A wearable thoracic monitoring band for physiological signal monitoring

Abstract

This invention relates to the field of wearable medical monitoring technology, specifically a wearable chest monitoring belt for monitoring physiological signals. The belt includes: synchronously extracting tension and respiratory phase data and performing temporal registration to generate a tension quantization matrix with belt phase markings; identifying and clustering key tension peaks based on this matrix to establish a local deformation mapping template; continuously matching real-time data using the template to dynamically construct and iteratively update a spatiotemporal multidimensional tensor field model of the thoracic cavity; decoupling the background component and respiratory activity component from this model, inputting the activity component into a lung function state inference network, and generating lung function simulation parameters frame by frame; integrating the above information to synthesize a multimodal thoracic cavity physiological state description map; and adaptively controlling the data acquisition and processing link based on this map. This invention achieves dynamic spatiotemporal modeling of thoracic cavity respiratory deformation and continuous inference of lung function parameters.

Description

Technical Field

[0001] This invention relates to the field of wearable medical monitoring technology, and more particularly to a wearable chest monitoring belt for monitoring physiological signals. Background Technology

[0002] Existing wearable chest monitoring technologies based on tension sensing primarily extract the temporal and amplitude characteristics of respiratory events by processing signal amplitude changes at single or multiple sensing points. These methods typically simplify chest deformation into a linear expansion and contraction model, either globally or locally, with data processing focused on filtering, peak detection, and envelope analysis of one-dimensional waveforms. Their limitation lies in their inability to describe the complex, asynchronous, and spatially correlated three-dimensional deformation dynamics of different regions of the chest surface during respiration. This results in the output signal sequence losing crucial spatiotemporal information about the distribution and propagation of deformation on the body surface, thus restricting its ability to characterize respiratory mechanics details.

[0003] Another limitation of existing technologies is the simplistic approach to deriving physiological parameters. They typically employ fixed conversion formulas based on amplitude calibration or superficial linear regression models, attempting to directly correlate raw tension signals with parameters such as lung volume. This method struggles to effectively isolate baseline tension drift introduced by factors such as changes in body position and variations in monitoring strap tightness, and it fails to construct a nonlinear, personalized mapping relationship between multidimensional deformation patterns on the body surface and the functional state of the lungs. Therefore, existing technologies cannot achieve continuous, robust, and quantitative inference from surface signals to deep pulmonary function parameters. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a wearable chest monitoring belt for monitoring physiological signals.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a wearable chest monitoring belt for physiological signal monitoring, comprising: Tension data units and respiratory phase data units are extracted synchronously from the raw data stream of the monitoring belt. The tension data units and the corresponding respiratory phase data units are time-domain registered to generate a tension quantization matrix with phase labels. Based on the tension quantization matrix with phase label, key tension peaks in respiratory events are identified and their morphological features are extracted. Based on the differences in morphological features, the key tension peaks are divided into several feature clusters, and a corresponding local deformation mapping template is established for each feature cluster. The local deformation mapping template is used to continuously match and map the real-time acquired thoracic tension data. Based on the result of feature matching, a multidimensional tensor field model reflecting the spatiotemporal changes of the thoracic cavity is dynamically constructed and iteratively updated. The background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation are decoupled from the multidimensional tensor field model. The active component is then input into a preset lung function state inference network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame. By integrating the baseline component, activity component, and lung function simulation parameters, a multimodal description map of the thoracic physiological state is synthesized. Based on the multimodal thoracic physiological state description map, the data acquisition mode and signal processing link of the thoracic monitoring belt are adaptively closed-loop controlled.

[0006] As a further aspect of the present invention, the step of simultaneously extracting tension data units and respiratory phase data units from the raw data stream of the monitoring belt includes: Deploy a parallel signal listening interface to capture the raw signal stream output from the sensor array of the monitoring band at a fixed sampling period; A high-frequency filtering channel and a low-frequency filtering channel are set after the parallel signal listening interface to separate tension signals and respiratory-related signals, respectively. Peak detection and baseline calibration are performed on the output of the high-frequency filter channel to obtain the calibrated tension data unit sequence; Zero-crossing detection and envelope extraction are performed on the output of the low-frequency filter channel to generate a sequence of respiratory phase data units characterizing the respiratory cycle; Under the control of the system clock, a unified timestamp is applied to each tension data unit and respiratory phase data unit; Establish a time-stamp alignment buffer, and store the tension data unit sequence and the respiratory phase data unit sequence in pairs according to the unified timestamp.

[0007] As a further aspect of the present invention, the step of performing time-domain registration of the tension data unit with the corresponding respiratory phase data unit to generate a tension quantization matrix with phase labels includes: Read the paired tension data units and their associated respiratory phase data units from the time-aligned buffer; Analyze the respiratory phase data unit to determine the respiratory phase category to which the current tension data unit belongs. The respiratory phase category includes early inspiratory phase, late inspiratory phase, early expiratory phase, and late expiratory phase. The tension data unit is normalized and quantized to map its amplitude to a preset quantization level range; The quantized tension amplitude is combined with its corresponding respiratory phase category label to form a tension quantization data point with phase label; The continuous tension quantization data points with phase markers are arranged in chronological order to form a two-dimensional tension quantization matrix; The two-dimensional tension quantization matrix is ​​filled with null values ​​and smoothed to generate a complete tension quantization matrix with phase labels.

[0008] As a further aspect of the present invention, the step of identifying key tension peaks in respiratory events and extracting their morphological features based on the phase-labeled tension quantization matrix includes: In the tension quantization matrix with phase labels, the search interval is from the end of expiration to the end of inspiration; A sliding window strategy is used within the search interval to detect local maximum points in the tension amplitude sequence; Threshold filtering is performed on the detected local maximum points to remove points whose amplitude is lower than the resting tension threshold, and the remaining points are marked as candidate tension peaks; For each candidate tension peak, a predetermined number of adjacent data points are extracted to form a tension peak data segment. Extract a set of morphological features from each tension peak data segment, including peak width, left slope, right slope, peak curvature, and peak area integral. All candidate tension peaks and their corresponding morphological feature sets are stored as a key tension peak feature table.

[0009] As a further aspect of the present invention, the step of dividing the key tension peaks into several feature clusters based on morphological feature differences and establishing a corresponding local deformation mapping template for each feature cluster includes: Read the set of morphological features of all key tension peaks from the key tension peak feature table; The morphological feature set was analyzed using an unsupervised clustering method, and the key tension peaks were divided into multiple non-overlapping feature clusters based on feature similarity. Calculate the average value of all tension peak morphological features within each feature cluster, and use it as the cluster center feature vector of the feature cluster; For each feature cluster, trace back the original spatial location information of the tension peak within the cluster on the chest monitoring band; Based on the spatial location information and the cluster center feature vector, a local deformation mapping template describing the typical tension distribution pattern of the feature cluster is constructed; The local deformation mapping template is stored in the form of a data grid, and the grid node value represents the probability weight of the corresponding feature cluster deformation occurring in a spatial location.

[0010] As a further aspect of the present invention, the step of using a local deformation mapping template to perform continuous feature matching and mapping on the real-time acquired thoracic tension data includes: A snapshot of the chest tension distribution at the current moment is obtained from the sensor array of the chest monitoring belt; Extract feature values ​​corresponding to the morphological characteristics of key tension peaks from the snapshot of the thoracic tension distribution; The similarity between the extracted feature values ​​and the cluster center feature vectors of all feature clusters is calculated. Identify the feature clusters with the highest similarity and call their corresponding local deformation mapping templates; The data from the snapshot of the thoracic tension distribution is spatially convolved with the invoked local deformation mapping template. The convolution result is output to obtain a probability distribution map that highlights the deformation region that best matches the current tension pattern.

[0011] As a further aspect of the present invention, the step of dynamically constructing and iteratively updating a multidimensional tensor field model reflecting the spatiotemporal changes of the thoracic cavity based on the feature matching results includes: The probability distribution map is used as the input stimulus for model updating; A three-dimensional spatial grid is set to cover the entire chest cavity monitoring area, and time is used as the fourth dimension; At each node of the four-dimensional grid, a tensor is initialized to record the tension mean, variance, and correlation coefficient with adjacent nodes at the spatiotemporal points. Map the probability distribution data at the current moment to a three-dimensional spatial grid according to spatial location, and update the tension mean data of the corresponding spatial nodes; Based on the current data and previous historical data, recalculate the tension variance of each node and its correlation coefficient with adjacent nodes; The tensor data of all nodes are smoothed and filtered in the time dimension to complete one iterative update of the multidimensional tensor field model. The process of continuously receiving new probability distribution maps and repeating the update process enables the multidimensional tensor field model to evolve dynamically.

[0012] As a further aspect of the present invention, the step of decoupling the background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation from the multidimensional tensor field model includes: Spatiotemporal frequency domain decomposition is performed on the iteratively updated multidimensional tensor field model; Low-pass filtering was used to extract signal components in the multidimensional tensor field model whose variation period exceeded the length of a single respiratory cycle, and these components were defined as the background components representing the basic tension of the thoracic cavity. Bandpass filtering was used to extract the signal components in the multidimensional tensor field model whose change period matched the length of a typical respiratory cycle, and these components were defined as active components representing respiratory deformation. The background component and the active component are strictly separated in the frequency domain to ensure that the background component and the active component are orthogonal. The static distribution map and dynamic change sequence of the background component and the active component on the three-dimensional spatial grid are reconstructed respectively.

[0013] As a further aspect of the present invention, the step of inputting the activity component into a preset lung function state inference network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame includes: Construct a lung function state inference network whose input layer dimension matches the data dimension of the activity components; The lung function state inference network contains multiple hidden layers for learning high-order nonlinear mappings between spatiotemporal patterns of activity components and lung function parameters. The dynamic change sequence of the active components representing respiratory deformation is input into the trained lung function state inference network in time frame order. The lung function state inference network processes each frame of activity component data and outputs a set of lung function simulation parameters in real time. The simulated lung function parameters include at least simulated tidal volume, simulated lung volume change rate, and simulated airway resistance index.

[0014] As a further aspect of the present invention, the adaptive closed-loop control of the data acquisition mode and signal processing link of the thoracic monitoring belt based on the multimodal thoracic physiological state description map includes: The static distribution map of the background component, the dynamic change sequence of the active component, and the time series data of the lung function simulation parameters are fused in a unified spatiotemporal coordinate system to generate a multimodal thoracic physiological state description map. Real-time analysis of the multimodal thoracic physiological state description map to monitor the occurrence of specific physiological events or state transitions; When a predefined specific physiological event is detected, a mode switching command is generated to adjust the sampling frequency or working mode of the thoracic monitoring belt sensor array; Meanwhile, based on the continuous changing trend of lung function simulation parameters, the cutoff frequency of the filter or the sensitivity threshold of feature extraction in the signal processing link is dynamically adjusted. The adjusted data acquisition and processing results are fed back into the generation process of the multimodal thoracic physiological state description map, forming a continuously optimized adaptive closed loop.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By constructing a phase-labeled tension quantization matrix from synchronously registered tension and respiratory phase data, and further identifying and clustering the morphological features of key tension peaks to establish local deformation mapping templates, this method achieves spatiotemporal correlation encoding of discrete tension signals on the thoracic surface. Using these templates, continuous feature matching and mapping of real-time data is performed, and a multidimensional tension field model is dynamically constructed and iteratively updated accordingly. This transforms a traditional one-dimensional signal sequence into a spatiotemporally continuous field reflecting the synergistic and evolving tension patterns at various points on the thoracic surface. This model decouples the background component representing static or slowly changing states from the purely respiratory-driven activity component, thereby separating motion artifacts from core physiological signals and providing a high-fidelity deformation data foundation for subsequent analysis.

[0016] The thoracic respiratory deformation components, precisely decoupled from the multidimensional tensor field model, are used as input to a pre-defined lung function state inference network, driving the network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame. This process avoids interference from basal tension fluctuations and directly uncovers the complex correlation between the spatiotemporal patterns of respiratory deformation and lung biomechanical parameters. By learning this deep mapping relationship, the neural network can output continuous sequences of functional parameters such as simulated tidal volume and simulated airway resistance, realizing an indirect but continuous quantitative inference from surface biomechanical deformation information measured by wearable devices to the intrathoracic lung function state, which traditionally requires invasive or desktop equipment for assessment. Attached Figure Description

[0017] Figure 1 This is a flowchart of the wearable chest monitoring belt for physiological signal monitoring described in this invention; Figure 2 A flowchart for generating a tension quantization matrix with phase labels; Figure 3 A heatmap showing the correlation between respiratory phase and morphological features; Figure 4 The time-domain comparison verification curve between the reconstructed signal of the active component and the original tension signal; Figure 5 Pearson correlation heatmap of lung function simulation parameters and thoracic tension parameters. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 The data processing unit synchronously extracts tension data units and respiratory phase data units from the raw data stream of the monitoring belt, and performs temporal registration between the tension data units and the corresponding respiratory phase data units to generate a tension quantization matrix with phase labels. Based on this matrix, key tension peaks in respiratory events are identified and their morphological features are extracted. Key tension peaks are divided into several feature clusters according to differences in morphological features, and a corresponding local deformation mapping template is established for each feature cluster. The local deformation mapping template is used to continuously match and map the real-time acquired thoracic tension data. Based on the feature matching results, a multidimensional tensor field model reflecting the spatiotemporal changes of the thoracic cavity is dynamically constructed and iteratively updated. The background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation are decoupled from the multidimensional tensor field model. The active component is input into a preset lung function state inference network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame. The background component, active component, and lung function simulation parameters are integrated to synthesize a multimodal thoracic cavity physiological state description map. Based on the multimodal thoracic physiological state description map, the data acquisition mode and signal processing link of the thoracic monitoring belt are adaptively closed-loop controlled.

[0021] In one embodiment of the present invention, see [reference] Figure 2A wearable chest monitoring belt for physiological signal monitoring deploys a parallel signal listening interface in its data processing unit. This interface captures the raw signal stream output from the monitoring belt's sensor array at a fixed sampling period. The raw signal stream contains mixed physiological signals simultaneously acquired by multiple sensing nodes. In some embodiments, the fixed sampling period is set to 1000 Hz, and the raw signal stream is transmitted in real time as an array, with each array element corresponding to the raw voltage value of a sensing node at a sampling moment. A high-frequency filtering channel and a low-frequency filtering channel are provided after the parallel signal listening interface. The high-frequency filtering channel uses a high-pass filter with a cutoff frequency of 10 Hz to separate tension signals reflecting chest wall muscle tension and movement from the mixed signal. The low-frequency filtering channel uses a low-pass filter with a cutoff frequency of 0.5 Hz to separate respiratory-related signals reflecting periodic changes in chest cavity volume. Peak detection and baseline calibration are performed on the output of the high-frequency filter channel. Peak detection uses a sliding window comparison method to identify local amplitude maxima. Baseline calibration is achieved by calculating the moving average of the signal from each sensor node over a period of time and subtracting the moving average baseline from the original signal to obtain a calibrated tension data unit sequence. Each tension data unit in the sequence contains the calibrated tension amplitude of all sensor nodes within a time segment. Zero-crossing detection and envelope extraction are performed on the output of the low-frequency filter channel. Zero-crossing detection is used to locate the phase reversal point of the respiratory waveform, and envelope extraction uses Hilbert transform to obtain the instantaneous amplitude of the respiratory signal, thereby generating a respiratory phase data unit sequence representing the respiratory cycle. Each respiratory phase data unit contains the main respiratory phase identifier corresponding to a time segment.

[0022] Under system clock control, each tension data unit and respiratory phase data unit is assigned a unified timestamp with millisecond-level precision, ensuring comparability of data from different processing channels on the timeline. A time-stamp alignment buffer is established, a circular data buffer based on timestamp indexing, which pairs and stores tension data unit sequences with respiratory phase data unit sequences according to the unified timestamp. The pairing logic associates tension data units and respiratory phase data units with timestamp differences within a preset tolerance range as a data pair. For example, a tension data unit with timestamp T will be paired with the respiratory phase data unit with the closest timestamp to T, and both will be stored in a storage unit of the time-stamp alignment buffer. It can be understood that the design of the time-stamp alignment buffer effectively handles minor delays caused by different signal processing paths, achieving precise synchronization between tension data and respiratory phase data.

[0023] The paired tension data units and their associated respiratory phase data units are read from the time-stamp alignment buffer. Pairing means that each tension data unit has a clearly defined corresponding respiratory phase information. The respiratory phase data units are parsed to determine the respiratory phase category to which the current tension data unit belongs. Respiratory phase categories include early inspiration, late inspiration, early expiration, and late expiration. The parsing process is based on logical judgments using the slope and amplitude of the respiratory signal envelope. For example, when the envelope slope is positive and the amplitude is less than 30% of the maximum amplitude, it is determined to be early inspiration; when the envelope slope is positive and the amplitude is greater than 70% of the maximum amplitude, it is determined to be late inspiration. The tension data units are then normalized and quantized, mapping their amplitude to a preset quantization level range, which is 0 to 255 levels. The mapping process is completed using a linear transformation formula: ; Where: Q represents the quantization level, and S represents the original tension amplitude. and These represent the preset minimum and maximum reference values ​​for tension amplitude, respectively. This indicates a rounding down operation. The quantized tension amplitude is combined with its corresponding respiratory phase category label to form a phase-labeled tension quantization data point. This data point is a structure containing a quantization level value and a respiratory phase category enumeration value. The consecutive phase-labeled tension quantization data points are arranged in chronological order to form a two-dimensional tension quantization matrix. The row indices of the two-dimensional tension quantization matrix correspond to the time series, and the column indices correspond to the sensor nodes at different spatial locations on the chest monitoring band. Each element in the matrix is ​​a phase-labeled tension quantization data point.

[0024] The two-dimensional tension quantization matrix undergoes null-filling and smoothing. Null-filling addresses matrix gaps caused by transient signal loss by using linear interpolation between adjacent time points and spatial nodes. Smoothing employs a time-dimensional moving average filter to suppress random noise, ultimately generating a complete tension quantization matrix with phase labels. In some embodiments, the window length of the moving average filter is set to the number of time data points corresponding to 0.1 seconds. The complete tension quantization matrix with phase labels forms the foundational data structure for subsequent respiratory event analysis and feature extraction. It can be understood that through the aforementioned time-domain registration and quantization process, the original, asynchronous tension and respiratory signals are integrated into a structured data matrix with a unified time reference, clear phase labels, and standardized amplitudes.

[0025] In one embodiment of the invention, the operation of identifying key tension peaks in a respiratory event and extracting their morphological features based on a complete phase-labeled tension quantization matrix is ​​carried out within a respiratory cycle. In the phase-labeled tension quantization matrix, a search interval is defined as the period from end-expiration to end-inspiration. This search interval is defined by locating data points with the respiratory phase category of end-expiration as the starting point and the immediately following data points with the end-inspiration as the ending point. The search interval covers the complete exertion phase from the end of one exhalation to the completion of the next inhalation. Within the search interval, a sliding window strategy is used to detect local maxima in the tension amplitude sequence. For the time-series data of each spatially located sensor node in the matrix, the sliding window moves along the time axis with a fixed length. In a specific implementation, the window length is set to the number of time data points corresponding to 0.05 seconds. When the amplitude of the center point of the window is greater than the amplitudes of all other points within the window, that center point is recorded as a local maxima.

[0026] Thresholding is applied to the detected local maximum points, removing points with amplitudes below the resting tension threshold. The resting tension threshold is a pre-set constant value (e.g., 150 quantization units) based on the 90th percentile of tension amplitudes statistically analyzed over a long period of time by the user in a calm breathing state. The remaining points are marked as candidate tension peaks. For each candidate tension peak, a tension peak data segment is formed by extracting a preset number of adjacent data points. The preset number of data points is typically 10 sampling points before and after the peak to ensure complete coverage of the rising and falling edges of the peak. A set of morphological features is extracted from each tension peak data segment. The extracted morphological features include peak width, left slope, right slope, peak curvature, and peak area integral. Peak width is defined as the full width at half the peak amplitude. The left slope is the slope of the linear fit of the rising edge on the left side of the peak, and the right slope is the slope of the linear fit of the falling edge on the right side of the peak. Peak curvature is approximated by calculating the second derivative of the data near the peak. Peak area integral is calculated by numerically integrating the area of ​​the amplitude within the tension peak data segment relative to the baseline using the trapezoidal rule. All candidate tension peaks and their corresponding morphological feature sets are stored as a key tension peak feature table. The key tension peak feature table is a structured database, and each record contains a timestamp, spatial location identifier, and a morphological feature vector consisting of five specific values.

[0027] The morphological feature sets of all key tension peaks are read from the key tension peak feature table. These morphological feature sets are organized in vector form. An unsupervised clustering method is used to analyze the morphological feature sets, dividing the key tension peaks into multiple non-overlapping feature clusters based on feature similarity. In specific implementation, the K-means clustering algorithm is used, with the number of clusters K set to 3. Feature similarity is measured by the Euclidean distance between feature vectors; the smaller the distance, the higher the similarity. The average value of the morphological features of all tension peaks within each feature cluster is calculated as the cluster center feature vector, representing the typical morphological feature pattern of that feature cluster. In essence, the clustering process categorizes a large number of discrete tension peaks based on their shape characteristics. For example, it might generate a cluster characterized by "rapid rise, narrow and high," and another cluster characterized by "slow rise, wide and flat."

[0028] For each feature cluster, the original spatial location information of its tension peaks on the chest monitoring band is traced back. This original spatial location information comes from the physical number of the sensor node and the preset geometric coordinate mapping relationship. Based on the spatial location information and the cluster center feature vector, a local deformation mapping template describing the typical tension distribution pattern of the feature cluster is constructed. The construction process first discretizes the chest surface covered by the chest monitoring band into a fine two-dimensional grid. For each node in the grid, the spatial distance from the node to all tension peaks belonging to the current feature cluster is calculated. The closer the tension peak is, the greater its influence weight on the node. The local deformation mapping template is stored in the form of a data grid. The grid node value represents the probability weight of the corresponding feature cluster deformation occurring at a spatial location. The probability weight is calculated according to a distance decay-based kernel function. ; in: Represents grid nodes The probability weight at each point, where N is the total number of tension peaks belonging to the current feature cluster. It is a grid node coordinates These are the spatial coordinates of the i-th tension peak belonging to this feature cluster. This is the Gaussian kernel width parameter that controls the range of influence. It can be understood that, through the above process, each feature cluster defined by morphological features is associated with a spatial probability distribution map. This distribution map is the local deformation mapping template, which quantifies the prior probability of a specific type of tension deformation occurring in different regions of the thoracic cavity. In some embodiments, the local deformation mapping templates of different feature clusters exhibit significant morphological differences. For example, the mapping template of one feature cluster may show high-weight regions concentrated around the sternum, while the mapping template of another feature cluster may show high-weight regions evenly distributed along the bilateral costal arches.

[0029] In one embodiment of the present invention, the process of continuously matching and mapping features of the real-time acquired thoracic tension data using a local deformation mapping template is performed cyclically. A snapshot of the thoracic tension distribution at the current moment is acquired in real time from the sensor array of the thoracic monitoring belt. The snapshot is a two-dimensional array, where each element corresponds to the tension amplitude of a sensing node at the latest sampling moment. Feature values ​​corresponding to the morphological features of key tension peaks are extracted from the snapshot. For each spatial location in the snapshot, the extraction operation calculates five morphological feature values—peak width, left slope, right slope, peak curvature, and peak area integral—within a local window centered on that location, thereby generating a multi-channel feature map with the same spatial resolution as the snapshot.

[0030] The extracted feature values ​​are compared with the cluster center feature vectors of all feature clusters using cosine similarity. For each point in space, the cosine value is calculated between its five-dimensional feature vector and the cluster center feature vector of each feature cluster. The feature cluster with the highest similarity is identified, and its corresponding local deformation mapping template is called. For each point in space, the one with the largest cosine similarity value from all feature clusters is selected, and this point is marked as belonging to that feature cluster. At the same time, the local deformation mapping template corresponding to that feature cluster, stored in a data grid, is loaded. Spatial convolution is performed between the data from the snapshot of the thoracic tension distribution and the called local deformation mapping template. For all spatial points marked as belonging to a specific feature cluster at the current time, their tension amplitude data is convolved with the probability weights of the corresponding local deformation mapping template in the neighborhood of that point. The convolution result is output, resulting in a probability distribution map that highlights the deformation region that best matches the current tension pattern. The brightness value of each pixel in the probability distribution map represents the degree to which the spatial location matches a specific deformation pattern learned in the past.

[0031] The probability distribution map is used as the input stimulus for model updates. Each time a new probability distribution map is obtained, an iterative update of the multidimensional tensor field model is triggered. A three-dimensional spatial grid is set to cover the entire chest cavity monitoring area. The three-dimensional spatial grid is aligned with the spatial arrangement of the sensor array in the monitoring belt in the horizontal and vertical directions. Three virtual layers are set in the depth direction perpendicular to the body surface, and time is set as the fourth dimension. At each node of the four-dimensional grid, a tensor is initialized. The tensor is a data structure used to record the tension mean, variance, and correlation coefficient with neighboring nodes at spatiotemporal points. At initialization, the tension mean is set to zero, the variance is set to a large preset value, and the correlation coefficient is set to zero.

[0032] The probability distribution map data at the current moment is mapped to a 3D spatial grid according to spatial location, and the tension mean data of the corresponding spatial nodes is updated. The mapping process is to assign each pixel value of the probability distribution map on the 2D plane to the node with the corresponding (X, Y) coordinates and a depth of the middle layer (second layer) in the 3D spatial grid, as the observation value of that node at the current moment. The observed values ​​are then used to update the historical tension mean of each node. Based on the current data and previous historical data, the tension variance of each node and its correlation coefficient with neighboring nodes are recalculated. The variance calculation uses an online update algorithm, and the correlation coefficient is calculated using the Pearson correlation coefficient between the current node and its four directly adjacent nodes (east, south, west, and north) within the most recent time window. Temporal smoothing filtering is applied to the tensor data of all nodes. A first-order low-pass filter is used to smooth the tension mean sequence to reduce the impact of instantaneous fluctuations, completing one iterative update of the multidimensional tensor field model. New probability distribution maps are continuously received and the update process is repeated, allowing the multidimensional tensor field model to evolve dynamically. Each iteration makes the tensor data in the model better reflect the spatiotemporal statistical laws of thoracic deformation. In some embodiments, the temporal smoothing filtering formula is expressed as: ; in: Represents the spatial node at time t. The updated mean tension value, It is the observation value mapped to the horizontal position of this node at time t (when z is an intermediate layer). For other depth layers, Obtained through interpolation. It is the average tension value at the previous moment. This is a smoothing factor between 0 and 1, controlling the weighting of new observations. This formula allows the model's mean tension estimate to track changes while filtering noise. In this way, the multidimensional tensor field model evolves from a static grid structure into a dynamic four-dimensional probabilistic model that characterizes how thoracic surface tension is distributed spatially, changes temporally, and correlates with different locations. Optionally, the correlation coefficient update also employs a similar exponential smoothing approach to capture dynamic changes in spatial correlation. In some embodiments, after iterative updates across multiple respiratory cycles, nodes in the multidimensional tensor field model corresponding to the main respiratory muscle activity areas exhibit higher mean tension, lower variance, and higher positive correlation coefficients with neighboring nodes, while nodes in relatively static areas maintain lower mean and correlation.

[0033] See Figure 3In the multimodal fusion phase of the respiratory phase-morphological feature correlation analysis, the quantitative correlation strength between different respiratory phases (early inspiratory phase, late inspiratory phase, early expiratory phase, and late expiratory phase) and five morphological features (peak width, left slope, right slope, peak curvature, and peak area integral) of the key tension peak was presented. Specifically, the intensity of the feature values ​​is visually represented by a color gradient: dark blue represents the highest correlation strength (feature value approximately 1.4–1.5), and light yellow represents the lowest correlation strength (feature value approximately 0.6–0.7). It is clear from the figure that the peak curvature feature value in early inspiratory phase reaches 1.50, which is the strongest correlation among all combinations, indicating that the morphological performance of peak curvature is most significant in this phase; while the peak width feature value in early inspiratory phase is 1.38, and the peak curvature feature value in late inspiratory phase is 1.32, which also belong to high correlation combinations. In addition, combinations such as the right slope at the end of inspiration (1.18) and the left slope at the end of expiration (1.23) also showed strong correlations, reflecting the specific manifestation of these morphological features under specific respiratory phases.

[0034] In one embodiment of the present invention, the process of decoupling the background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation from the iteratively updated multidimensional tensor field model is based on the spatiotemporal frequency domain decomposition method. The iteratively updated multidimensional tensor field model is decomposed in the spatiotemporal frequency domain. This operation first extracts the tension mean sequence of the model in the time dimension from each spatial grid node to form a four-dimensional spatiotemporal data block. Then, a three-dimensional spatial Fourier transform and a one-dimensional time Fourier transform are applied to this four-dimensional data block to transform it into the spatial wavenumber domain and the time frequency domain. The transformed complex spectrum signal reflects the distribution of tension changes in spatial structure and temporal rhythm.

[0035] Low-pass filtering is used to extract signal components in the multidimensional tensor field model whose variation period exceeds the length of a single respiratory cycle. The low-pass filter operates in the time-frequency domain, and its time cutoff frequency is set according to the lower limit of a typical respiratory cycle. For example, when the typical respiratory cycle is 3 seconds (corresponding to a frequency of about 0.33 Hz), the time cutoff frequency of the low-pass filter is set to 0.1 Hz to retain slow-change components with variation periods greater than 10 seconds. This part of the signal component is separated from the frequency domain signal and inversely transformed, and it is defined as the background component representing the basic tension of the thoracic cavity. The background component mainly reflects the slow-changing or nearly static tension background caused by body position, long-term muscle tension, and the physical constraints of the monitoring strap. Bandpass filtering is used to extract signal components in the multidimensional tensor field model whose variation period matches the length of a typical respiratory cycle. Bandpass filtering also operates in the time-frequency domain, and its passband range is set according to the typical respiratory frequency range of the target population. For example, the lower limit of the passband frequency is set to 0.15 Hz and the upper limit of the passband frequency is set to 0.5 Hz to capture respiratory rhythm signals with a period between 2 seconds and 6.67 seconds. This part of the signal components is separated and inversely transformed, and defined as the active component representing respiratory deformation. The active component concentrates the periodic tension fluctuations caused by respiratory motion.

[0036] The background component and the active component are strictly separated in the frequency domain to ensure orthogonality. This orthogonality is guaranteed by designing non-overlapping ideal filter responses in the frequency domain. The frequency response functions of the low-pass and band-pass filters do not overlap in the passband and stopband, ensuring that the two components extracted from the original signal are uncorrelated in the time domain. The static distribution map and dynamic change sequence of the background component and the active component on a three-dimensional spatial grid are reconstructed separately. The reconstruction process involves performing inverse Fourier transforms on the frequency domain components retained after filtering to restore them to the spatiotemporal domain. For the background component, since its time change is extremely slow, its average value in the time dimension is usually taken to generate a three-dimensional static distribution map. For the active component, its complete time evolution information is retained to generate a three-dimensional dynamic sequence that changes over time.

[0037] A lung function state inference network is constructed, with the input layer dimension matching the data dimension of the activity components. Specifically, the number of nodes in the input layer equals the total number of nodes in the 3D spatial grid, allowing for the reception of one complete frame of 3D spatial distribution data of the activity components at a time. The lung function state inference network contains multiple hidden layers, employing fully connected or convolutional structures to learn high-order nonlinear mappings between the spatiotemporal patterns of the activity components and lung function parameters. The network undergoes supervised training using historically acquired synchronous activity component data and real parameters measured by standard pulmonary function instruments (such as spirometers) as training samples until its output prediction error converges to an acceptable range.

[0038] The dynamic change sequence of the active components representing respiratory deformation is input sequentially into the trained pulmonary function state inference network (PFSWN) in time frame order. Each frame of input data is a numerical vector of the active component corresponding to all nodes of the three-dimensional spatial grid at a certain moment. The PFSWN processes the active component data of each frame and outputs a set of pulmonary function simulation parameters in real time. The forward propagation process of the network maps the input spatial pattern to a set of scalar output values. The pulmonary function simulation parameters include at least simulated tidal volume, simulated lung volume change rate, and simulated airway resistance index. These parameters are continuous real-time estimates generated synchronously with the respiratory cycle. In some embodiments, different spatial distribution patterns of active components will correspond to different combinations of pulmonary function simulation parameter outputs. For example, an active component pattern dominated by the upper chest with high amplitude of change may be mapped by the PFSWN to a larger simulated tidal volume and a higher simulated lung volume change rate. It can be understood that the PFSWN acts as a nonlinear converter from chest deformation patterns to mechanical parameters. The nonlinear transformation process of the hidden layer of the PFSWN can be described as follows: ; in: Representing the The output vector of the hidden layer. It is the output vector of the previous layer (for the first layer, That is, the input active component data vector). It is the connection of the first Layer and First The weight matrix of the layer, It is the first The layer's bias vector, This represents the activation function, such as the ReLU function. Optionally, the lung function state inference network can employ convolutional layers to process the active component input with spatial topology to better capture local deformation features. See Table 1, which shows the lung function simulation parameters corresponding to the output layer of the lung function state inference network and their physical meaning.

[0039] Table 1: Table of Lung Function Simulation Parameters Parameter name Brief description of the physical meaning Output range Simulated tidal volume Estimated volume of gas inhaled or exhaled in a single breath 300-800 ml Simulated lung volume change rate Estimated instantaneous rate of change of lung volume over time -2.0 to +2.0 liters per second Simulated airway drag index A relative index that reflects the pressure required for airflow through the respiratory tract. 0.5-3.0 dimensionless units It is understood that, through the above decoupling and reasoning process, dynamic components directly related to respiration are separated from the original, mixed tension signals, and further transformed into quantitative parameters with clear physiological significance. In some embodiments, when the activity component exhibits a symmetrical and uniform chest-abdominal combined movement pattern, the simulated tidal volume value output by the lung function state inference network is within the normal range, while the simulated airway resistance index is at a low level; when the activity component pattern exhibits an asymmetrical or rapid superficial pattern mainly involving accessory respiratory muscles, the output simulated tidal volume may decrease, while the simulated airway resistance index increases.

[0040] See Figure 4 This paper presents a time-domain comparison of the original tension signal and the reconstructed signal consisting of the background and active components. Specifically, the original tension signal (gray curve) is a mixed signal collected by the sensor array of the chest monitoring belt, containing background tension, respiratory deformation, and noise. By performing spatiotemporal frequency domain decomposition on the multidimensional tensor field model, the tension mean sequence of spatial grid nodes is first extracted to form a four-dimensional spatiotemporal data block. This block is then transformed to the spatial wavenumber domain and time frequency domain through three-dimensional spatial Fourier transform and one-dimensional time Fourier transform. Subsequently, non-overlapping ideal low-pass and band-pass filters are used to achieve orthogonal separation of the background and active components. Finally, the background and active component signal (cyan curve) is reconstructed through inverse Fourier transform. As shown in the figure, the reconstructed signal completely preserves the periodic tension fluctuation characteristics synchronized with the respiratory cycle in the original signal, while filtering out high-frequency noise and non-respiratory transient disturbances. This verifies the effectiveness of the spatiotemporal frequency domain decomposition and frequency domain filtering strategy, providing a reliable signal basis for the subsequent generation of lung function simulation parameters.

[0041] In one embodiment of the present invention, the operation of integrating the background component, the activity component, and lung function simulation parameters to synthesize a multimodal thoracic physiological state description map is performed by a dedicated fusion module of the data processing unit. The static distribution map of the background component, the dynamic change sequence of the activity component, and the time series data of the lung function simulation parameters are fused in a unified spatiotemporal coordinate system to generate a multimodal thoracic physiological state description map. The unified spatiotemporal coordinate system is established based on a three-dimensional spatial grid, and the time axis is aligned with an absolute timestamp. The fusion process assigns a multidimensional feature vector to each spatiotemporal point. This vector contains the baseline tension value of the point, the respiratory deformation amplitude value at the current moment, and all lung function simulation parameters generated by the lung function state inference network at that moment. The multimodal thoracic physiological state description map is internally represented as a five-dimensional data cube that evolves over time, and its dimensions include three spatial dimensions, a time dimension, and a feature channel dimension.

[0042] This system performs real-time analysis of multimodal thoracic physiological state profiles, monitoring for specific physiological events or state transitions. Real-time analysis is achieved by scanning the feature vectors at each time step using a set of predefined logical rules or a lightweight classifier. The determination of specific physiological events is based on a combination of threshold values ​​from multidimensional features and time-duration conditions. When a predefined specific physiological event is detected, a mode-switching command is generated to adjust the sampling frequency or operating mode of the thoracic monitoring sensor array. This command is a structured instruction containing specific parameter settings. For example, when the simulated tidal volume in the lung function simulation parameters is below the hypoventilation threshold for three consecutive cycles and the spatial entropy value of the activity component decreases, the predefined "shallow and rapid breathing event" is triggered. This triggers an instruction to increase the sensor array's sampling frequency from the standard 100 Hz to 200 Hz and switch the operating mode from "standard monitoring" to "high-resolution event capture."

[0043] Simultaneously, based on the continuous changing trend of lung function simulation parameters, the cutoff frequency of filters or the sensitivity threshold of feature extraction in the signal processing link are dynamically adjusted. The continuous changing trend of lung function simulation parameters is quantified by calculating the deviation between their short-term moving average and long-term moving average. For example, when the short-term trend value of the simulated airway resistance index is consistently higher than its long-term baseline value by more than one standard deviation, it indicates that the airway may be narrowing and the respiratory signal spectrum may be shifting to higher frequencies. In this case, the cutoff frequency of the filter separating respiratory-related signals in the low-frequency filtering channel is dynamically lowered to better capture possible rapid shallow breathing components; or, the sensitivity of the resting tension threshold in the detection of key tension peaks is increased to avoid misjudging weak abnormal muscle activity as noise filtering.

[0044] The adjusted data acquisition and processing results are fed back into the generation process of the multimodal thoracic physiological state description map, forming a continuously optimizing adaptive closed loop. Feedback means that a new round of sensor data generated by mode switching and parameter adjustment will re-enter the complete signal processing chain from time-domain registration to map generation, and be calculated using updated processing parameters to generate an updated multimodal thoracic physiological state description map. This updated map will then be used for a new round of real-time analysis and control decisions. In some embodiments, this adaptive closed-loop control enables the system to automatically increase the spatiotemporal resolution of signal acquisition to capture details when the user's breathing pattern changes drastically due to strenuous exercise, and to automatically reduce power consumption and data volume when the user returns to calm sleep. It can be understood that the core logic of the closed loop lies in using high-level state interpretation to inversely optimize the underlying signal acquisition and preprocessing. This closed-loop control process can be formally expressed as: ; in: This represents a multimodal description of the thoracic physiological state generated in the next moment. B represents the entire signal processing and fusion function from raw data to a graph, and B represents the static distribution map of the background components. A dynamic sequence representing the changes in the active components at the current moment. Time-series data representing lung function simulation parameters at the current moment. This represents a regulatory function that describes the multimodal thoracic physiological state at the current moment. As input, the output is a set of adjustment parameters for the sensor array data acquisition mode and signal processing link.

[0045] See Figure 5 In the correlation analysis between simulated lung function parameters and thoracic tension parameters, the heatmap quantified the strength of linear associations among multiple sets of parameters using the Pearson correlation coefficient. Specifically, simulated tidal volume showed a significant positive correlation with respiratory deformation amplitude (correlation coefficient 0.66), indicating that an increase in thoracic deformation amplitude during respiration is usually accompanied by an increase in tidal volume; the airway resistance index showed a significant positive correlation with baseline tension (correlation coefficient 0.63), suggesting that an increase in baseline thoracic tension may correspond to an increase in airway resistance; while simulated tidal volume showed a negative correlation with baseline tension (correlation coefficient -0.45), reflecting the potential inhibitory effect of baseline tension on tidal volume. The correlation coefficients of other parameters, such as lung volume change rate, with other indicators were close to 0, indicating a weak linear association with these parameters.

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A wearable chest monitoring belt for monitoring physiological signals, characterized in that, Includes a data processing unit, which is used to perform: Tension data units and respiratory phase data units are extracted synchronously from the raw data stream of the monitoring belt. The tension data units and the corresponding respiratory phase data units are time-domain registered to generate a tension quantization matrix with phase labels. Based on the phase-labeled tension quantization matrix, key tension peaks in respiratory events are identified and their morphological features are extracted. Based on the differences in morphological characteristics, the key tension peaks are divided into several feature clusters, and a corresponding local deformation mapping template is established for each feature cluster. The local deformation mapping template is used to continuously match and map the real-time acquired thoracic tension data. Based on the result of feature matching, a multidimensional tensor field model reflecting the spatiotemporal changes of the thoracic cavity is dynamically constructed and iteratively updated. The background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation are decoupled from the multidimensional tensor field model. The active component is then input into a preset lung function state inference network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame. By integrating the baseline component, activity component, and lung function simulation parameters, a multimodal description map of the thoracic physiological state is synthesized. Based on the multimodal thoracic physiological state description map, the data acquisition mode and signal processing link of the thoracic monitoring belt are adaptively closed-loop controlled.

2. The wearable chest monitoring belt for physiological signal monitoring according to claim 1, characterized in that, The method of synchronously extracting tension data units and respiratory phase data units from the raw data stream of the monitoring belt includes: Deploy a parallel signal listening interface to capture the raw signal stream output from the sensor array of the monitoring band at a fixed sampling period; A high-frequency filtering channel and a low-frequency filtering channel are set after the parallel signal listening interface to separate tension signals and respiratory-related signals, respectively. Peak detection and baseline calibration are performed on the output of the high-frequency filter channel to obtain the calibrated tension data unit sequence; Zero-crossing detection and envelope extraction are performed on the output of the low-frequency filter channel to generate a sequence of respiratory phase data units characterizing the respiratory cycle; Under the control of the system clock, a unified timestamp is applied to each tension data unit and respiratory phase data unit; Establish a time-stamp alignment buffer, and store the tension data unit sequence and the respiratory phase data unit sequence in pairs according to the unified timestamp.

3. A wearable chest monitoring belt for physiological signal monitoring according to claim 2, characterized in that, The step of performing time-domain registration of tension data units with corresponding respiratory phase data units to generate a tension quantization matrix with phase labels includes: Read the paired tension data units and their associated respiratory phase data units from the time-stamp alignment buffer; Analyze the respiratory phase data unit to determine the respiratory phase category to which the current tension data unit belongs. The respiratory phase category includes early inspiratory phase, late inspiratory phase, early expiratory phase, and late expiratory phase. The tension data unit is normalized and quantized to map its amplitude to a preset quantization level range; The quantized tension amplitude is combined with its corresponding respiratory phase category label to form a tension quantization data point with phase label; The continuous tension quantization data points with phase markers are arranged in chronological order to form a two-dimensional tension quantization matrix; The two-dimensional tension quantization matrix is ​​filled with null values ​​and smoothed to generate a complete tension quantization matrix with phase labels.

4. A wearable chest monitoring belt for physiological signal monitoring according to claim 3, characterized in that, The process of identifying key tension peaks in respiratory events and extracting their morphological features based on the phase-labeled tension quantization matrix includes: In the tension quantization matrix with phase labels, the search interval is from the end of expiration to the end of inspiration; A sliding window strategy is used within the search interval to detect local maximum points in the tension amplitude sequence; Threshold filtering is performed on the detected local maximum points to remove points whose amplitude is lower than the resting tension threshold, and the remaining points are marked as candidate tension peaks; For each candidate tension peak, a predetermined number of adjacent data points are extracted to form a tension peak data segment. Extract a set of morphological features from each tension peak data segment, including peak width, left slope, right slope, peak curvature, and peak area integral. All candidate tension peaks and their corresponding morphological feature sets are stored as a key tension peak feature table.

5. A wearable chest monitoring belt for physiological signal monitoring according to claim 4, characterized in that, The process of dividing key tension peaks into several feature clusters based on morphological differences and establishing a corresponding local deformation mapping template for each feature cluster includes: Read the set of morphological features of all key tension peaks from the key tension peak feature table; The morphological feature set was analyzed using an unsupervised clustering method, and the key tension peaks were divided into multiple non-overlapping feature clusters based on feature similarity. Calculate the average value of all tension peak morphological features within each feature cluster, and use it as the cluster center feature vector of the feature cluster; For each feature cluster, trace back the original spatial location information of the tension peak within the cluster on the chest monitoring band; Based on the spatial location information and the cluster center feature vector, a local deformation mapping template describing the typical tension distribution pattern of the feature cluster is constructed; The local deformation mapping template is stored in the form of a data grid, and the grid node value represents the probability weight of the corresponding feature cluster deformation occurring in a spatial location.

6. A wearable chest monitoring belt for physiological signal monitoring according to claim 5, characterized in that, The method of using a local deformation mapping template to perform continuous feature matching and mapping on real-time acquired thoracic tension data includes: A snapshot of the chest tension distribution at the current moment is obtained from the sensor array of the chest monitoring belt; Extract feature values ​​corresponding to the morphological characteristics of key tension peaks from the snapshot of the thoracic tension distribution; The similarity between the extracted feature values ​​and the cluster center feature vectors of all feature clusters is calculated. Identify the feature clusters with the highest similarity and call their corresponding local deformation mapping templates; The data from the snapshot of the thoracic tension distribution is spatially convolved with the invoked local deformation mapping template. The convolution result is output to obtain a probability distribution map that highlights the deformation region that best matches the current tension pattern.

7. A wearable chest monitoring belt for physiological signal monitoring according to claim 6, characterized in that, The process of dynamically constructing and iteratively updating a multidimensional tensor field model reflecting the spatiotemporal changes of the thoracic cavity based on the feature matching results includes: The probability distribution map is used as the input stimulus for model updating; A three-dimensional spatial grid is set to cover the entire chest cavity monitoring area, and time is used as the fourth dimension; At each node of the four-dimensional grid, a tensor is initialized to record the tension mean, variance, and correlation coefficient with adjacent nodes at the spatiotemporal points. Map the probability distribution data at the current moment to a three-dimensional spatial grid according to spatial location, and update the tension mean data of the corresponding spatial nodes; Based on the current data and previous historical data, recalculate the tension variance of each node and its correlation coefficient with adjacent nodes. The tensor data of all nodes are smoothed and filtered in the time dimension to complete one iterative update of the multidimensional tensor field model. The process of continuously receiving new probability distribution maps and repeating the update process enables the multidimensional tensor field model to evolve dynamically.

8. A wearable chest monitoring belt for physiological signal monitoring according to claim 7, characterized in that, The process of decoupling the background component representing the basic tension of the thoracic cavity and the active component representing respiratory deformation from the multidimensional tensor field model includes: Spatiotemporal frequency domain decomposition is performed on the iteratively updated multidimensional tensor field model; Low-pass filtering was used to extract signal components in the multidimensional tensor field model whose variation period exceeded the length of a single respiratory cycle, and these components were defined as the background components representing the basic tension of the thoracic cavity. Bandpass filtering was used to extract the signal components in the multidimensional tensor field model whose change period matched the length of a typical respiratory cycle, and these components were defined as active components representing respiratory deformation. The background component and the active component are strictly separated in the frequency domain to ensure that the background component and the active component are orthogonal. The static distribution map and dynamic change sequence of the background component and the active component on the three-dimensional spatial grid are reconstructed respectively.

9. A wearable chest monitoring belt for monitoring physiological signals according to claim 8, characterized in that, The step of inputting the activity components into a preset lung function state inference network to generate lung function simulation parameters synchronized with the respiratory cycle frame by frame includes: Construct a lung function state inference network, whose input layer dimension matches the data dimension of the activity component; The lung function state inference network contains multiple hidden layers for learning high-order nonlinear mappings between spatiotemporal patterns of activity components and lung function parameters. The dynamic change sequence of the active components representing respiratory deformation is input into the trained lung function state inference network in time frame order. The lung function state inference network processes each frame of activity component data and outputs a set of lung function simulation parameters in real time. The simulated lung function parameters include at least simulated tidal volume, simulated lung volume change rate, and simulated airway resistance index.

10. A wearable chest monitoring belt for physiological signal monitoring according to claim 9, characterized in that, The adaptive closed-loop control of the data acquisition mode and signal processing link of the thoracic monitoring belt, based on the multimodal thoracic physiological state description atlas, includes: The static distribution map of the background component, the dynamic change sequence of the active component, and the time series data of the lung function simulation parameters are fused in a unified spatiotemporal coordinate system to generate a multimodal thoracic physiological state description map. Real-time analysis of the multimodal thoracic physiological state description map to monitor the occurrence of specific physiological events or state transitions; When a predefined specific physiological event is detected, a mode switching command is generated to adjust the sampling frequency or working mode of the thoracic monitoring belt sensor array; Meanwhile, based on the continuous changing trend of lung function simulation parameters, the cutoff frequency of the filter or the sensitivity threshold of feature extraction in the signal processing link is dynamically adjusted. The adjusted data acquisition and processing results are fed back into the generation process of the multimodal thoracic physiological state description map, forming a continuously optimized adaptive closed loop.

Images