A method, system, device and medium for automatic temperature control of a moxibustion device that adapts to human posture.

Abstract

This application relates to a method, system, device, and medium for automatic temperature control of a moxibustion device that adapts to human posture. The method includes: acquiring multi-dimensional physiological indicators to construct an initial physiological dataset; extracting baseline features of human posture through deep learning and inputting them into a machine learning classifier to obtain posture change type labels; calculating the variation values ​​of contact area and pressure distribution based on these labels, adjusting and calibrating them through a heat distribution model to obtain a heat distribution map; then calculating and optimizing the temperature parameters of each heating unit based on the heat distribution map, and generating a drive signal to control the heating elements to perform heating actions. This method can achieve accurate identification and quantitative characterization of human posture, respond in real time to the influence of posture changes on the moxibustion contact state, complete zoned adaptive temperature control through precise heat distribution, improve temperature control accuracy and uniformity, reduce the risk of local overheating, and improve the safety, stability, and comfort of moxibustion therapy.

Description

Technical Field

[0001] This invention belongs to the field of intelligent temperature control technology, and in particular relates to a method, system, device and medium for automatic temperature control of a moxibustion device that adapts to human posture. Background Technology

[0002] Traditional moxibustion therapy devices mostly use fixed power and fixed-area heating methods. When the user's body position or posture changes, they cannot sense and adjust the heating strategy in real time, which can easily lead to problems such as excessively high local temperatures, uneven heating, or reduced therapeutic effects. Existing temperature control solutions mostly rely on simple temperature feedback and do not take into account changes in body posture, contact state, and pressure distribution for comprehensive judgment. This makes it difficult to achieve dynamic, precise, and safe adaptive temperature adjustment, resulting in low levels of safety, comfort, and intelligence in use. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, system, device, and medium for automatic temperature control of moxibustion devices that can achieve adaptive temperature control for heating, improve the accuracy and uniformity of moxibustion temperature control, and reduce the risk of local overheating.

[0004] In a first aspect, this application provides a method for automatic temperature control of a moxibustion device that adapts to human posture, including:

[0005] The system acquires multi-dimensional human physiological indicators of the user collected by the built-in sensor array of the moxibustion device to obtain an initial physiological dataset; the human physiological indicators include joint angles, skin contact point location information and muscle tension indicators.

[0006] Deep learning network features were used to extract features from the initial physiological dataset to obtain baseline posture features for recognizing human posture baseline patterns.

[0007] Based on the baseline posture features, feature matching of human physiological indicators is performed to obtain posture change vectors. The posture change vectors are then input into a machine learning classifier for classification to obtain change type labels.

[0008] The variation values ​​of contact area and pressure distribution are calculated based on the change type label, and the heat distribution is adjusted by combining it with a pre-built heat distribution model to obtain a heat distribution map.

[0009] The temperature output parameters of each heating unit of the moxibustion device are calculated based on the heat distribution map. The optimized temperature output parameters are used to control the heating elements and generate corresponding drive signals.

[0010] In one embodiment, a deep learning network is used to extract features from an initial physiological dataset to obtain baseline pose features for recognizing human pose baseline patterns, including:

[0011] A deep learning network was used to extract features from the time-series physiological signals in the initial physiological dataset, resulting in high-dimensional feature vectors.

[0012] Dimensionality reduction is performed on the high-dimensional feature vectors to obtain low-dimensional feature representations.

[0013] Cluster analysis based on low-dimensional feature representation identifies several baseline attitude feature clusters.

[0014] If the number of reference attitude feature clusters exceeds a preset threshold, the reference attitude feature clusters are merged to obtain merged feature clusters.

[0015] The central feature vectors corresponding to each feature cluster are calculated based on the merged feature clusters, and each central feature vector is used as the reference pose feature.

[0016] In one embodiment, the central feature vector corresponding to each feature cluster is calculated using the following formula:

[0017]

[0018] in, This represents the central eigenvector of the merged feature cluster. This represents the number of samples representing low-dimensional features in the merged feature cluster. Indicates the first The confidence weight vector, represented by a sensor array with low-dimensional features, is obtained by normalizing the signal integrity, temporal stability, and contact fit of the multi-dimensional sensors built into the moxibustion device. Indicates the first The posture feature dimension contribution weight vector, represented by several low-dimensional features, is quantified and assigned by the mutual information entropy values ​​of various physiological indicators, such as joint angles, muscle tension, and skin contact point positions, in relation to human posture representation. Represents the Hadamard product of vectors. Indicates the first The feature vector is represented by a low-dimensional feature representation. This represents the mean vector of all low-dimensional feature representations within the merged feature cluster. Indicates the first The Euclidean distance between each feature vector and the cluster mean vector This represents the outlier suppression coefficient, a pre-defined positive real number based on the human posture monitoring scenario of the moxibustion device, used to weight and attenuate outlier feature vectors within a cluster. This represents the Gaussian outlier penalty term.

[0019] In one embodiment, a posture change vector is obtained by feature matching of human physiological indicators based on baseline posture features. This posture change vector is then input into a machine learning classifier for classification to obtain change type labels, including:

[0020] Based on the baseline posture features, feature matching of human physiological indicators is performed, and the cosine distance algorithm is used to calculate the deviation value after matching, and the effective deviation value exceeding the preset threshold is screened out.

[0021] The effective deviation values ​​of each dimension are constructed into vector dimensions and quantized into features to obtain the posture change vector that represents the changes in human posture during the use of the moxibustion device.

[0022] The vector dimension is constructed by mapping and structuring the effective deviation values ​​according to the dimensional attributes of human physiological indicators, resulting in a dimension-matched deviation data sequence.

[0023] Feature quantization encoding involves normalizing and numerically quantizing the structured, arranged bias data sequence to obtain a standardized feature vector.

[0024] The posture change vector is input into a pre-trained machine learning classifier to perform classification and recognition of human posture change types, and the classification and recognition results are obtained.

[0025] Based on the classification and recognition results, output the corresponding human posture change type label.

[0026] In one embodiment, the variation values ​​of contact area and pressure distribution are calculated based on the variation type label, and the heat distribution is adjusted in conjunction with a pre-built heat distribution model to obtain a heat distribution map, including:

[0027] Extract the contact area variation data and pressure distribution variation data corresponding to the change type labels to obtain the initial set of variation values.

[0028] Contact area variation data includes the amount of change in the contact area, the rate of change in the area, and the offset of the contact profile.

[0029] Pressure distribution variation data includes the pressure value change, pressure change rate, and pressure center offset coordinates in the contact area.

[0030] The initial set of variation values ​​is input into the pre-constructed heat distribution model, and the preliminary adjustment parameters of the heat distribution are obtained by solving.

[0031] The heat distribution model is constructed based on the radial basis function neural network algorithm.

[0032] The weights and bias parameters of the heat distribution model are iteratively calibrated using the backpropagation algorithm after initial parameter adjustment until the mean square error between the model's output heat distribution prediction value and the benchmark reference value is less than a preset threshold, thus obtaining the adjusted heat distribution matrix.

[0033] The heat distribution matrix is ​​divided according to the area affected by moxibustion, resulting in several heat distribution sub-regions.

[0034] If the variation value of the heat distribution sub-region exceeds the preset threshold, a secondary parameter calibration is performed on the heat distribution sub-region, and the calibrated heat distribution sub-regions are stitched together to obtain the heat distribution map.

[0035] The heat distribution map includes several temperature data points, the spatial coordinates of each temperature data point, the temperature value of each temperature data point, and the corresponding mapping relationship between the heating unit of the moxibustion device and the contact area of ​​the human body.

[0036] In one embodiment, the temperature output parameters of each heating unit of the moxibustion device are calculated based on the heat distribution map, and the optimized temperature output parameters are used to control the heating element to generate a corresponding drive signal, including:

[0037] The temperature output parameters of each heating unit are calculated based on the heat distribution map to form a dynamic temperature parameter sequence.

[0038] Temperature smoothing optimization is performed based on a dynamic temperature parameter sequence combined with a preset temperature change rate constraint to obtain preliminary optimized temperature parameters.

[0039] The initially optimized temperature parameters are subjected to regional adaptive weighting based on the area of ​​moxibustion application to generate optimized temperature parameters that adapt to changes in human posture.

[0040] Drive signals corresponding to each heating unit are generated based on optimized temperature parameters.

[0041] Each heating element performs heating operations according to the corresponding drive signal.

[0042] Secondly, this application also provides a human posture-adaptive automatic temperature control system for moxibustion devices, the system comprising:

[0043] The physiological data acquisition module is used to acquire multi-dimensional human physiological indicators of the user collected by the built-in sensor array of the moxibustion device to obtain an initial physiological dataset; human physiological indicators include joint angles, skin contact point location information and muscle tension indicators.

[0044] The pose feature extraction module is used to extract features from the initial physiological dataset using a deep learning network to obtain baseline pose features for recognizing human pose reference patterns.

[0045] The posture change recognition module is used to perform feature matching on human physiological indicators based on baseline posture features to obtain posture change vectors. The posture change vectors are then input into a machine learning classifier for classification to obtain change type labels.

[0046] The heat distribution generation module is used to calculate the variation values ​​of contact area and pressure distribution based on the change type label, and adjust the heat distribution by combining it with the pre-built heat distribution model to obtain the heat distribution map.

[0047] The temperature adaptive control module is used to calculate the temperature output parameters of each heating unit of the moxibustion device based on the heat distribution map, and to control the heating element using the optimized temperature output parameters, thereby generating the corresponding drive signal.

[0048] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0049] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0050] The aforementioned method, system, computer equipment, and storage medium for automatic temperature control of a moxibustion device based on human posture adaptation acquire multi-dimensional human physiological indicators collected by the built-in sensor array of the moxibustion device to form an initial physiological dataset. These physiological indicators include joint angles, skin contact point location information, and muscle tension indicators. Feature extraction is performed on the initial physiological dataset using a deep learning network to obtain baseline posture features for identifying human posture reference patterns. Based on the baseline posture features, feature matching is performed on real-time human physiological indicators to obtain posture change vectors. These vectors are then input into a machine learning classifier for classification to obtain posture change type labels. Based on the posture change type labels, the variation values ​​of the contact area and pressure distribution are calculated. Combined with a pre-constructed heat distribution model, the heat distribution is adjusted and calibrated to obtain a heat distribution map containing several temperature data points, their corresponding spatial coordinates, temperature values, and the mapping relationship between the moxibustion device heating unit and the human contact area. The temperature output parameters of each heating unit are calculated based on the heat distribution map, and the temperature output parameters are optimized to generate corresponding drive signals to control the heating elements to perform corresponding heating actions. This method achieves accurate identification of human posture and quantitative representation of posture changes through multi-dimensional physiological index collection and deep learning feature extraction, enabling real-time response to the impact of human posture changes on the moxibustion contact state. By constructing a heat distribution map that includes temperature data points, spatial coordinates, temperature values, and the mapping relationship between the heating unit and the contact area, and optimizing temperature output parameters and generating drive signals based on the heat distribution, it can realize the zoning and adaptive temperature control of the heating unit of the moxibustion device, improve the accuracy and uniformity of moxibustion temperature control, reduce the risk of local overheating, and improve the safety, stability, and comfort of the moxibustion therapy process. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies 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.

[0052] Figure 1 A flowchart of an automatic temperature control method for a moxibustion device that adapts to human posture is provided in an embodiment of the present invention;

[0053] Figure 2 The diagram below shows a structural block diagram of a human posture-adaptive automatic temperature control system for a moxibustion device, as provided in an embodiment of the present invention. Detailed Implementation

[0054] 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.

[0055] In one embodiment, such as Figure 1 As shown, this application provides a human posture-adaptive automatic temperature control method for moxibustion devices, which may include the following steps:

[0056] Step S101: Obtain multi-dimensional human physiological indicators of the user collected by the built-in sensor array of the moxibustion device to obtain the initial physiological dataset; the human physiological indicators include joint angles, skin contact point location information and muscle tension indicators.

[0057] Specifically, the moxibustion device has a built-in sensor array that collects multi-dimensional human physiological indicators in real time during the user's use. Among them, joint angles are used to characterize the degree of bending and posture of the human limbs, skin contact point information is used to reflect the contact position and coverage of the moxibustion device with the human skin, and muscle tension index is used to help judge the stability of the human posture. All the collected physiological indicator data are sorted and summarized to form an initial physiological dataset.

[0058] Step S102: Deep learning network feature extraction is performed on the initial physiological dataset to obtain the baseline posture features used to identify the baseline pattern of human posture.

[0059] The initial physiological dataset is input into a pre-defined deep learning network. Through operations such as convolution and pooling, redundant information in the initial physiological dataset is filtered out, and key data that can characterize the core features of human posture are extracted. After feature fusion and normalization, a baseline posture feature is obtained to identify the baseline pattern of human posture. This baseline posture feature serves as a reference standard for subsequent judgment of changes in human posture.

[0060] Step S103: Based on the baseline posture features, perform feature matching on human physiological indicators to obtain posture change vectors, input the posture change vectors into a machine learning classifier for classification, and obtain change type labels.

[0061] Using baseline posture features as a reference, feature matching is performed on real-time collected human physiological indicators to calculate the deviation between real-time physiological indicators and baseline posture features. The deviation values ​​of each dimension are then used to construct vector dimensions and encode features to obtain a posture change vector representing changes in human posture. This posture change vector is then input into a pre-trained machine learning classifier, which identifies and classifies the type of posture change and outputs the corresponding change type label. This label is used to represent the specific form of human posture change.

[0062] Step S104: Calculate the variation values ​​of contact area and pressure distribution based on the change type label, and adjust the heat distribution by combining it with the pre-built heat distribution model to obtain a heat distribution map.

[0063] Based on the output change type label, the influence of human posture changes on the contact state between the moxibustion device and the human body is determined, and then the variation values ​​of the contact area and pressure distribution are calculated. The above variation values ​​are input into the pre-constructed heat distribution model, and the heat distribution is dynamically adjusted and calibrated through model calculation, and finally a heat distribution map is obtained. The heat distribution map contains several temperature data points, the spatial coordinates corresponding to each temperature data point, the temperature value of each temperature data point, and the corresponding mapping relationship between the heating unit of the moxibustion device and the contact area of ​​the human body.

[0064] Step S105: Calculate the temperature output parameters of each heating unit of the moxibustion device according to the heat distribution map, control the heating element with the optimized temperature output parameters, and generate the corresponding drive signal.

[0065] Based on the heat distribution map, and combining the temperature data points, spatial coordinates, temperature values, and the corresponding mapping relationship between the heating unit and the human body contact area contained therein, the temperature output parameters corresponding to each heating unit of the moxibustion device are calculated. The calculated temperature output parameters are then smoothed and subjected to regional adaptive weighting to obtain optimized temperature output parameters. Based on the optimized temperature output parameters, a drive signal corresponding to each heating unit is generated to control each heating unit to perform the corresponding heating operation, thereby achieving adaptive temperature adjustment.

[0066] The aforementioned method for automatic temperature control of a moxibustion device based on human posture adaptation involves acquiring multi-dimensional human physiological indicators collected by the built-in sensor array of the moxibustion device to form an initial physiological dataset. These physiological indicators include joint angles, skin contact point location information, and muscle tension indicators. Feature extraction is performed on the initial physiological dataset using a deep learning network to obtain baseline posture features for identifying human posture reference patterns. Based on these baseline posture features, feature matching is performed on real-time human physiological indicators to obtain posture change vectors. These vectors are then input into a machine learning classifier for classification to obtain posture change type labels. Based on these labels, the variation values ​​of contact area and pressure distribution are calculated. Combined with a pre-built heat distribution model, the heat distribution is adjusted and calibrated to obtain a heat distribution map containing several temperature data points, their corresponding spatial coordinates, temperature values, and the mapping relationship between the moxibustion device's heating unit and the human contact area. The temperature output parameters of each heating unit are calculated based on the heat distribution map. These parameters are then optimized, and corresponding drive signals are generated to control the heating elements to perform corresponding heating actions. This method achieves accurate identification of human posture and quantitative representation of posture changes through multi-dimensional physiological index collection and deep learning feature extraction, enabling real-time response to the impact of human posture changes on the moxibustion contact state. By constructing a heat distribution map that includes temperature data points, spatial coordinates, temperature values, and the mapping relationship between the heating unit and the contact area, and optimizing temperature output parameters and generating drive signals based on the heat distribution, it can realize the zoning and adaptive temperature control of the heating unit of the moxibustion device, improve the accuracy and uniformity of moxibustion temperature control, reduce the risk of local overheating, and improve the safety, stability, and comfort of the moxibustion therapy process.

[0067] In one embodiment, deep learning network feature extraction is performed on the initial physiological dataset to obtain baseline pose features for recognizing human pose baseline patterns, which may include the following steps:

[0068] Step S201: Use a deep learning network to extract features from each time series physiological signal in the initial physiological dataset to obtain a high-dimensional feature vector.

[0069] Step S202: Perform dimensionality reduction processing on the high-dimensional feature vector to obtain a low-dimensional feature representation.

[0070] Step S203: Cluster analysis is performed based on low-dimensional feature representation to identify several baseline attitude feature clusters.

[0071] Step S204: If the number of reference attitude feature clusters exceeds a preset threshold, then each reference attitude feature cluster is merged to obtain a merged feature cluster.

[0072] Step S205: Calculate the central feature vector corresponding to each feature cluster based on the merged feature clusters, and use each central feature vector as the reference pose feature.

[0073] Specifically, a deep learning network is used to extract features from the time-series physiological signals in the initial physiological dataset, resulting in high-dimensional feature vectors. Each time-series physiological signal corresponds to real-time data collection of joint angles, skin contact point locations, and muscle tension indices in the initial physiological dataset. These high-dimensional feature vectors contain multi-dimensional core features of human posture. Dimensionality reduction is then performed on the extracted high-dimensional feature vectors to filter redundant features and simplify feature dimensions, resulting in low-dimensional feature representations that accurately characterize key information about human posture. Clustering analysis is then performed based on these low-dimensional feature representations. Clustering algorithms categorize low-dimensional feature representations with similar features, identifying several baseline posture feature clusters. Each baseline posture feature cluster corresponds to a different basic human posture pattern. If the number of identified baseline posture feature clusters exceeds a preset threshold, it indicates that the baseline posture division is too fine. In this case, the baseline posture feature clusters need to be merged, removing highly similar feature clusters to obtain merged feature clusters with a number that meets the preset threshold. Based on the merged feature clusters, a central feature vector is calculated for each merged feature cluster. This central feature vector represents the core posture feature of the corresponding merged feature cluster, and each central feature vector is used as a baseline posture feature.

[0074] This embodiment extracts features from the time-series signals of initial physiological data using deep learning, which fully uncovers the core features of human posture and improves the accuracy of feature representation. Dimensionality reduction simplifies the feature dimensions, reducing the computational load of subsequent data processing and improving processing efficiency. Cluster analysis and cluster merging ensure the representativeness and rationality of the baseline posture features, avoiding posture recognition bias caused by too many or too few baseline posture divisions. The resulting baseline posture features provide a solid foundation for the accurate identification and quantitative representation of human posture changes.

[0075] In one embodiment, the central feature vector corresponding to each feature cluster can be calculated using the following formula:

[0076]

[0077] in, This represents the central eigenvector of the merged feature cluster. This represents the number of samples representing low-dimensional features in the merged feature cluster. Indicates the first The confidence weight vector, represented by a sensor array with low-dimensional features, is obtained by normalizing the signal integrity, temporal stability, and contact fit of the multi-dimensional sensors built into the moxibustion device. Indicates the first The posture feature dimension contribution weight vector, represented by several low-dimensional features, is quantified and assigned by the mutual information entropy values ​​of various physiological indicators, such as joint angles, muscle tension, and skin contact point positions, in relation to human posture representation. Represents the Hadamard product of vectors. Indicates the first The feature vector is represented by a low-dimensional feature representation. This represents the mean vector of all low-dimensional feature representations within the merged feature cluster. Indicates the first The Euclidean distance between each feature vector and the cluster mean vector This represents the outlier suppression coefficient, a pre-defined positive real number based on the human posture monitoring scenario of the moxibustion device, used to weight and attenuate outlier feature vectors within a cluster. This represents the Gaussian outlier penalty term.

[0078] This embodiment extracts features from the time-series signals of initial physiological data using deep learning, which fully uncovers the core features of human posture and improves the accuracy of feature representation. Dimensionality reduction simplifies feature dimensions, reducing the computational load of subsequent data processing and improving efficiency. Cluster analysis and cluster merging ensure the representativeness and rationality of the baseline posture features, avoiding posture recognition bias. In the calculation of the central feature vector, the confidence weight of sensor acquisition and the contribution weight of posture feature dimensions are integrated, and the Hadamard product weighting and Gaussian outlier penalty term are used to suppress interference from outlier features within clusters, significantly improving the reliability and robustness of the central feature vector. The resulting baseline posture features are highly accurate and resistant to interference, providing a solid foundation for the accurate identification and quantitative representation of human posture changes.

[0079] In one embodiment, feature matching of human physiological indicators based on baseline posture features is used to obtain a posture change vector. The posture change vector is then input into a machine learning classifier for classification to obtain a change type label. This may include the following steps:

[0080] Step S301: Based on the baseline posture features, perform feature matching on human physiological indicators, use the cosine distance algorithm to calculate the deviation value after matching, and filter out the valid deviation values ​​that exceed the preset threshold.

[0081] Preferably, using the generated baseline posture features as a reference standard, real-time collected human physiological indicators (joint angles, skin contact point position information, muscle tension indicators) are input. Feature matching is used to achieve the correspondence between real-time physiological indicators and baseline posture features. A cosine distance algorithm is used to calculate the cosine value of the angle between the feature vector corresponding to the real-time human physiological indicators and the feature vector of the baseline posture. This cosine value is converted into a deviation value that represents the degree of difference between the two. The larger the deviation value, the more significant the difference between the real-time human posture and the baseline posture. A preset deviation threshold is used to distinguish between effective posture changes and invalid interference. By comparing the calculated deviation value with the preset threshold, invalid deviations less than or equal to the preset threshold (such as deviations caused by small sensor fluctuations or small human posture jitters) are eliminated, and effective deviation values ​​exceeding the preset threshold are selected. This effective deviation value can truly represent the actual changes in human posture during the use of the moxibustion device.

[0082] Step S302: Construct vector dimensions and quantize and encode features for the effective deviation values ​​of each dimension to obtain the posture change vector that represents the changes in human posture during the use of the moxibustion device.

[0083] Preferably, the vector dimension is constructed by mapping and structuring the effective deviation values ​​according to the dimensional attributes of human physiological indicators, resulting in a dimension-matched deviation data sequence.

[0084] Feature quantization encoding involves normalizing and numerically quantizing the structured, arranged bias data sequence to obtain a standardized feature vector.

[0085] Step S303: Input the posture change vector into a pre-trained machine learning classifier to perform human posture change type classification and recognition, and obtain the classification and recognition result.

[0086] For example, the posture change vector can accurately represent the changes in human posture during the use of the moxibustion device, and it matches the input dimension of the pre-trained machine learning classifier. The pre-trained machine learning classifier has been trained with a large number of human posture change samples. The training samples contain posture change vectors and corresponding labels corresponding to different posture transformation types, and have the ability to accurately identify human posture transformation types. After the posture change vector is input into the classifier, the classifier performs feature parsing, category comparison and probability calculation on the input vector through built-in algorithms to determine the specific type of human posture transformation, and finally outputs the classification recognition result. The classification recognition result includes the category label of the posture transformation type and the corresponding recognition confidence, ensuring the accuracy and reliability of posture change type recognition.

[0087] Step S304: Based on the classification and recognition results, output the corresponding human posture change type label.

[0088] Specifically, feature matching is performed on real-time collected human physiological indicators based on baseline posture features. A cosine distance algorithm is used to calculate the deviation between the matched real-time physiological indicators and the baseline posture features. The calculated deviation values ​​are then filtered using a preset deviation threshold to remove invalid deviation data and select valid deviation values ​​exceeding the threshold. Vector dimension construction and feature quantization encoding are performed on the valid deviation values ​​corresponding to each dimension of the human physiological indicators. Vector dimension construction involves mapping and structuring the valid deviation values ​​according to the dimensional attributes of the human physiological indicators (joint angles, skin contact point location information, muscle tension indicators) to obtain a deviation data sequence matching the dimensions of the physiological indicators. Feature quantization encoding normalizes the structured deviation data sequence, mapping the data to a preset standardized range, and simultaneously performing numerical quantization to transform it into a standardized feature vector that can be used for classification and recognition. This standardized feature vector is the posture change vector representing the changes in human posture during the use of the moxibustion device. The posture change vector is input into a pre-trained machine learning classifier, which performs classification and recognition of the human posture transformation type and outputs the classification and recognition results. Based on the classification and recognition results, corresponding labels representing the specific forms of human posture changes are output.

[0089] This embodiment uses the cosine distance algorithm to calculate the deviation value, which can accurately quantify the difference between real-time physiological indicators and baseline posture features, improving the accuracy of deviation calculation. By filtering valid deviation values ​​and eliminating invalid interference data, the redundancy of subsequent data processing is reduced, and processing efficiency is improved. The vector dimension is constructed and arranged in a structured manner according to the physiological indicator dimension to ensure the orderliness and correlation of the deviation data. Feature quantization encoding is performed through normalization and numerical quantization to standardize the deviation data, ensuring the effectiveness and uniformity of posture change vectors. The standardized posture change vectors are input into a pre-trained machine learning classifier, which can achieve accurate classification and recognition of human posture transformation types, ensuring that the output change type labels are accurate and reliable.

[0090] In one embodiment, the variation values ​​of contact area and pressure distribution are calculated based on the variation type label, and the heat distribution is adjusted in conjunction with a pre-built heat distribution model to obtain a heat distribution map. This may include the following steps:

[0091] Step S401: Extract the contact area variation data and pressure distribution variation data corresponding to the change type label to obtain the initial set of variation values.

[0092] Preferably, the contact area variation data includes the amount of change in the contact area, the rate of change in the area, and the offset of the contact contour.

[0093] Pressure distribution variation data includes the pressure value change, pressure change rate, and pressure center offset coordinates in the contact area.

[0094] Step S402: Input the initial set of variation values ​​into the pre-constructed heat distribution model and solve for the preliminary adjustment parameters of heat distribution.

[0095] The heat distribution model is constructed based on the radial basis function neural network algorithm.

[0096] Furthermore, after inputting the initial set of variation values ​​into the heat distribution model, the model completes forward propagation calculations based on the input variation characteristics, performs preliminary fitting and calculation of the heat distribution change trend in the contact area between the moxibustion device and the human body, and outputs the corresponding preliminary adjustment parameters of the heat distribution. These parameters are used to characterize the preliminary adjustment direction and adjustment range of the heat distribution.

[0097] Step S403: The weights and bias parameters of the heat distribution model are iteratively calibrated using the backpropagation algorithm after initial parameter adjustment until the mean square error between the predicted heat distribution value and the benchmark reference value output by the model is less than a preset threshold, thus obtaining the adjusted heat distribution matrix.

[0098] Schematic, the preliminary adjustment parameters are used as model input constraints. The weights and bias parameters of the pre-constructed radial basis function neural network-type heat distribution model are iteratively updated and calibrated based on the error backpropagation algorithm. During the iteration process, the mean square error between the predicted heat distribution value output by the model and the preset benchmark reference value is used as the convergence criterion. The formula for calculating the mean square error is:

[0099]

[0100] in, Indicates mean square error. Indicates the number of data points for heat distribution. The first output of the model represents the... One predicted value for heat distribution, Indicates the first A baseline reference value for heat distribution.

[0101] When the mean square error obtained by iterative calculation is less than the preset error threshold, the model is determined to have converged and the iterative calibration is stopped. At this time, the converged heat distribution model outputs stable and accurate heat distribution data. The data is then organized into a matrix according to the correspondence between the spatial location and the heating unit, which yields the adjusted heat distribution matrix.

[0102] Step S404: Divide the heat distribution matrix according to the area of ​​moxibustion effect to obtain several heat distribution sub-regions.

[0103] Step S405: If the variation value of the heat distribution sub-region exceeds the preset threshold, perform secondary parameter calibration on the heat distribution sub-region, and stitch and merge the calibrated heat distribution sub-regions to obtain the heat distribution map.

[0104] The heat distribution map includes several temperature data points, the spatial coordinates of each temperature data point, the temperature value of each temperature data point, and the corresponding mapping relationship between the heating unit of the moxibustion device and the contact area of ​​the human body.

[0105] Specifically, the contact area variation data and pressure distribution variation data corresponding to the change type labels are extracted to obtain an initial set of variation values. The contact area variation data includes the area change, area change rate, and contact contour offset of the contact area. The pressure distribution variation data includes the pressure value change, pressure change rate, and pressure center offset coordinates of the contact area. The initial set of variation values ​​is input into a heat distribution model constructed based on a radial basis function neural network algorithm to obtain preliminary adjustment parameters for heat distribution. The weights and bias parameters of the heat distribution model are iteratively calibrated using the preliminary adjustment parameters through an error backpropagation algorithm until the mean square error between the model's output heat distribution prediction value and the benchmark reference value is less than a preset threshold, resulting in an adjusted heat distribution matrix. The heat distribution matrix is ​​divided according to the moxibustion action area to obtain several heat distribution sub-regions. If the variation value of a heat distribution sub-region exceeds the preset threshold, a secondary parameter calibration is performed on the heat distribution sub-region. The calibrated heat distribution sub-regions are then spliced ​​and merged to obtain a heat distribution map containing several temperature data points, the spatial coordinates of each temperature data point, the temperature value, and the mapping relationship between the moxibustion device heating unit and the human body contact area.

[0106] This embodiment extracts multi-dimensional contact area and pressure distribution variation data that match the type of posture change, which can accurately quantify the impact of human posture changes on the moxibustion contact state; it uses a radial basis function neural network combined with an error backpropagation algorithm to construct and iteratively calibrate a heat distribution model, which can improve the accuracy of heat distribution prediction and the model convergence stability; through region division and secondary parameter calibration, it achieves fine correction of heat distribution and avoids local heat distribution anomalies; the final generated heat distribution map has complete temperature, spatial and mapping relationship information, which improves the adaptability and reliability of the moxibustion temperature control system.

[0107] In one embodiment, calculating the temperature output parameters of each heating unit of the moxibustion device based on the heat distribution map, controlling the heating element using the optimized temperature output parameters, and generating a corresponding drive signal may include the following steps:

[0108] Step S501: Calculate the temperature output parameters of each heating unit based on the heat distribution diagram to form a dynamic temperature parameter sequence.

[0109] The heating unit includes an independent temperature control zone, zoned heating channels, and a zone control module.

[0110] Step S502: Based on the dynamic temperature parameter sequence and the preset temperature change rate constraint, temperature smoothing optimization is performed to obtain the preliminary optimized temperature parameters.

[0111] Step S503: Perform regional adaptive weighting on the initially optimized temperature parameters according to the area of ​​moxibustion action to generate optimized temperature parameters that adapt to changes in human posture.

[0112] Preferably, the process takes the initially optimized temperature parameters as the processing object, and according to the moxibustion effect area division rules preset by the moxibustion device, combined with the heat distribution characteristics, contact state and human posture change type of each area, assigns corresponding adaptive weighting coefficients to different areas, and adjusts the initially optimized temperature parameters in a regionalized manner through weighted calculation, so that the temperature parameters match the actual temperature control needs of each area, and finally generates optimized temperature parameters that can adapt to human posture changes.

[0113] Step S504: Generate drive signals corresponding to each heating unit based on the optimized temperature parameters.

[0114] The optimized temperature parameters are converted into control parameters that the moxibustion device's drive module can recognize. Based on the mapping relationship between the heating unit and the moxibustion area, a matching drive signal is generated independently for each heating unit. The drive signal contains control information such as heating power and working sequence, ensuring that the drive signal corresponds one-to-one with the temperature control requirements of each heating unit, and providing accurate instructions for the execution of the heating element.

[0115] In step S505, each heating element performs a heating operation according to the corresponding drive signal.

[0116] Preferably, the heating element includes a flexible heating film, a ceramic heating plate, a PTC constant temperature heating element, and a carbon fiber heating wire.

[0117] Each heating element receives the drive signal corresponding to its heating unit and performs the corresponding heating action according to the heating power, working sequence and other parameters specified by the drive signal, so that the actual output temperature of each area of ​​the moxibustion device is consistent with the optimized temperature parameters, realizing zoned adaptive heating based on changes in human posture, and ensuring the accuracy and uniformity of the moxibustion temperature output.

[0118] Specifically, based on the obtained heat distribution map, combined with several temperature data points contained in the map, the spatial coordinates of each temperature data point, the temperature value, and the corresponding mapping relationship between the heating unit of the moxibustion device and the contact area of ​​the human body, the temperature output parameters corresponding to each heating unit are calculated. The temperature output parameters of each heating unit are organized according to the time sequence and unit number to form a dynamic temperature parameter sequence. Based on this dynamic temperature parameter sequence, combined with the preset temperature change rate constraint (used to limit the temperature rise and fall range and avoid sudden temperature changes), the dynamic temperature parameter sequence is subjected to temperature smoothing optimization processing to filter parameter fluctuation interference and obtain preliminary optimized temperature parameters with satisfactory stability. The preliminary optimized temperature parameters are divided according to the moxibustion action area. According to the physiotherapy needs and contact state differences of each area, regional adaptive weighting processing is performed to assign appropriate weight coefficients to different action areas, generating optimized temperature parameters that can accurately adapt to changes in human posture. Based on the optimized temperature parameters, a drive signal corresponding to each heating unit is generated to ensure that the parameters of the drive signal are accurately matched with the temperature control requirements of the heating unit. Each heating element receives the corresponding drive signal and executes the heating operation according to the signal instruction to achieve accurate output of moxibustion temperature.

[0119] This embodiment is based on a precise heat distribution map to ensure the accuracy of the temperature output parameters calculated for each heating unit. The dynamic temperature parameter sequence can respond in real time to the differences in heat distribution caused by changes in human posture. Temperature smoothing optimization combined with preset temperature change rate constraints can effectively avoid sudden temperature rises and falls, improve the stability of temperature output, and reduce the risk of local overheating or underheating. Regional adaptive weighting adjusts parameters according to the differences in the moxibustion area to achieve precise temperature control in different zones, adapting to the differences in contact state caused by changes in human posture and improving the adaptability of temperature control. By optimizing temperature parameters to generate corresponding drive signals, it ensures that the heating operation of the heating element is precisely matched with the temperature control requirements, realizing the coordinated control of the heating unit and the heating element.

[0120] In one embodiment, such as Figure 2 As shown, this application also provides a human posture-adaptive automatic temperature control system for moxibustion devices, the system including:

[0121] The physiological data acquisition module 601 is used to acquire multi-dimensional human physiological indicators of the user collected by the built-in sensor array of the moxibustion device to obtain an initial physiological dataset; the human physiological indicators include joint angles, skin contact point location information and muscle tension indicators.

[0122] The pose feature extraction module 602 is used to extract features from the initial physiological dataset using a deep learning network to obtain baseline pose features for recognizing human pose baseline patterns.

[0123] The posture change recognition module 603 is used to perform feature matching on human physiological indicators based on baseline posture features to obtain posture change vectors, and input the posture change vectors into a machine learning classifier for classification to obtain change type labels.

[0124] The heat distribution generation module 604 is used to calculate the variation values ​​of contact area and pressure distribution based on the change type label, and adjust the heat distribution by combining it with the pre-built heat distribution model to obtain a heat distribution map.

[0125] The temperature adaptive control module 605 is used to calculate the temperature output parameters of each heating unit of the moxibustion device according to the heat distribution map, control the heating element with the optimized temperature output parameters, and generate the corresponding drive signal.

[0126] The aforementioned automatic temperature control system for a moxibustion device with human posture adaptation includes a physiological data acquisition module, a posture feature extraction module, a posture change recognition module, a heat distribution generation module, and a temperature adaptive control module. The physiological data acquisition module uses the built-in sensor array of the moxibustion device to acquire multi-dimensional human physiological indicators such as joint angles, skin contact point positions, and muscle tension, forming an initial physiological dataset. The posture feature extraction module performs deep learning network feature extraction on the initial physiological dataset to obtain baseline posture features for recognizing human posture reference patterns. The posture change recognition module performs feature analysis on real-time human physiological indicators based on the baseline posture features. The system matches and constructs posture change vectors, inputs these vectors into a machine learning classifier for classification and recognition, and obtains corresponding human posture change type labels. The heat distribution generation module calculates the variation values ​​of contact area and pressure distribution based on the change type labels, and combines them with a pre-built heat distribution model to dynamically adjust and calibrate the heat distribution, resulting in a heat distribution map containing temperature data points, spatial coordinates, temperature values, and the mapping relationship between the heating unit and the human body contact area. The temperature adaptive control module calculates the temperature output parameters of each heating unit of the moxibustion device based on the heat distribution map, optimizes the temperature output parameters, and generates corresponding drive signals to control the heating elements to perform heating actions.

[0127] This system, through the collaborative work of multiple modules, constructs a complete closed-loop control process from physiological signal acquisition to end-effector heating execution, enabling real-time perception and quantitative recognition of changes in human posture. It leverages deep learning and machine learning algorithms to improve the accuracy of posture feature extraction and change type determination, and uses a heat distribution model to dynamically adjust heat distribution to adapt to posture changes. A temperature adaptive control module then completes zoned and refined temperature output and execution control. The overall system effectively solves problems such as uneven temperature control and localized overheating caused by changes in human posture in traditional moxibustion devices, improving the accuracy, stability, and adaptability of temperature control, and enhancing the safety and comfort of the moxibustion therapy process.

[0128] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0129] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above-described method for automatic temperature control of a moxibustion device based on human posture adaptation.

[0130] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0131] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0132] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for automatic temperature control of a moxibustion device that adapts to human posture, characterized in that, The method includes: The user's multi-dimensional human physiological indicators were collected by the built-in sensor array of the moxibustion device to obtain an initial physiological dataset; the human physiological indicators include joint angles, skin contact point location information and muscle tension indicators. Deep learning network features are extracted from the initial physiological dataset to obtain baseline posture features for recognizing human posture baseline patterns; Based on the baseline posture features, feature matching is performed on the human physiological indicators to obtain a posture change vector. The posture change vector is then input into a machine learning classifier for classification to obtain a change type label. The variation values ​​of contact area and pressure distribution are calculated based on the change type labels, and the heat distribution is adjusted in combination with the pre-constructed heat distribution model to obtain a heat distribution map; The temperature output parameters of each heating unit of the moxibustion device are calculated based on the heat distribution diagram. The optimized temperature output parameters are used to control the heating element and generate the corresponding drive signal.

2. The method according to claim 1, characterized in that, The step of performing deep learning network feature extraction on the initial physiological dataset to obtain baseline posture features for recognizing human posture baseline patterns includes: A deep learning network is used to extract features from each time-series physiological signal in the initial physiological dataset to obtain a high-dimensional feature vector. The high-dimensional feature vector is subjected to dimensionality reduction processing to obtain a low-dimensional feature representation; Cluster analysis is performed based on the low-dimensional feature representation to identify several baseline pose feature clusters; If the number of the reference attitude feature clusters exceeds a preset threshold, the reference attitude feature clusters are merged to obtain merged feature clusters. The central feature vector corresponding to each feature cluster is calculated based on the merged feature clusters, and each central feature vector is used as the reference pose feature.

3. The method according to claim 2, characterized in that, The central feature vector corresponding to each feature cluster is calculated using the following formula: in, This represents the central eigenvector of the merged feature cluster. This represents the number of samples representing low-dimensional features in the merged feature cluster. Indicates the first The confidence weight vector, represented by a sensor array with low-dimensional features, is obtained by normalizing the signal integrity, temporal stability, and contact fit of the multi-dimensional sensors built into the moxibustion device. Indicates the first The posture feature dimension contribution weight vector, represented by several low-dimensional features, is quantified and assigned by the mutual information entropy values ​​of various physiological indicators, such as joint angles, muscle tension, and skin contact point positions, in relation to human posture representation. Represents the Hadamard product of vectors. Indicates the first The feature vector is represented by a low-dimensional feature representation. This represents the mean vector of all low-dimensional feature representations within the merged feature cluster. Indicates the first The Euclidean distance between each feature vector and the cluster mean vector This represents the outlier suppression coefficient, a pre-defined positive real number based on the human posture monitoring scenario of the moxibustion device, used to weight and attenuate outlier feature vectors within a cluster. This represents the Gaussian outlier penalty term.

4. The method according to claim 2, characterized in that, The posture change vector is obtained by feature matching of the human physiological indicators based on the baseline posture features. The posture change vector is then input into a machine learning classifier for classification to obtain change type labels, including: Based on the baseline posture features, feature matching of human physiological indicators is performed, and the cosine distance algorithm is used to calculate the deviation value after matching, and the effective deviation values ​​exceeding the preset threshold are screened out. The effective deviation values ​​of each dimension are constructed into vector dimensions and quantized and encoded to obtain a posture change vector that represents the changes in human posture during the use of the moxibustion device. The vector dimension is constructed by mapping and arranging the effective deviation values ​​according to the dimensional attributes of the human physiological indicators, resulting in a dimension-matched deviation data sequence. The feature quantization encoding involves normalizing and numerically quantizing the structured and arranged deviation data sequence to obtain a standardized feature vector. The posture change vector is input into a pre-trained machine learning classifier to perform human posture change type classification and recognition, and the classification and recognition results are obtained. Based on the classification and recognition results, the corresponding human posture change type label is output.

5. The method according to claim 1, characterized in that, The variation values ​​of contact area and pressure distribution are calculated based on the change type labels, and the heat distribution is adjusted in conjunction with a pre-constructed heat distribution model to obtain a heat distribution map, including: Extract the contact area variation data and pressure distribution variation data corresponding to the change type labels to obtain an initial set of variation values; The contact area variation data includes the amount of change in the contact area, the rate of change in the area, and the offset of the contact contour. The pressure distribution variation data includes the pressure value change, pressure change rate, and pressure center offset coordinates in the contact area. The initial set of variation values ​​is input into the pre-constructed heat distribution model to solve for the preliminary adjustment parameters of the heat distribution. The heat distribution model is constructed based on the radial basis function neural network algorithm; The weights and bias parameters of the heat distribution model are iteratively calibrated using the initial adjustment parameters and the backpropagation algorithm until the mean square error between the predicted heat distribution value and the benchmark reference value output by the model is less than a preset threshold, thus obtaining the adjusted heat distribution matrix. The heat distribution matrix is ​​divided according to the area of ​​moxibustion action to obtain several heat distribution sub-regions; If the variation value of the heat distribution sub-region exceeds the preset threshold, a secondary parameter calibration is performed on the heat distribution sub-region, and the calibrated heat distribution sub-regions are spliced ​​and merged to obtain a heat distribution map. The heat distribution map includes several temperature data points, the spatial coordinates corresponding to each temperature data point, the temperature value of each temperature data point, and the corresponding mapping relationship between the heating unit of the moxibustion device and the contact area of ​​the human body.

6. The method according to claim 1, characterized in that, The process of calculating the temperature output parameters of each heating unit of the moxibustion device based on the heat distribution map, controlling the heating element using the optimized temperature output parameters, and generating corresponding drive signals includes: Calculate the temperature output parameters of each heating unit based on the heat distribution diagram to form a dynamic temperature parameter sequence; Temperature smoothing optimization is performed based on the dynamic temperature parameter sequence combined with a preset temperature change rate constraint to obtain preliminary optimized temperature parameters. The preliminary optimized temperature parameters are subjected to regional adaptive weighting based on the area of ​​moxibustion action to generate optimized temperature parameters that adapt to changes in human posture. Based on the optimized temperature parameters, a drive signal corresponding to each of the heating units is generated; Each heating element performs a heating operation according to the corresponding drive signal.

7. A human posture-adaptive automatic temperature control system for moxibustion devices, characterized in that, The system includes: The physiological data acquisition module is used to acquire multi-dimensional human physiological indicators of the user collected by the built-in sensor array of the moxibustion device to obtain an initial physiological dataset; the human physiological indicators include joint angles, skin contact point location information and muscle tension indicators. The pose feature extraction module is used to perform deep learning network feature extraction on the initial physiological dataset to obtain baseline pose features for recognizing human pose baseline patterns. The posture change recognition module is used to perform feature matching on the human physiological indicators based on the baseline posture features to obtain a posture change vector, and input the posture change vector into a machine learning classifier for classification to obtain a change type label. The heat distribution generation module is used to calculate the variation values ​​of contact area and pressure distribution based on the change type label, and adjust the heat distribution by combining it with the pre-built heat distribution model to obtain a heat distribution map; The temperature adaptive control module is used to calculate the temperature output parameters of each heating unit of the moxibustion device according to the heat distribution map, control the heating element with the optimized temperature output parameters, and generate the corresponding drive signal.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, 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 6.

9. 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 6.

Images