Methods, systems, computer equipment and storage media for dual-person sleep monitoring

Abstract

This application relates to a method, system, computer device, and storage medium for dual-person sleep monitoring. This application converts the acquired mixed multimodal signals from both individuals into a time-frequency graph matrix, extracts the spatial features of body movement amplitude and posture angle changes from the time-frequency graph matrix, captures the time-series features of respiratory signals, and decouples the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual. This overcomes the limitations of traditional Fourier transform and other algorithms, achieving accurate separation of dual-person sleep multimodal signals, solving the problem of signal cross-interference, and through multi-algorithm collaborative processing, achieving high-precision extraction of key parameters such as respiratory rate and body movement during dual-person sleep, reducing the monitoring error rate to within 3%. The method of analyzing each individual's sleep quality and generating individual sleep monitoring reports can adapt to different individual body types, sleep habits, and spatial location changes, improving the algorithm's generalization ability and effectively reducing monitoring blind spots and the risk of misjudgment.

Description

Technical Field

[0001] This application relates to the field of dual-person sleep monitoring technology, and in particular to a dual-person sleep monitoring method, system, computer device, and storage medium. Background Technology

[0002] Currently, in the field of dual-person sleep monitoring, although millimeter-wave radar has become an important technological means due to its advantages such as non-contact operation and strong penetration, existing technologies have significant limitations in dealing with complex interference and signal decoupling. Existing millimeter-wave radar dual-person sleep monitoring systems typically consist of a millimeter-wave radar sensor, a signal processing unit, and a data analysis module. In actual operation, the millimeter-wave radar sensor first continuously transmits millimeter-wave signals into the sleep area. Upon encountering the sleeping human body, the signal is reflected back, forming an echo signal. The echo signal enters the signal processing unit, where it undergoes low-pass filtering to remove high-frequency noise, and then is amplified to increase the signal strength. Subsequently, the echo signal is transmitted to the data analysis module. Traditional methods often employ basic signal processing algorithms such as Fourier transform to analyze the spectrum of the echo signal, attempting to separate the breathing and body movement information of each individual from the mixed signal, and then calculate parameters such as sleep respiratory rate and body movement frequency. Summary of the Invention

[0003] Based on this, a dual-person sleep monitoring method, system, computer device, and storage medium are provided to address the technical problems that traditional preprocessing methods such as filtering are insufficient to completely eliminate interference from electromagnetic interference and background noise in the environment, leading to a decline in signal quality; and that traditional algorithms based on Fourier transform cannot effectively separate the signal characteristics of each person and accurately extract key sleep parameters such as sleep respiratory rate and body movement for complex echo signals generated during dual-person sleep due to mutual obstruction and movement interference, thus failing to meet the requirements of practical applications in terms of anti-interference capability and monitoring accuracy.

[0004] On the one hand, a method for dual-person sleep monitoring is provided, the method comprising:

[0005] Millimeter wave signals are transmitted to the sleeping area of ​​two people, and echo signals reflected by the human body are received to obtain a mixed multimodal signal of two people containing information on breathing, body movement and posture changes of two people;

[0006] The dual-person mixed multimodal signal is converted into a time-frequency graph matrix. The spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix are extracted. The time-series features of the respiratory signal are captured and obtained. The respiratory signal sequence, body motion signal sequence and posture signal sequence of each person are decoupled.

[0007] The respiratory rate is obtained from the respiratory signal sequence, the body movement parameters are obtained from the body movement signal sequence, and the sleep posture is identified from the posture signal sequence.

[0008] The sleep quality of each person is analyzed based on the respiratory rate, body movement parameters, and sleep posture, and a sleep monitoring report for each person is generated.

[0009] In one embodiment, the step of converting the dual-person mixed multimodal signal into a time-frequency map matrix, extracting the spatial features of body motion amplitude and posture angle changes in the time-frequency map matrix, capturing the time-series features of the respiratory signal, and decoupling to obtain the respiratory signal sequence, body motion signal sequence, and posture signal sequence for each person includes:

[0010] When converting the dual-person hybrid multimodal signal into a time-frequency diagram matrix, the time-frequency domain and spatial array phase information of the signal are preserved, wherein the dimension of the time-frequency diagram matrix is ​​[time × frequency × number of antenna channels];

[0011] The time-frequency graph matrix is ​​input into a convolutional neural network to extract the spatial features of body motion amplitude and attitude angle changes in the time-frequency graph matrix. The convolutional neural network includes at least two convolutional layers, and the dimension of the features output by the convolutional neural network is [batch size × time step × number of feature channels].

[0012] By using a bidirectional gated recurrent unit to model the temporal features of the features output by the convolutional neural network, the periodic micro-motion features of the respiratory signal are captured, and the time-series features of the respiratory signal are obtained.

[0013] The respiratory, body movement, and posture changes of each person in the dual-person mixed multimodal signal are separated and decoupled to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person.

[0014] In one embodiment, after acquiring the dual-person hybrid multimodal signal, the method further includes:

[0015] The dual-person hybrid multimodal signal is preprocessed by using a bandpass filter to remove electromagnetic interference in specific frequency bands of the environment and using an adaptive gain amplifier to dynamically adjust the gain according to the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal.

[0016] In one embodiment, after decoupling to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual, the method further includes:

[0017] Collect signal characteristics of electromagnetic interference and background noise, and establish an interference signal characteristic database;

[0018] The respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person are compared with the interference signal feature database. The interference type is identified using a support vector machine classifier. The interference types include narrowband interference, broadband interference, human body occlusion clutter, and periodic body movement artifacts.

[0019] When the identified interference type is narrowband interference, a tunable notch filter is used for frequency suppression, and the center frequency is dynamically adjusted according to the interference frequency.

[0020] When the identified interference type is broadband interference, wavelet threshold denoising algorithm is used for wavelet packet denoising, and the number of decomposition layers is adaptively selected based on noise energy.

[0021] When the identified interference type is human body obstruction clutter, the spatial spectrum estimation algorithm is used to reconstruct the target position using the phase difference of a multi-antenna array;

[0022] Based on real-time data from environmental sensors, the parameters of the anti-interference algorithm are dynamically adjusted to achieve adaptive suppression of interference signals.

[0023] In one embodiment, the step of collecting signal characteristics of electromagnetic interference and background noise and establishing an interference signal characteristic database includes:

[0024] Pre-collect various typical interference signals, including narrowband electromagnetic interference, broadband noise, and human body obstruction clutter.

[0025] Extract the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signal. The characteristic parameters include center frequency, energy entropy, and wavelet coefficient variance.

[0026] By pre-collecting signal characteristics of various background noises and combining them with the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signals, an interference signal feature database is established.

[0027] In one embodiment, the steps of obtaining respiratory rate based on the respiratory signal sequence, obtaining body movement parameters based on the body movement signal sequence, and identifying sleep posture based on the posture signal sequence include:

[0028] The respiratory signal sequence is processed by Hilbert-Huang transform, noise modes are separated by empirical mode decomposition, and the intrinsic mode function components containing the respiratory frequency are selected for Hilbert spectral transform to extract the instantaneous frequency of the respiratory signal to form the respiratory frequency.

[0029] The amplitude change of the body movement signal sequence is detected by amplitude threshold, the time interval of the amplitude change is detected by time interval threshold, and the strong body movement and weak body movement are distinguished by the slope change of the time domain waveform of the body movement signal. The number of body movements and the intensity of body movements are counted as body movement parameters.

[0030] A three-dimensional attitude angle calculation model is constructed based on the phase difference of the echo signal from a multi-antenna array. The mapping relationship between attitude angle and signal features is fitted by a support vector regression algorithm. The attitude signal sequence is input into the three-dimensional attitude angle calculation model to identify attitude angle features by pitch, roll, and azimuth. The corresponding sleep posture is identified based on the attitude angle features. The sleep posture includes supine, lateral, and prone positions.

[0031] On the other hand, a dual-person sleep monitoring system is provided, the system comprising:

[0032] Millimeter-wave radar sensors are used to transmit millimeter-wave signals into the sleeping area of ​​two people, receive echo signals reflected by the human body, and acquire mixed multimodal signals of two people containing information on breathing, body movement, and posture changes of two people.

[0033] The multimodal signal decoupling module is used to convert the dual-person mixed multimodal signal into a time-frequency diagram matrix, extract the spatial features of body motion amplitude and posture angle changes in the time-frequency diagram matrix, capture the time series features of the respiratory signal, and decouple to obtain the respiratory signal sequence, body motion signal sequence and posture signal sequence of each person.

[0034] The sleep parameter extraction module is used to obtain the respiratory rate based on the respiratory signal sequence, obtain the body movement parameters based on the body movement signal sequence, and identify the sleep posture based on the posture signal sequence.

[0035] The data processing terminal is used to analyze the sleep quality of each person based on the respiratory rate, the body movement parameters, and the sleep posture, and to generate a sleep monitoring report for each person.

[0036] In one embodiment, the dual-person sleep monitoring system further includes:

[0037] The signal preprocessing module is used to preprocess the dual-person hybrid multimodal signal, using a bandpass filter to remove electromagnetic interference in specific frequency bands of the environment, and using an adaptive gain amplifier to dynamically adjust the gain according to the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal.

[0038] An adaptive anti-interference module is used to establish an interference signal feature database. It compares each person's respiratory signal sequence, body movement signal sequence, and posture signal sequence with the interference signal feature database, uses a support vector machine classifier to identify the interference type, and adopts the corresponding anti-interference strategy according to the identified interference type.

[0039] In another aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a dual-person sleep monitoring method.

[0040] In another aspect, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of a dual-person sleep monitoring method.

[0041] The aforementioned dual-person sleep monitoring method, system, computer equipment, and storage medium convert the acquired dual-person mixed multimodal signals into a time-frequency graph matrix, extract the spatial features of body movement amplitude and posture angle changes from the time-frequency graph matrix, capture the time-series features of respiratory signals, and decouple each person's respiratory signal sequence, body movement signal sequence, and posture signal sequence. This overcomes the limitations of traditional Fourier transform and other algorithms, achieving accurate separation of dual-person sleep multimodal signals, resolving signal cross-interference issues, and through multi-algorithm collaborative processing, achieving high-precision extraction of key parameters such as respiratory rate and body movement during dual-person sleep. The monitoring error rate is reduced to within 3%, meeting the stringent requirements of the medical and health field. The method of analyzing each person's sleep quality and generating individual sleep monitoring reports can adapt to different individual body types, sleep habits, and spatial location variations, improving algorithm generalization ability and effectively reducing monitoring blind spots and the risk of misjudgment. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating a dual-person sleep monitoring method in one embodiment of this application;

[0044] Figure 2 This is a structural block diagram of a dual-person sleep monitoring system in one embodiment of this application;

[0045] Figure 3 This is an internal structural diagram of a computer device in one embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] As described in the background section, existing technologies suffer from multi-dimensional technical bottlenecks in dual-person sleep monitoring applications, with the following specific shortcomings:

[0048] 1. Insufficient signal decoupling capability: Existing solutions use traditional signal processing algorithms such as Fourier transform, which lack an effective decoupling mechanism for multimodal signals (breathing, body movement, posture changes, etc.) in two-person sleep scenarios. It is difficult to separate mixed signals caused by mutual occlusion and motion interference, resulting in cross-contamination of target signals between two people and making it impossible to achieve independent and accurate extraction of individual sleep parameters.

[0049] 2. Limitations of the anti-interference mechanism: Relying on basic preprocessing methods such as low-pass filtering and amplification, it cannot cope with dynamic interference and background noise in complex electromagnetic environments. It lacks adaptive interference identification and suppression strategies and cannot adjust anti-interference parameters in real time according to environmental changes, resulting in a low signal-to-noise ratio of the echo signal, and critical sleep information is easily overwhelmed by interference signals.

[0050] 3. Poor monitoring accuracy and stability: No dedicated algorithm model has been developed for dual-person sleep scenarios, making it difficult to adapt to changes in signal characteristics under different sleep postures and movement patterns. In complex and noisy environments, traditional methods exhibit high error rates in monitoring key parameters such as sleep respiratory rate and body movement frequency, failing to meet the high-precision and high-reliability application requirements of the medical and health field for sleep monitoring.

[0051] 4. Insufficient scene adaptability: Existing technologies do not fully consider the special characteristics of two-person sleep scenarios and lack the ability to dynamically adapt to individual differences (body size, sleep habits) and changes in spatial location, resulting in poor algorithm generalization and the risk of monitoring blind spots and misjudgments in practical applications.

[0052] These technical shortcomings directly limit the application effectiveness of millimeter-wave radar in the field of dual-person sleep monitoring, and breakthroughs are urgently needed through innovative multimodal signal decoupling and adaptive anti-interference technologies.

[0053] To address the technical bottlenecks in existing millimeter-wave radar dual-person sleep monitoring technologies, such as insufficient multimodal signal decoupling capability, limited anti-interference mechanisms, poor monitoring accuracy and stability, and insufficient scene adaptability, this application proposes a novel dual-person sleep monitoring method. By constructing a deep learning-based multimodal signal decoupling model, it achieves precise separation of mixed signals such as breathing, body movement, and posture changes during dual-person sleep. Simultaneously, an adaptive interference identification and suppression algorithm is introduced, dynamically adjusting the anti-interference strategy based on environmental parameters to effectively eliminate the influence of electromagnetic interference and background noise on the echo signal. Furthermore, by establishing a personalized parameter adaptation mechanism for dual-person sleep scenarios, the algorithm's adaptability to different individual characteristics and spatial position changes is improved. Ultimately, this achieves high-precision and high-stability monitoring of key parameters such as respiratory rate and body movement during dual-person sleep, meeting the stringent requirements of the medical and health field for dual-person sleep monitoring.

[0054] In one embodiment, such as Figure 1 As shown, a method for dual-person sleep monitoring is provided, including the following steps:

[0055] Step S1: Transmit millimeter wave signals to the sleeping area of ​​the two people, receive the echo signals reflected by the human body, and obtain the mixed multimodal signal of the two people containing information on breathing, body movement, and posture changes of the two people.

[0056] Step S3: Convert the dual-person mixed multimodal signal into a time-frequency graph matrix, extract the spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix, capture the time series features of the respiratory signal, and decouple to obtain the respiratory signal sequence, body motion signal sequence and posture signal sequence of each person.

[0057] Step S5: Obtain respiratory rate based on the respiratory signal sequence, obtain body movement parameters based on the body movement signal sequence, and identify sleep posture based on the posture signal sequence;

[0058] Step S6: Analyze each person's sleep quality based on the respiratory rate, body movement parameters, and sleep posture, and generate a sleep monitoring report for each person.

[0059] This application converts the acquired mixed multimodal signals of two individuals into a time-frequency graph matrix, extracts the spatial features of body movement amplitude and posture angle changes from the time-frequency graph matrix, captures the time-series features of respiratory signals, and decouples the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual. It overcomes the limitations of traditional Fourier transform and other algorithms, achieving accurate separation of multimodal sleep signals from two individuals and solving the problem of signal cross-interference. Through multi-algorithm collaborative processing, it achieves high-precision extraction of key parameters such as respiratory rate and body movement during sleep for both individuals, reducing the monitoring error rate to within 3%, meeting the stringent requirements of the medical and health field. The method of analyzing each individual's sleep quality and generating individual sleep monitoring reports can adapt to different individual body types, sleep habits, and spatial location changes, improving the algorithm's generalization ability and effectively reducing monitoring blind spots and the risk of misjudgment.

[0060] Specifically, by converting the acquired mixed multimodal signals of two individuals into a time-frequency graph matrix, the spatial features of body movement amplitude and posture angle changes in the time-frequency graph matrix are extracted, and the time-series features of respiratory signals are captured. This decouples the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual, overcoming the limitations of traditional Fourier transform and other algorithms. This achieves accurate separation of multimodal sleep signals from two individuals, resolving signal cross-interference issues. Through multi-algorithm collaborative processing, high-precision extraction of key parameters such as respiratory rate and body movement during sleep is achieved, reducing the monitoring error rate to within 3%, meeting the stringent requirements of the medical and health field. The method of analyzing each individual's sleep quality and generating individual sleep monitoring reports can adapt to different individual body types, sleep habits, and spatial location variations, improving the algorithm's generalization ability and effectively reducing monitoring blind spots and the risk of misjudgment.

[0061] In this embodiment, the process of converting the dual-person mixed multimodal signal into a time-frequency graph matrix, extracting the spatial features of body motion amplitude and posture angle changes from the time-frequency graph matrix, capturing the time-series features of the respiratory signal, and decoupling to obtain the respiratory signal sequence, body motion signal sequence, and posture signal sequence for each individual includes:

[0062] When converting the dual-person hybrid multimodal signal into a time-frequency diagram matrix, the time-frequency domain and spatial array phase information of the signal are preserved, wherein the dimension of the time-frequency diagram matrix is ​​[time × frequency × number of antenna channels];

[0063] The time-frequency graph matrix is ​​input into a convolutional neural network to extract the spatial features of body motion amplitude and attitude angle changes in the time-frequency graph matrix. The convolutional neural network includes at least two convolutional layers, and the dimension of the features output by the convolutional neural network is [batch size × time step × number of feature channels].

[0064] The features output by the convolutional neural network are modeled using a bidirectional gated recurrent unit (Bi-GRU) to capture the periodic micro-motion features of the respiratory signal and obtain the time-series features of the respiratory signal.

[0065] The respiratory, body movement, and posture changes of each person in the dual-person mixed multimodal signal are separated and decoupled to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person.

[0066] This embodiment breaks through the limitations of traditional linear methods such as Fourier transform and short-time Fourier transform, solves the problem of cross-interference of nonlinear mixed signals, and improves the signal decoupling accuracy by more than 40%.

[0067] The multimodal signal decoupling accuracy is improved by more than 40%. Utilizing a CNN-LSTM fusion neural network, this method achieves deep separation of nonlinear features from the mixed signals of breathing (0.1-0.5Hz micro-movements), body movement (1-5Hz amplitude variations), and posture (spatial angular features) during sleep in two individuals. Compared to traditional Fourier transform methods (decoupling error ≥25%), this application extracts spatial features such as body movement amplitude and posture angles using CNN (recognition accuracy ≥85%), combined with LSTM capturing the periodic temporal features of breathing signals (correlation coefficient ≥0.92), reducing the separation error of the two-person target signal to below 15% (measured data), laying a high-purity data foundation for subsequent parameter extraction.

[0068] like Figure 1 As shown, in this embodiment, after acquiring the dual-person mixed multimodal signal, the method further includes:

[0069] Step S2 involves preprocessing the dual-person hybrid multimodal signal by using a bandpass filter to remove electromagnetic interference in specific frequency bands of the environment and using an adaptive gain amplifier to dynamically adjust the gain based on the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal.

[0070] like Figure 1 As shown, in this embodiment, after decoupling to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual, the following steps are also included:

[0071] Step S4: Establish an interference signal feature database. Compare each person's breathing signal sequence, body movement signal sequence, and posture signal sequence with the interference signal feature database. Use a support vector machine classifier to identify the interference type and adopt the corresponding anti-interference strategy based on the identified interference type.

[0072] This embodiment constructs an interference identification and dynamic suppression mechanism, which intelligently adjusts the anti-interference strategy according to environmental changes, significantly improving signal purity in complex environments compared to traditional fixed filtering methods.

[0073] The step of establishing an interference signal feature database involves comparing each person's respiratory signal sequence, body movement signal sequence, and posture signal sequence with the interference signal feature database, using a support vector machine classifier to identify the interference type, and adopting corresponding anti-interference strategies based on the identified interference type.

[0074] Collect signal characteristics of electromagnetic interference and background noise, and establish an interference signal characteristic database;

[0075] The respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person are compared with the interference signal feature database. The interference type is identified using a support vector machine (SVM) classifier. The interference types include narrowband interference, broadband interference, human body occlusion clutter, and periodic body movement artifacts.

[0076] When the identified interference type is narrowband interference, a tunable notch filter is used for frequency suppression, and the center frequency is dynamically adjusted according to the interference frequency.

[0077] When the identified interference type is broadband interference, wavelet threshold denoising algorithm is used for wavelet packet denoising, and the number of decomposition layers is adaptively selected based on noise energy.

[0078] When the identified interference type is human body obstruction clutter, the spatial spectrum estimation algorithm is used to reconstruct the target position using the phase difference of a multi-antenna array;

[0079] Based on real-time data from environmental sensors (electromagnetic intensity sensor, temperature and humidity sensor), the parameters of the anti-interference algorithm are dynamically adjusted to achieve adaptive suppression of interference signals.

[0080] Among them, the SVM classifier using the RBF kernel function is input with the feature vector of the current echo signal to achieve real-time identification of interference types (such as narrowband / wideband / periodic motion artifacts) with an accuracy of ≥95%.

[0081] This application achieves closed-loop adaptive processing of interference "identification-classification-suppression", improving anti-interference efficiency by more than 60%.

[0082] For narrowband electromagnetic interference (such as WiFi 2.4G harmonic interference), the interference suppression depth can be improved to below -30dB by using a tunable notch filter (the center frequency can be dynamically adjusted in steps ≤100kHz) (the suppression depth of traditional filters is ≤-15dB).

[0083] For broadband background noise (such as the vibration noise of an air conditioner motor), a wavelet packet denoising algorithm (adaptive decomposition layer ≤ 5 layers) is used to improve the signal-to-noise ratio (SNR) from 10dB to more than 25dB, effectively preserving the 0.1Hz level respiratory micro-motion signal.

[0084] For the mutual interference signals caused by dual-person occlusion, multi-sensor TDOA positioning technology (positioning accuracy ≤10cm) and spatial spectrum estimation algorithm are used to separate the cross echo signals, solving the signal loss problem caused by occlusion in traditional single sensors (the signal effectiveness in occlusion scenarios is increased from 75% to 92% in actual tests).

[0085] Environmental adaptability covers all scenarios. By integrating electromagnetic intensity sensors and temperature and humidity sensors, the system collects environmental parameters in real time and dynamically adjusts anti-interference algorithm parameters (such as notch filter bandwidth and wavelet threshold coefficients), enabling the system to maintain stable operation even in the following extreme scenarios:

[0086] In high electromagnetic interference scenarios (such as industrial control rooms): the bit error rate is reduced from 15% in traditional solutions to below 3%;

[0087] High humidity environment (humidity ≥ 85% RH): The distance measurement error is optimized from ±5cm to ±2cm by correcting the millimeter wave propagation delay through a temperature and humidity compensation model;

[0088] For complex sleep posture scenarios (such as side-lying curled up): relying on multi-angle radar deployment (head sensor tilt angle 15°-30°) and posture calculation model, the integrity of respiratory signal acquisition is improved from 60% to 85%.

[0089] In this embodiment, the step of collecting signal characteristics of electromagnetic interference and background noise and establishing an interference signal characteristic database includes:

[0090] Pre-collect various typical interference signals, including narrowband electromagnetic interference, broadband noise, and human body obstruction clutter.

[0091] Extract the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signal. The characteristic parameters include center frequency, energy entropy, and wavelet coefficient variance.

[0092] By pre-collecting signal characteristics of various background noises and combining them with the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signals, an interference signal feature database is established.

[0093] In this embodiment, obtaining the respiratory rate based on the respiratory signal sequence, obtaining the body movement parameters based on the body movement signal sequence, and identifying the sleep posture based on the posture signal sequence includes:

[0094] The respiratory signal sequence was processed by Hilbert-Huang transform (HHT), noise modes were separated by empirical mode decomposition (EMD), and intrinsic mode function (IMF) components containing the respiratory frequency were selected for Hilbert spectral transform to extract the instantaneous frequency of the respiratory signal to form the respiratory frequency.

[0095] The amplitude change of the body movement signal sequence is detected by amplitude threshold, the time interval of the amplitude change is detected by time interval threshold, and the strong body movement and weak body movement are distinguished by the slope change of the time domain waveform of the body movement signal. The number of body movements and the intensity of body movements are counted as body movement parameters.

[0096] A three-dimensional attitude angle calculation model is constructed based on the phase difference of the echo signal from a multi-antenna array. The mapping relationship between attitude angle and signal features is fitted by the support vector regression (SVR) algorithm. The attitude signal sequence is input into the three-dimensional attitude angle calculation model to identify attitude angle features by pitch angle, roll angle, and azimuth angle. The corresponding sleep posture is identified based on the attitude angle features. The sleep posture includes supine, lateral, and prone positions.

[0097] Among them, respiratory frequency extraction: the decoupled respiratory signal is processed by HHT, the noise mode is separated by empirical mode decomposition (EMD), the intrinsic mode function (IMF) is analyzed by Hilbert spectrum, and the instantaneous frequency of the respiratory signal is extracted with an accuracy of ±0.5 breaths / minute;

[0098] Among them, body movement parameter extraction: a dual-threshold dynamic detection algorithm (amplitude threshold + time interval threshold) is adopted, which combines the slope change of the time domain waveform of the body movement signal to distinguish between rolling over (strong body movement) and limb micro-movement (weak body movement). The accuracy of body movement recognition is ≥92%, providing refined data support for sleep cycle segmentation (light sleep / deep sleep / REM stage).

[0099] Among them, attitude recognition: based on the phase difference of the echo signal of the multi-antenna array, a three-dimensional attitude angle calculation model (pitch angle, roll angle, azimuth angle) is constructed. The mapping relationship between attitude angle and signal features is fitted by the support vector regression (SVR) algorithm, and the attitude classification accuracy reaches more than 85% (supine / lateral / prone), which solves the problem of respiratory signal attenuation caused by chest cavity obstruction when lying on the side.

[0100] This application addresses the problem of poor generalization ability of existing technologies for different sleep scenarios by using a two-dimensional adaptive mechanism of environmental parameters and individual characteristics, such as adjusting posture calculation model parameters according to user body shape.

[0101] This application automatically adjusts the human body size parameters (such as the chest cavity thickness compensation coefficient) in the posture calculation model for users of different body types, so that the monitoring error of users with a height of 150-200cm and a weight of 40-120kg remains consistent (error fluctuation ≤5%), solving the monitoring bias problem of obese / thin users in the existing technology; it supports dynamic bed layout, and when the sleeping positions of two people are offset (such as sleeping on one side / in the center), the target coordinates are updated in real time through multi-sensor TDOA positioning, and the generalization of the algorithm is improved by 30% (compared with the fixed position calibration scheme).

[0102] In the aforementioned dual-person sleep monitoring method, the acquired mixed multimodal signals from both individuals are converted into a time-frequency graph matrix. Spatial features of body movement amplitude and posture angle changes are extracted from this matrix, and the temporal sequence features of respiratory signals are captured. This decouples the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual. This method overcomes the limitations of traditional Fourier transform algorithms, achieving precise separation of dual-person sleep multimodal signals and resolving signal cross-interference issues. Through multi-algorithm collaborative processing, high-precision extraction of key parameters such as respiratory rate and body movement during sleep is achieved, reducing the monitoring error rate to below 3%, meeting the stringent requirements of the medical and health field. The method of analyzing each individual's sleep quality and generating individual sleep monitoring reports can adapt to different individual body types, sleep habits, and spatial location variations, improving the algorithm's generalization ability and effectively reducing monitoring blind spots and the risk of misjudgment.

[0103] In one embodiment, such as Figure 2 As shown, a dual-person sleep monitoring system 10 is provided, including: a millimeter-wave radar sensor 1, a multimodal signal decoupling module 2, a sleep parameter extraction module 3, and a data processing terminal 4.

[0104] The millimeter-wave radar sensor 1 is used to transmit millimeter-wave signals to the sleeping area of ​​two people, receive echo signals reflected by the human body, and acquire a mixed multimodal signal of two people containing information on breathing, body movement, and posture changes of two people.

[0105] The multimodal signal decoupling module 2 is used to convert the dual-person mixed multimodal signal into a time-frequency graph matrix, extract the spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix, capture the time series features of the respiratory signal, and decouple to obtain the respiratory signal sequence, body motion signal sequence and posture signal sequence of each person.

[0106] The sleep parameter extraction module 3 is used to obtain the respiratory rate based on the respiratory signal sequence, obtain the body movement parameters based on the body movement signal sequence, and identify the sleep posture based on the posture signal sequence.

[0107] The data processing terminal 4 is used to analyze the sleep quality of each person based on the respiratory rate, the body movement parameters, and the sleep posture, and to generate a sleep monitoring report for each person.

[0108] The dual-person sleep monitoring system 10 includes at least two 24GHz millimeter-wave radar sensors (such as Infineon BGT24MTR11), each sensor integrating a 4-8 channel planar antenna array (antenna spacing ≤λ / 2, where λ is the wavelength corresponding to 24GHz, 12.5mm), forming a fan-shaped beam coverage area.

[0109] The radar sensors are symmetrically arranged below the edge of the mattress. Specifically, one sensor is arranged on each side of the head of the mattress (40-45cm from the side edge of the bed and 70-80cm from the head of the bed) to cover the micro-movements of the upper body's breathing and chest cavity; and one sensor (optional) is arranged on each side of the foot of the mattress (20-25cm from the foot edge of the bed). The antenna array is horizontally pointed towards the sleeping area to cover the lower limb movement signals, forming a dual-view monitoring system.

[0110] The sensor achieves multi-source signal acquisition synchronization through a time synchronization module (accuracy ≤10ns), and establishes a unified coordinate system (with the geometric center of the mattress as the origin, and the coordinate system error ≤5mm) through a spatial calibration module.

[0111] The horizontal beamwidth (HPBW) of a single sensor antenna array is 60°-90° and the vertical beamwidth is 40°-60°. The tilt design allows the beam intersection to cover the entire sleeping area (200cm long × 180cm wide), eliminating blind spots at the head / foot of the bed.

[0112] When fusing signals from multiple sensors, spatial diversity reception technology is used: by utilizing the time difference of arrival (TDOA) and phase difference (PDOA) of echo signals from different sensors, the three-dimensional position coordinates (x, y, z) of the two-person target are constructed with a positioning accuracy of ≤10cm, which solves the signal attenuation problem caused by human body occlusion in traditional single sensors.

[0113] The sensors are isolated from each other by a metal shield (thickness ≥ 0.5 mm), with the shield height exceeding the sensor antenna array by 2-3 cm to suppress electromagnetic crosstalk between radars. Each sensor integrates an independent bandpass filter (passband 23.6-24.4 GHz) to suppress crosstalk noise between radars to below -40 dB, reducing co-channel interference by 80% compared to traditional unshielded solutions. Combined with a digital domain crosstalk cancellation algorithm, co-channel interference between multiple sensors is eliminated.

[0114] The 24GHz radar chip has a power consumption of ≤50mW / unit. With the time-division multiplexing acquisition mode (single sensor working cycle ≤20ms), the power consumption of the entire monitoring system is ≤200mW. It supports lithium battery power supply for portable devices (such as sleep monitoring mattresses) (battery life ≥8 hours), meeting the low power consumption requirements of home scenarios.

[0115] The multi-radar sensor collaborative monitoring system uses 24GHz radar sensors (such as the Infineon BGT24MTR11) symmetrically arranged below the edge of the mattress (10-15cm from the headboard on both sides, with an antenna tilt angle of 15°-30°), compared to traditional top-mounted radar solutions:

[0116] No blind spots in spatial coverage: Through the dual-view beam design (upper body monitoring + lower body motion coverage), the obstruction area at the head / foot of the bed is eliminated, and the effective signal coverage area is increased from 1.2㎡ to 1.8㎡ (covering a standard double bed area of ​​200cm×180cm);

[0117] Multi-source signal synchronization: By using a time synchronization module with an accuracy of ≤10ns and pulse triggering technology, the time deviation of the echo signals from multiple sensors is ensured to be ≤5ns, providing a spatiotemporally consistent data source for subsequent signal decoupling and avoiding phase calculation errors (error ≤0.5°) caused by traditional asynchronous acquisition.

[0118] The sensor placement parameters (spacing 10-15cm, tilt angle 15°-30°) and antenna array design (4-8 channels, spacing ≤λ / 2) are all based on the standardized design of electromagnetic characteristics in the 24GHz band, and are compatible with mainstream commercial radar chips (such as TI IWR1443, NXPTEF82xx), reducing mass production costs by more than 30%.

[0119] The signal processing flow (preprocessing → decoupling → anti-interference → parameter extraction) can be implemented in real time on an embedded DSP (such as TITMS320C6748), with a computational delay of ≤50ms, meeting the stringent requirements for real-time monitoring of medical devices (industry standard delay ≤100ms).

[0120] Table 1 Sensor Hardware Configuration

[0121]

[0122]

[0123] Signal acquisition: The radar sensor transmits FMCW signals at a sampling rate of 100Hz (sweep bandwidth 200MHz, sweep period 5ms), and acquires 4 channels × 2048 points of raw echo data in each cycle.

[0124] Bandpass filtering: A 5th-order Butterworth filter (passband 0.05-10Hz) is used to filter out power frequency (50Hz) and high-frequency environmental noise.

[0125] Adaptive gain: The PGA2311 programmable gain amplifier is used to dynamically adjust the gain (range 0-60dB) according to the signal amplitude, so that the output signal amplitude is stabilized at 1-2Vpp.

[0126] Taking the monitoring of a double bed as an example, the sensor is calibrated with a laser rangefinder, with the distance from the edge of the bed headboard 70cm±0.5cm and the tilt angle 20°±2° to ensure that the beam intersection covers the chest cavity area (error ≤5cm); the trigger signal of multiple sensors is measured with an oscilloscope, and the time deviation is ≤9ns, which meets the requirements of signal decoupling for spatiotemporal synchronization.

[0127] Extreme scenario testing

[0128] Electromagnetic interference (WiFi fully enabled): The bit error rate of this application is 3.1%, while the bit error rate of the traditional solution is 18.7%;

[0129] Side-lying curled-up position: The effective respiratory signal acquisition rate of this application was 84%, while that of the traditional method was only 55%;

[0130] Low power consumption test: The total power consumption of the system is 180mW (2 main sensors + FPGA processing), and the lithium battery (3.7V / 2000mAh) provides 8.2 hours of battery life.

[0131] A comparative experimental example was conducted using a commercially available single-radar sleep monitoring device (model A) to compare with the system described in this application:

[0132] Decoupling performance: This application uses a 24GHz dual sensor + CNN-LSTM decoupling, and the separation degree (cross-correlation coefficient) of the two people's breathing signals is ≤0.15; Device A uses a single sensor + FFT algorithm, with a separation degree ≥0.42, and there is significant signal crosstalk.

[0133] Interference immunity: Under an electromagnetic interference intensity of 10V / m, the SNR of this application is improved by 16dB, while the ASNR of the device is only improved by 5dB;

[0134] Engineering costs: The cost of a single hardware unit in this application is RMB 280 (for a batch of 1000 units), and the cost of equipment A is RMB 350, which gives it a cost advantage in mass production.

[0135] The above embodiments, through clearly defined hardware parameters, algorithm flow, and experimental data, demonstrate the feasibility and technical advantages of the technical solution of this application, meeting the requirements of invention patents for embodiments to be "specific, repeatable, and supporting the claims." In practical applications, the sensor spacing (±2cm) can be adjusted according to different mattress sizes (1.5m / 1.8m beds), and the interference feature database can be optimized through OTA upgrades.

[0136] In this embodiment, the process of converting the dual-person mixed multimodal signal into a time-frequency graph matrix, extracting the spatial features of body motion amplitude and posture angle changes from the time-frequency graph matrix, capturing the time-series features of the respiratory signal, and decoupling to obtain the respiratory signal sequence, body motion signal sequence, and posture signal sequence for each individual includes:

[0137] When converting the dual-person hybrid multimodal signal into a time-frequency diagram matrix, the time-frequency domain and spatial array phase information of the signal are preserved, wherein the dimension of the time-frequency diagram matrix is ​​[time × frequency × number of antenna channels];

[0138] The time-frequency graph matrix is ​​input into a convolutional neural network to extract the spatial features of body motion amplitude and attitude angle changes in the time-frequency graph matrix. The convolutional neural network includes at least two convolutional layers, and the dimension of the features output by the convolutional neural network is [batch size × time step × number of feature channels].

[0139] The features output by the convolutional neural network are modeled using a bidirectional gated recurrent unit (Bi-GRU) to capture the periodic micro-motion features of the respiratory signal and obtain the time-series features of the respiratory signal.

[0140] The respiratory, body movement, and posture changes of each person in the dual-person mixed multimodal signal are separated and decoupled to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person.

[0141] In this embodiment, as Figure 2 As shown, the provided dual-person sleep monitoring system 10 also includes a signal preprocessing module 5.

[0142] The signal preprocessing module 5 is used to preprocess the dual-person hybrid multimodal signal, using a bandpass filter to remove electromagnetic interference in specific frequency bands in the environment, and using an adaptive gain amplifier to dynamically adjust the gain according to the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal.

[0143] In this embodiment, as Figure 2 As shown, the provided dual-person sleep monitoring system 10 also includes an adaptive anti-interference module 6.

[0144] The adaptive anti-interference module 6 is used to establish an interference signal feature database after decoupling and obtaining the breathing signal sequence, body movement signal sequence and posture signal sequence of each person. It compares the breathing signal sequence, body movement signal sequence and posture signal sequence of each person with the interference signal feature database, uses a support vector machine classifier to identify the interference type, and adopts the corresponding anti-interference strategy according to the identified interference type.

[0145] The step of establishing an interference signal feature database involves comparing each person's respiratory signal sequence, body movement signal sequence, and posture signal sequence with the interference signal feature database, using a support vector machine classifier to identify the interference type, and adopting corresponding anti-interference strategies based on the identified interference type.

[0146] Collect signal characteristics of electromagnetic interference and background noise, and establish an interference signal characteristic database;

[0147] The respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person are compared with the interference signal feature database. The interference type is identified using a support vector machine (SVM) classifier. The interference types include narrowband interference, broadband interference, human body occlusion clutter, and periodic body movement artifacts.

[0148] When the identified interference type is narrowband interference, a tunable notch filter is used for frequency suppression, and the center frequency is dynamically adjusted according to the interference frequency.

[0149] When the identified interference type is broadband interference, wavelet threshold denoising algorithm is used for wavelet packet denoising, and the number of decomposition layers is adaptively selected based on noise energy.

[0150] When the identified interference type is human body obstruction clutter, the spatial spectrum estimation algorithm is used to reconstruct the target position using the phase difference of a multi-antenna array;

[0151] Based on real-time data from environmental sensors (electromagnetic intensity sensor, temperature and humidity sensor), the parameters of the anti-interference algorithm are dynamically adjusted to achieve adaptive suppression of interference signals.

[0152] In this embodiment, the step of collecting signal characteristics of electromagnetic interference and background noise and establishing an interference signal characteristic database includes:

[0153] Pre-collect various typical interference signals, including narrowband electromagnetic interference, broadband noise, and human body obstruction clutter.

[0154] Extract the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signal. The characteristic parameters include center frequency, energy entropy, and wavelet coefficient variance.

[0155] By pre-collecting signal characteristics of various background noises and combining them with the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signals, an interference signal feature database is established.

[0156] In this embodiment, obtaining the respiratory rate based on the respiratory signal sequence, obtaining the body movement parameters based on the body movement signal sequence, and identifying the sleep posture based on the posture signal sequence includes:

[0157] The respiratory signal sequence was processed by Hilbert-Huang transform (HHT), noise modes were separated by empirical mode decomposition (EMD), and intrinsic mode function (IMF) components containing the respiratory frequency were selected for Hilbert spectral transform to extract the instantaneous frequency of the respiratory signal to form the respiratory frequency.

[0158] The amplitude change of the body movement signal sequence is detected by amplitude threshold, the time interval of the amplitude change is detected by time interval threshold, and the strong body movement and weak body movement are distinguished by the slope change of the time domain waveform of the body movement signal. The number of body movements and the intensity of body movements are counted as body movement parameters.

[0159] A three-dimensional attitude angle calculation model is constructed based on the phase difference of the echo signal from a multi-antenna array. The mapping relationship between attitude angle and signal features is fitted by the support vector regression (SVR) algorithm. The attitude signal sequence is input into the three-dimensional attitude angle calculation model to identify attitude angle features by pitch angle, roll angle, and azimuth angle. The corresponding sleep posture is identified based on the attitude angle features. The sleep posture includes supine, lateral, and prone positions.

[0160] In the aforementioned dual-person sleep monitoring system, the acquired mixed multimodal signals from both individuals are converted into a time-frequency graph matrix. Spatial features of body movement amplitude and posture angle changes are extracted from this matrix, and the temporal sequence features of respiratory signals are captured. This decouples the respiratory, body movement, and posture signal sequences for each individual, overcoming the limitations of traditional Fourier transform algorithms. This achieves precise separation of dual-person sleep multimodal signals, resolves signal cross-interference issues, and through multi-algorithm collaborative processing, achieves high-precision extraction of key parameters such as respiratory rate and body movement during sleep. The monitoring error rate is reduced to less than 3%, meeting the stringent requirements of the medical and health field. The system analyzes each individual's sleep quality and generates individual sleep monitoring reports, adapting to variations in body shape, sleep habits, and spatial location, improving algorithm generalization ability, and effectively reducing monitoring blind spots and the risk of misjudgment.

[0161] Specific limitations regarding the dual-person sleep monitoring system can be found in the limitations of the dual-person sleep monitoring method described above, and will not be repeated here. Each module in the aforementioned dual-person sleep monitoring system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0162] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores dual-person sleep monitoring data. The network interface communicates with external terminals via a network connection. When the processor executes the computer program, it implements a dual-person sleep monitoring method.

[0163] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0164] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0165] Millimeter wave signals are transmitted to the sleeping area of ​​two people, and echo signals reflected by the human body are received to obtain a mixed multimodal signal of two people containing information on breathing, body movement and posture changes of two people;

[0166] The dual-person mixed multimodal signal is converted into a time-frequency graph matrix. The spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix are extracted. The time-series features of the respiratory signal are captured and obtained. The respiratory signal sequence, body motion signal sequence and posture signal sequence of each person are decoupled.

[0167] The respiratory rate is obtained from the respiratory signal sequence, the body movement parameters are obtained from the body movement signal sequence, and the sleep posture is identified from the posture signal sequence.

[0168] The sleep quality of each person is analyzed based on the respiratory rate, body movement parameters, and sleep posture, and a sleep monitoring report for each person is generated.

[0169] For specific limitations on the steps a processor takes when executing a computer program, please refer to the limitations on the method for dual-person sleep monitoring mentioned above, which will not be repeated here.

[0170] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0171] Millimeter wave signals are transmitted to the sleeping area of ​​two people, and echo signals reflected by the human body are received to obtain a mixed multimodal signal of two people containing information on breathing, body movement and posture changes of two people;

[0172] The dual-person mixed multimodal signal is converted into a time-frequency graph matrix. The spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix are extracted. The time-series features of the respiratory signal are captured and obtained. The respiratory signal sequence, body motion signal sequence and posture signal sequence of each person are decoupled.

[0173] The respiratory rate is obtained from the respiratory signal sequence, the body movement parameters are obtained from the body movement signal sequence, and the sleep posture is identified from the posture signal sequence.

[0174] The sleep quality of each person is analyzed based on the respiratory rate, body movement parameters, and sleep posture, and a sleep monitoring report for each person is generated.

[0175] For specific limitations on the steps implemented when a computer program is executed by a processor, please refer to the limitations on the method of dual-person sleep monitoring mentioned above, which will not be repeated here.

[0176] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0177] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0178] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for dual-person sleep monitoring, characterized in that, include: Millimeter wave signals are transmitted to the sleeping area of ​​two people, and echo signals reflected by the human body are received to obtain a mixed multimodal signal of two people containing information on breathing, body movement and posture changes of two people; The dual-person mixed multimodal signal is converted into a time-frequency graph matrix. The spatial features of body motion amplitude and posture angle changes in the time-frequency graph matrix are extracted. The time-series features of the respiratory signal are captured and obtained. The respiratory signal sequence, body motion signal sequence and posture signal sequence of each person are decoupled. The respiratory rate is obtained from the respiratory signal sequence, the body movement parameters are obtained from the body movement signal sequence, and the sleep posture is identified from the posture signal sequence. Based on the respiratory rate, body movement parameters, and sleep posture, the sleep quality of each individual is analyzed, and a sleep monitoring report is generated for each individual. The dual-person mixed multimodal signal is converted into a time-frequency plot matrix. Spatial features of body motion amplitude and posture angle changes are extracted from the time-frequency plot matrix. The time-series features of the respiratory signal are captured, and the respiratory signal sequence, body motion signal sequence, and posture signal sequence of each person are decoupled and obtained, including: When converting the dual-person hybrid multimodal signal into a time-frequency diagram matrix, the time-frequency domain and spatial array phase information of the signal are preserved, wherein the dimension of the time-frequency diagram matrix is ​​[time × frequency × number of antenna channels]; The time-frequency graph matrix is ​​input into a convolutional neural network to extract the spatial features of body motion amplitude and attitude angle changes in the time-frequency graph matrix. The convolutional neural network includes at least two convolutional layers, and the dimension of the features output by the convolutional neural network is [batch size × time step × number of feature channels]. By using a bidirectional gated recurrent unit to model the temporal features of the features output by the convolutional neural network, the periodic micro-motion features of the respiratory signal are captured, and the time-series features of the respiratory signal are obtained. The respiratory, body movement, and posture changes of each individual in the dual-person mixed multimodal signal are separated and decoupled to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence of each individual. After decoupling to obtain the respiratory signal sequence, body movement signal sequence, and posture signal sequence for each individual, the following is also included: Collect signal characteristics of electromagnetic interference and background noise, and establish an interference signal characteristic database; The respiratory signal sequence, body movement signal sequence, and posture signal sequence of each person are compared with the interference signal feature database. The interference type is identified using a support vector machine classifier. The interference types include narrowband interference, broadband interference, human body occlusion clutter, and periodic body movement artifacts. When the identified interference type is narrowband interference, a tunable notch filter is used for frequency suppression, and the center frequency is dynamically adjusted according to the interference frequency. When the identified interference type is broadband interference, wavelet threshold denoising algorithm is used for wavelet packet denoising, and the number of decomposition layers is adaptively selected based on noise energy. When the identified interference type is human body obstruction clutter, the spatial spectrum estimation algorithm is used to reconstruct the target position using the phase difference of a multi-antenna array; Based on real-time data from environmental sensors, the parameters of the anti-interference algorithm are dynamically adjusted to achieve adaptive suppression of interference signals.

2. The dual-person sleep monitoring method according to claim 1, characterized in that, After acquiring the dual-person hybrid multimodal signal, the following is also included: The dual-person hybrid multimodal signal is preprocessed by using a bandpass filter to remove electromagnetic interference in specific frequency bands of the environment and using an adaptive gain amplifier to dynamically adjust the gain according to the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal.

3. The dual-person sleep monitoring method according to claim 1, characterized in that, The process of collecting signal characteristics of electromagnetic interference and background noise, and establishing an interference signal characteristic database, includes: Pre-collect various typical interference signals, including narrowband electromagnetic interference, broadband noise, and human body obstruction clutter. Extract the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signal. The characteristic parameters include center frequency, energy entropy, and wavelet coefficient variance. By pre-collecting signal characteristics of various background noises and combining them with the time-domain peak value, frequency-domain energy, time-frequency domain entropy value, and characteristic parameters of the typical interference signals, an interference signal feature database is established.

4. The dual-person sleep monitoring method according to claim 1, characterized in that, The steps of obtaining respiratory rate based on the respiratory signal sequence, obtaining body movement parameters based on the body movement signal sequence, and identifying sleep posture based on the posture signal sequence include: The respiratory signal sequence is processed by Hilbert-Huang transform, noise modes are separated by empirical mode decomposition, and the intrinsic mode function components containing the respiratory frequency are selected for Hilbert spectral transform to extract the instantaneous frequency of the respiratory signal to form the respiratory frequency. The amplitude change of the body movement signal sequence is detected by amplitude threshold, the time interval of the amplitude change is detected by time interval threshold, and the strong body movement and weak body movement are distinguished by the slope change of the time domain waveform of the body movement signal. The number of body movements and the intensity of body movements are counted as body movement parameters. A three-dimensional attitude angle calculation model is constructed based on the phase difference of the echo signal from a multi-antenna array. The mapping relationship between attitude angle and signal features is fitted by a support vector regression algorithm. The attitude signal sequence is input into the three-dimensional attitude angle calculation model to identify attitude angle features by pitch, roll, and azimuth. The corresponding sleep posture is identified based on the attitude angle features. The sleep posture includes supine, lateral, and prone positions.

5. A dual-person sleep monitoring system for performing the dual-person sleep monitoring method according to any one of claims 1-4, characterized in that, The dual-person sleep monitoring system includes: Millimeter-wave radar sensors are used to transmit millimeter-wave signals into the sleeping area of ​​two people, receive echo signals reflected by the human body, and acquire mixed multimodal signals of two people containing information on breathing, body movement, and posture changes of two people. The multimodal signal decoupling module is used to convert the dual-person mixed multimodal signal into a time-frequency diagram matrix, extract the spatial features of body motion amplitude and posture angle changes in the time-frequency diagram matrix, capture the time series features of the respiratory signal, and decouple to obtain the respiratory signal sequence, body motion signal sequence and posture signal sequence of each person. The sleep parameter extraction module is used to obtain the respiratory rate based on the respiratory signal sequence, obtain the body movement parameters based on the body movement signal sequence, and identify the sleep posture based on the posture signal sequence. The data processing terminal is used to analyze the sleep quality of each person based on the respiratory rate, the body movement parameters, and the sleep posture, and to generate a sleep monitoring report for each person.

6. The dual-person sleep monitoring system according to claim 5, characterized in that, The dual-person sleep monitoring system also includes: The signal preprocessing module is used to preprocess the dual-person hybrid multimodal signal, using a bandpass filter to remove electromagnetic interference in specific frequency bands of the environment, and using an adaptive gain amplifier to dynamically adjust the gain according to the signal strength in order to improve the signal-to-noise ratio of the dual-person hybrid multimodal signal. An adaptive anti-interference module is used to establish an interference signal feature database. It compares each person's respiratory signal sequence, body movement signal sequence, and posture signal sequence with the interference signal feature database, uses a support vector machine classifier to identify the interference type, and adopts the corresponding anti-interference strategy according to the identified interference type.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images