify when the user is in a highly static state. The smoothness of the pressure center trajectory is extremely high, indicating that the user's body has not undergone macroscopic and rapid displacement. At this time, the pressure sensor can stably and accurately capture subtle changes in the center of gravity. The basis for setting T2 is to calculate the distribution of the smoothness of the pressure trajectory during the period when the user is confirmed to be static (such as the deep sleep period confirmed by video), and take the higher quantile (e.g., the 95th percentile) as the threshold. When the smoothness is higher than T2, there is a very high confidence that the pressure trajectory is reliable. The third threshold T3 (the lowest reliable value for the smoothness of the pressure center trajectory): This threshold is used to determine whether the pressure signal is inaccurate due to violent and rapid body movements. When a user quickly turns over or sits up, the instantaneous acceleration of the body and the separation of contact with the mattress will cause the calculated pressure center to jump, and the trajectory smoothness will drop sharply. The T3 is set based on the analysis of the distribution of pressure trajectory smoothness during violent body movements, taking a lower quantile (e.g., the 5th percentile). When the smoothness is lower than T3, the pressure trajectory is determined to be unreliable. The fourth threshold T4 (the highest reliable value of the average signal-to-noise ratio of the micro-motion trajectory): This threshold is used to determine whether the radar signal is in a high-quality state. When the signal-to-noise ratio of the micro-motion trajectory is higher than T4, it indicates that the radar can clearly and stably track limb movements, and the micro-motion information it provides is highly reliable. The basis for setting T4 is to statistically analyze the distribution of the signal-to-noise ratio of the micro-motion trajectory under ideal conditions of good radar field of view and no obstruction, and take a higher quantile (e.g., the 95th percentile). S33. Determine the dominant reliable trajectory based on the quality indicators of the micro-motion trajectory and the pressure center trajectory: If the average signal-to-noise ratio of the micro-motion trajectory is lower than the first threshold and the trajectory smoothness of the pressure center trajectory is higher than the second threshold, then the pressure center trajectory is determined to be the dominant reliable trajectory; if the trajectory smoothness of the pressure center trajectory is lower than the third threshold and the average signal-to-noise ratio of the micro-motion trajectory is higher than the fourth threshold, then the micro-motion trajectory is determined to be the dominant reliable trajectory; specifically. S34. Based on the determination result of the dominant reliable trajectory, different strategies are used to generate the body motion fusion confidence: when the pressure center trajectory is determined to be dominant, the body motion fusion confidence is obtained based on the overall displacement of the trajectory within the determination time window, where the overall displacement of the trajectory is the weighted sum of the Euclidean distance between the trajectory start point and the end point and the sum of the magnitudes of all displacement vectors within the time window. When the micro-motion trajectory is determined to be dominant, the confidence level of body motion fusion is calculated based on the product of the spatial distribution dispersion of all dynamic points of the trajectory within the determination time window and the average motion velocity. When a single dominant trajectory cannot be determined, the confidence level of body-motion fusion is obtained by calculating the geometric mean of the cosine similarity between the micro-motion trajectory and the pressure center trajectory in the displacement direction and the Pearson correlation coefficient in the motion velocity. Understandably, when a single dominant trajectory cannot be determined, a body motion fusion confidence score can be generated using an attention-weighted multimodal feature fusion method: extract the feature vectors of the micro-motion trajectory and the pressure center trajectory respectively; input the average signal-to-noise ratio of the micro-motion trajectory and the trajectory smoothness of the pressure center trajectory into a lightweight neural network to generate attention weights for weighting the two spatiotemporal feature vectors; use the attention weights to perform a weighted summation of their respective spatiotemporal feature vectors to obtain a fused feature vector; input the fused feature vector into a pre-trained confidence regression model to output the final body motion fusion confidence score.

[0021] It should be further explained that, in the specific implementation process, in step S4, the initial vital sign signal and the pressure fluctuation signal are fused in the frequency domain to generate a fused vital sign signal; simultaneously, the process of identifying the type of macroscopic bodily movement includes: Short-time Fourier transforms were performed on the initial vital signs signal and the pressure fluctuation signal, respectively, to obtain the time spectrum of the initial vital signs signal, i.e., the radar source time spectrum. The time spectrum of the pressure fluctuation signal, i.e., the time spectrum of the pressure source. ,in For frequency, For time frames; Define adaptive fusion weights, where the expression for adaptive fusion weights is: ; In the formula, Indicated at each frequency and time frame The time spectrum allocated to the radar source The fusion weights; correspondingly, the spectrum allocated to the pressure source. The fusion weight is ; The frequency band preference base function is a frequency function preset based on prior knowledge of the dominant frequency bands of different physiological signals. Its value is between 0 and 1, and the principle for setting its function value is: within the pre-calibrated dominant heartbeat frequency band, A value close to 1 indicates that initial vital signs are preferentially trusted within this frequency band; within the pre-calibrated respiratory dominant frequency band, The value is set to approximately 0.5, indicating that within this frequency band, it provides an equal basis for the fusion of both initial vital sign signals and pressure fluctuation signals; in other non-physiologically dominant frequency bands, The value of approaches 0, which is used to suppress noise in non-physiological frequency bands; The time spectrum of the radar source at the frequency point The instantaneous signal quality factor at the location, When the pressure source is at a certain frequency point The instantaneous signal quality factor at a given frequency point, wherein the instantaneous signal quality factor is determined by the signal quality within the frequency neighborhood ( ,in The neighborhood window sizes used to calculate local signal quality are defined in both the frequency and time domains. The radius of the frequency neighborhood. The signal-to-noise ratio of the signal amplitude (within the time neighborhood radius) and the spectral peak significance are calculated together. The hyperbolic tangent function is used to map the difference in instantaneous signal quality factors to the interval [-1, 1] to ensure the fusion weights. The changes are smooth and continuous; Using the calculated adaptive fusion weights, the time spectrum of the radar source and the time spectrum of the pressure source are fused at the frequency point level to generate a fused time spectrum. The calculation formula is as follows: ; For the fused time spectrum Perform inverse short-time Fourier transform to reconstruct the fused vital sign signal in the time domain: B1, fused time spectrum A1. Perform inverse short-time Fourier transform to obtain a series of local time-domain signal frames; B2. Use window function complementarity processing, that is, multiply the inverse transformed time-domain signal frames by a synthetic window function with complementary rising and falling edges (for example, replace the Hanning window used in the analysis stage with a synthetic window that meets the perfect reconstruction condition, such as a double-length Hanning window and perform overlapping and cropping); B3. Superimpose and overlap the signal frames processed by the window function according to their corresponding time positions. The length of the overlapping region is not less than 50% of the length of a single frame. By adding the signal amplitudes of the overlapping regions, the abrupt changes between frames are eliminated, and finally, a complete and smooth time-domain fused vital sign signal is generated. In the motion analysis pathway, the micro-motion trajectory and the pressure center trajectory are synchronously divided into multiple consecutive short-time windows on the time axis with the same starting point and duration. The duration of each window is L seconds, and adjacent windows overlap by 50% (L is the preset window duration). The trajectory data within each short-term window is encoded to generate a fused trajectory segment descriptor. This descriptor includes the direction of movement of the pressure center within the segment, the spatial distribution entropy of the micro-movement points, and the mutual information of the velocity changes of the two trajectories. Specifically, the trajectory segment descriptor is generated as follows: the direction angle θ (θ∈[0°,360°)) of the line connecting the starting and ending points of the pressure center trajectory within the short-term window is calculated and used as the direction of movement of the pressure center; the spatial coordinates of all micro-movement trajectory points within the short-term window are divided into a grid on the mattress plane, the number of points in each grid is counted, and their probability distribution is calculated. Based on this probability distribution, the Shannon entropy is calculated, ultimately yielding the spatial distribution entropy of the micro-movement points; the average velocity sequence of the micro-movement trajectory and the velocity sequence of the pressure center trajectory within the short-term window are calculated separately, and the joint probability distribution and marginal probability distribution of the two velocity sequences are estimated using histogram statistics. Based on this, the mutual information value between them is calculated, thereby obtaining the mutual information of the velocity changes of the two trajectories. The trajectory segment descriptors generated in all time windows are dynamically clustered using an online clustering algorithm. The online clustering algorithm adopts the CluStream streaming data clustering algorithm, which maintains the data summary online through micro-cluster structure and generates macro-clusters periodically. Trajectory segments with the same motion mode are grouped into the same cluster. The cluster center of each macro-cluster is defined as a behavioral primitive to represent a typical, fine-grained cross-modal motion mode. The sequence of behavioral primitives arranged in chronological order is matched with a predefined body movement grammar library. The body movement grammar library consists of a set of deterministic finite automata, each corresponding to a macroscopic body movement type, and its state transitions are triggered by a predefined sequence of behavioral primitives. Specifically, in the body movement grammar library, the rolling over event is defined as a pattern containing a specific sequence of lateral pressure movement primitives and high spatial entropy micro-movement primitives; the sitting up event is defined as a pattern dominated by longitudinal pressure movement primitives and subsequently accompanied by micro-movement silence primitives; the getting out of bed event is defined as a pressure trajectory sequence matching the getting out of bed pattern and the micro-movement trajectory entering a continuous silence state within the bed body. The currently generated sequence of behavioral primitives is used as the input string and fed in parallel into the automata corresponding to all macroscopic body movement types. When an automaton reaches its terminal state after going through a series of state transitions from the initial state, the macroscopic body movement type corresponding to that automaton is determined and the recognition result is output.

[0022] It should be further explained that, in the specific implementation process, step S5, which involves dynamically determining the user's sleep stage by establishing a sleep depth potential field, includes: Define a one-dimensional potential field for sleep depth. Where DS represents the sleep depth coordinates; potential energy field It consists of multiple predefined stable potential wells stacked together, each potential well corresponding to a sleep stage and defined by its center position and width parameters: for a wakefulness potential well, the center is located at... Width is Its effective range of action is For the potential well during rapid eye movement (REM) sleep, the center is located at... Width is Its effective range of action is For the potential well during deep sleep, the center is located at... Width is Its effective range of action is Among them, when the sleep depth state When a potential well falls into its effective range, it is determined to enter the corresponding sleep stage. As the potential well center during rapid eye movement, For the potential well half-width during rapid eye movement (REM) It is the potential trap center during deep sleep. This represents the upper limit of the sleep depth coordinates. The potential trap width is half-width during deep sleep; At each processing time v, based on the physiological parameters parsed from the fused vital signs signals and the macroscopic body movement types identified by the body movement analysis pathway, the net force driving the evolution of the sleep state in the sleep depth potential field is calculated and analyzed. : ; In the formula, Vital signs are calculated from physiological parameters (heart rate, respiratory rate, and heart rate variability) obtained by fusing vital sign signals. Their direction and magnitude depend on the vital sign status and current sleep depth. Deviation from expected state; specifically, when heart rate and respiratory rate are detected to be lower than their individualized baseline, and heart rate variability indicators (such as the LF / HF ratio) are reduced. This is a positive force pointing towards deeper sleep (increased DS); when a sudden increase in heart rate or respiratory rate, or an abnormally increased variability is detected, This is a negative force pointing towards lighter sleep (reduced DS); understandably, heart rate, respiratory rate, and heart rate variability indices are extracted from the fused vital signs signal. Specifically: spectral analysis is performed on the fused vital signs signal by finding spectral peaks in the 0.8Hz to 3.0Hz frequency band, converting the frequencies corresponding to the peaks into heart rate per minute; spectral analysis is also performed on the fused vital signs signal by finding spectral peaks in the 0.1Hz to 0.8Hz frequency band, converting the frequencies corresponding to the peaks into respiratory rates per minute; and heart rate interval sequences are extracted from the fused vital signs signal, and the variability of the heart rate interval sequences in the time or frequency domain is calculated to obtain the heart rate variability index. The body motion impact force is triggered by the macroscopic body motion type identified by the body motion analysis pathway. The body motion impact force is an instantaneous negative force whose amplitude is related to the macroscopic body motion type and intensity and whose direction is directed towards the lucid period. As a biological clock regulating force, it is a slowly varying force function related to cumulative sleep time; it provides a positive force to promote sleep in the early stages of sleep, and gradually turns into a negative force to promote wakefulness as sleep time increases; Based on net force The sleep depth state is obtained by evolving the dynamic equations of simulated inertia and damping. : ; In the formula, m is the mass parameter simulating the inertia of the sleep state, which characterizes the ease or difficulty of changing the state. The damping coefficient of the system is used to simulate the resistance to changes in state. For the potential energy field of deep sleep The gradient represents the restoring force of the sleep depth potential field itself on the sleep depth state, driving the sleep depth state. Slide towards the nearest stable potential well; The sleep depth state is obtained by solving the dynamic equations through numerical integration. The evolutionary trajectory;

[0023] The sleep stage determination is based on the currently locked stable potential well; changes in sleep stage are modeled as sleep depth states. The dynamic process of transitions between different stable potential wells.

[0024] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. The focus of each embodiment is on its differences from other embodiments. In particular, the apparatus embodiments are described simply because they are fundamentally based on the method embodiments; relevant details can be found in the descriptions of the method embodiments.

[0025] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0026] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0027] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0029] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0030] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A mattress-based, non-invasive sleep detection method integrating millimeter-wave radar and pressure sensing, characterized by: S1. Simultaneously collect the user's radar echo signal and body pressure distribution signal through a millimeter-wave radar sensor array and a distributed thin-film pressure sensor array laid inside the mattress; S2. Process the radar echo signal to extract the initial vital signs signal and micro-motion trajectory; at the same time, process the body pressure distribution signal to extract the pressure center trajectory and pressure fluctuation signal. S3. Based on the micro-motion trajectory and the pressure center trajectory, establish a trajectory consistency fusion processing mechanism and calculate the body motion fusion confidence level; S4. Based on the confidence level of body movement fusion, adaptively select and execute different signal processing pathways, and obtain fused vital sign signals and macroscopic body movement types; among which, different signal processing pathways include frequency domain signal fusion pathways and body movement analysis pathways; S5. Based on the fusion of vital signs signals and macroscopic body movement types, a sleep depth potential energy field is established; by simulating the evolution process of sleep state driven by dynamic equations in this potential energy field, the user's sleep stage is dynamically determined.

2. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 1, characterized in that, In step S2, the process of processing the radar echo signal to extract the initial vital signs signal and micro-motion trajectory includes: S21a. Preprocess the multi-channel radar echo signals synchronously acquired by the millimeter-wave radar sensor array; S22a. Perform joint spatial-temporal processing on the preprocessed multi-channel radar echo signal and output the signal after spatial enhancement. S23a. For the signal after spatial enhancement, perform a fast Fourier transform in the slow time dimension to generate a high signal-to-noise ratio range-Doppler spectrum and construct a range-Doppler-time three-dimensional data cube. S24a. Using an energy entropy-based region of interest automatic detection algorithm, the distance-Doppler unit corresponding to the human chest cavity is located and its phase history sequence is extracted; after phase unwrapping and bandpass filtering, the initial signal of vital signs is separated. S25a. Perform dynamic target detection on the three-dimensional data cube to identify dynamic point clouds; for the dynamic point cloud detected at each time step, find the nearest point cloud in the dynamic point cloud set of adjacent time steps, associate them to form trajectory segments, and form micro-motion trajectories after splicing and smoothing.

3. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 1, characterized in that, In step S2, the process of processing the body pressure distribution signal and extracting the pressure center trajectory and pressure fluctuation signals includes: S21b. For each frame of body pressure distribution signal acquired by the distributed thin-film pressure sensor array, construct a two-dimensional pressure matrix, where each element of the matrix corresponds to the pressure value of a sensor unit; calculate the centroid coordinates of the two-dimensional pressure matrix. ; S22b. Calculate the centroid coordinates of each frame of body pressure distribution signal in chronological order. Connect them sequentially to form a continuous pressure center trajectory; S23b. In each frame of the body pressure distribution signal, using the centroid coordinates of the current frame... Based on this, a rectangular region of interest corresponding to the user's chest cavity is automatically defined. The pressure values ​​of all sensor units within this rectangular region of interest are summed to obtain the total pressure sequence for that region. The total pressure sequence is then bandpass filtered to extract the pressure fluctuation signals caused by respiratory activity.

4. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 1, characterized in that, In step S3, the process of establishing a trajectory consistency fusion processing mechanism based on the micro-motion trajectory and the pressure center trajectory, and calculating the body motion fusion confidence level, includes: S31. Calculate the quality indices of the micro-motion trajectory and the pressure center trajectory respectively: the quality index of the micro-motion trajectory is its average signal-to-noise ratio within the sliding time window; the quality index of the pressure center trajectory is its trajectory smoothness within the sliding time window. S32. Based on quality indicators, set a first threshold, a second threshold, a third threshold, and a fourth threshold respectively; S33. Determine the dominant reliable trajectory based on the quality indicators of the micro-motion trajectory and the pressure center trajectory: If the average signal-to-noise ratio of the micro-motion trajectory is lower than the first threshold and the trajectory smoothness of the pressure center trajectory is higher than the second threshold, then the pressure center trajectory is determined to be the dominant reliable trajectory; if the trajectory smoothness of the pressure center trajectory is lower than the third threshold and the average signal-to-noise ratio of the micro-motion trajectory is higher than the fourth threshold, then the micro-motion trajectory is determined to be the dominant reliable trajectory. S34. Based on the determination results of the dominant reliable trajectory, different strategies are used to generate the confidence level of body motion fusion.

5. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 4, characterized in that, In step S34, the process of generating body motion fusion confidence scores using different strategies based on the determination result of the dominant reliable trajectory includes: When the pressure center trajectory is determined to be dominant, the body motion fusion confidence is obtained based on the overall displacement of the trajectory within the determination time window. The overall displacement of the trajectory is the weighted sum of the Euclidean distance between the trajectory start and end points and the sum of the magnitudes of all displacement vectors within the time window. When the micro-motion trajectory is determined to be dominant, the confidence level of body motion fusion is calculated based on the product of the spatial distribution dispersion of all dynamic points of the trajectory within the determination time window and the average motion velocity. When a single dominant trajectory cannot be determined, the confidence level of body-motion fusion is obtained by calculating the geometric mean of the cosine similarity between the micro-motion trajectory and the pressure center trajectory in the displacement direction and the Pearson correlation coefficient in the motion velocity.

6. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 1, characterized in that, Step S4 further includes: if the confidence level of body movement fusion is lower than the preset pathway switching threshold, the user is determined to be in the static-dominant stage, and the frequency domain signal fusion pathway is activated to perform frequency domain fusion of the initial vital sign signal and the pressure fluctuation signal to generate a fused vital sign signal; if the confidence level of body movement fusion is higher than or equal to the preset pathway switching threshold, the user is determined to be in the body movement-dominant stage, and the body movement analysis pathway is activated to identify the macroscopic body movement type based on the micro-motion trajectory and the pressure center trajectory.

7. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 6, characterized in that, The process of frequency domain fusion of initial vital signs signals and pressure fluctuation signals to generate fused vital signs signals includes: Short-time Fourier transforms were performed on the initial vital signs signal and the pressure fluctuation signal respectively to obtain the time spectrum of the initial vital signs signal, i.e., the radar source time spectrum, and the time spectrum of the pressure fluctuation signal, i.e., the pressure source time spectrum. Define adaptive fusion weights and use them to perform frequency-point weighted fusion of the radar source time spectrum and the pressure source time spectrum to generate a fused time spectrum; The fused time-domain fusion vital signs signal is reconstructed by performing an inverse short-time Fourier transform on the fused time-domain spectrum.

8. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 6, characterized in that, The process of identifying macroscopic body movement types includes: In the body motion analysis pathway, the micro-motion trajectory and the pressure center trajectory are synchronously divided into multiple consecutive short-time windows on the time axis with the same starting point and duration; The trajectory data within each short time window is encoded to generate a trajectory segment descriptor; An online clustering algorithm is used to dynamically cluster trajectory segment descriptors generated in all time windows. Specifically, a micro-cluster structure is used to maintain the data summary online and generate macro-clusters periodically. Trajectory segments with the same motion mode are grouped into the same cluster, and the cluster center is defined as the behavioral primitive representing cross-modal motion mode. The chronological sequence of action primitives is matched against a predefined body movement grammar; the body movement grammar consists of a set of automata, each corresponding to a macroscopic body movement type. The currently generated sequence of behavioral primitives is used as the input string and fed in parallel into the automata corresponding to all macroscopic body movement types. When an automaton reaches its terminal state after undergoing a state transition from the initial state, the macroscopic body movement type corresponding to that automaton is determined and the recognition result is output.

9. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 1, characterized in that, In step S5, the process of dynamically determining the user's sleep stage by establishing a sleep depth potential field includes: Define a one-dimensional potential field for sleep depth. ; At each processing time Based on the physiological parameters parsed from fused vital sign signals and the macroscopic body movement types identified by the body movement analysis pathway, the net force driving the evolution of sleep state in the sleep depth potential field is calculated and analyzed. ; Based on net force The sleep depth state is obtained by evolving the dynamic equations of simulated inertia and damping. ; The sleep depth state is obtained by solving the dynamic equations through numerical integration. The evolutionary trajectory; The sleep stage determination is based on the currently locked stable potential well; changes in sleep stage are modeled as sleep depth states. The dynamic process of transitions between different stable potential wells.

10. The mattress-type imperceptible sleep detection method integrating millimeter-wave radar and pressure sensing according to claim 9, characterized in that, In the defined potential energy field of sleep depth In the diagram, DS represents the sleep depth coordinates; potential energy field It consists of multiple predefined stable potential wells stacked together, each potential well corresponding to a sleep stage and defined by its center position and width parameters: for a wakefulness potential well, the center is located at... Width is Its effective range of action is For the potential well during rapid eye movement (REM) sleep, the center is located at... Width is Its effective range of action is For the potential well during deep sleep, the center is located at... Width is Its effective range of action is .

Images