Thermosensitive moxibustion tracking algorithm and system based on long-acting moxibustion therapy sensing and posture three-dimensional sensing

Abstract

The invention relates to the technical field of intelligent traditional Chinese medicine diagnosis and treatment, and discloses a thermosensitive moxibustion tracking algorithm and system based on long-acting moxibustion therapy sensing and posture three-dimensional sensing, and the algorithm comprises a long-acting flexible thermosensitive sensing subsystem, a three-dimensional posture sensing module, a multi-modal data fusion processing unit, a dynamic thermal field modeling algorithm, and a self-adaptive moxibustion applying control subsystem. Through full-chain innovation of long-acting sensing, three-dimensional sensing, coupling modeling and intelligent control, the performance stability problem (drift error is less than or equal to + / -0.3 DEG C) of the flexible sensor under long-term work (more than or equal to 72 hours) is solved; establishing a posture-thermal field-time four-dimensional coupling model, and realizing submillimeter thermosensitive point tracking under a dynamic posture; a millisecond-level response (less than or equal to 50ms) closed-loop control system is constructed, and the adaptability to complex clinical scenes is improved; therefore, the key technical problem in the dynamic moxibustion therapy scene is systematically solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent traditional Chinese medicine diagnosis and treatment technology, and in particular to a thermosensitive moxibustion tracking algorithm and system based on long-acting moxibustion therapy sensing and three-dimensional body posture perception. Background Art

[0002] Traditional thermosensitive moxibustion technology has the following problems:

[0003] 1. Existing flexible sensors experience drift errors (≥±1.2°C) after continuous operation for more than 24 hours, and their poor air permeability reduces long-term comfort.

[0004] 2. Dynamic body position changes (such as turning over, flexion and extension) cause the spatial offset of the thermal sensitive point to exceed 5mm, and the thermal field distribution distortion error to exceed 1.5℃;

[0005] 3. The sensing, computing, and execution modules lack millisecond-level coordination, and the system response delay exceeds 300ms.

[0006] The existing technology proposes a dynamic thermal field compensation solution to solve the above three problems, but it still has the following shortcomings:

[0007] ① The long-term stability of the sensor was not considered, and the accuracy of thermal point recognition dropped by 37% after 8 hours;

[0008] ② The use of two-dimensional positioning technology ignores the three-dimensional surface features of the human body, and the deviation of the moxibustion angle leads to a 40% reduction in heat penetration efficiency;

[0009] ③ The robotic arm control strategy lacks the ability to predict body posture, and the burn rate in scenarios with sudden body position changes is as high as 6.8%.

[0010] To this end, we propose an intelligent moxibustion therapy system and algorithm that integrates long-lasting flexible thermal sensing technology, real-time three-dimensional body posture perception, and dynamic thermal field modeling. Through multimodal data fusion and adaptive control strategies, we solve the problems of monitoring distortion, positioning offset, and control lag caused by sensor performance degradation and dynamic changes in body position in traditional moxibustion therapy, and achieve precise moxibustion in complex clinical scenarios. Summary of the Invention

[0011] The purpose of the present invention is to solve the shortcomings of the existing technology mentioned in the background technology, and to propose an intelligent moxibustion system and algorithm that integrates long-term flexible thermal sensing technology, three-dimensional body real-time perception and dynamic thermal field modeling. Through multimodal data fusion and adaptive control strategy, it solves the monitoring distortion, positioning offset and control lag problems caused by sensor performance attenuation and dynamic changes in body position in traditional moxibustion therapy, and realizes precise moxibustion in complex clinical scenarios.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] The thermal moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception includes:

[0014] Long-lasting flexible thermal sensing array, integrating temperature and bioimpedance sensors, supporting ≥72 hours of continuous monitoring;

[0015] A three-dimensional posture perception device, including a depth camera, a flexible strain sensor, and an inertial measurement unit;

[0016] Multimodal data fusion processor to achieve spatiotemporal registration and coupled modeling of body posture and thermal data;

[0017] Dynamic thermal field modeling unit, solving the unsteady heat conduction equation related to body shape;

[0018] The six-degree-of-freedom moxibustion robotic arm has sub-millimeter trajectory tracking capabilities;

[0019] Edge computing platform, deploys reinforcement learning control algorithms and security protection mechanisms.

[0020] As a further step in the present invention, the long-lasting flexible thermal sensing array:

[0021] Adopt microporous breathable structure, pore size 80-200μm, porosity ≥45%;

[0022] Integrated temperature drift self-calibration circuit, long-term stability error ≤±0.25℃;

[0023] The serpentine silver nanowire routing layout can withstand ≥20% tensile deformation.

[0024] As a further embodiment of the present invention, the three-dimensional posture sensing device includes:

[0025] TOF depth camera, resolution 1280×720@60fps, depth accuracy ±0.8mm;

[0026] Liquid metal flexible strain sensor, stretching rate ≥ 180%, resistance change rate ≤ 0.5% / time;

[0027] 9-axis IMU unit, angle measurement error ≤ 0.3°, sampling frequency ≥ 200Hz.

[0028] As a further step in the present invention, the multimodal data fusion processor, its processing method includes:

[0029] Extended Kalman filter aligns multi-source data timestamps, with synchronization error ≤ 0.8ms;

[0030] Establish anatomical coordinate system mapping relationship based on feature point propagation algorithm;

[0031] Construct a four-dimensional space-time tensor T(x,y,z,t,θ,ε).

[0032] As a further step in the present invention, the dynamic thermal field modeling unit has a modeling method as follows:

[0033] Surface layer: Improved Kriging interpolation generates 0.5mm 2 Resolution temperature distribution;

[0034] Subcutaneous layer: Finite element method is used to solve the body-related heat conduction equation, with a grid size of 0.2 mm 3 ;

[0035] Real-time inversion of thermal conductivity parameters K(T,Z)=K0+αT+βZ.

[0036] As a further step in the present invention, the control strategy of the six-degree-of-freedom moxibustion robotic arm includes:

[0037] Optimize the moxibustion path based on model predictive control (MPC);

[0038] Impedance control maintains a constant moxibustion pressure (3N±0.2N);

[0039] Emergency braking response time ≤ 20ms, braking distance accuracy ± 0.3mm.

[0040] As a further step in the present invention, the security protection module:

[0041] Dual redundant temperature detection channels, triggering shutdown when the difference is >0.5℃;

[0042] Dynamic safety fences are generated based on real-time point cloud expansion of 3mm;

[0043] Three-stage gradient cooling mechanism with adjustable cooling rate (1-3℃ / s).

[0044] As a further step in the present invention, a thermosensitive moxibustion tracking algorithm based on long-term moxibustion therapy sensing and three-dimensional body posture perception includes:

[0045] Obtain the temperature matrix T(x,y,t) and impedance matrix Z(x,y,t) through the long-term sensor array;

[0046] The 3D posture sensing device captures joint angles θ, muscle strains ε, and body surface point clouds;

[0047] Multimodal data spatiotemporal registration to generate coupled model input tensors;

[0048] Solve the body-related heat conduction equation and predict the future thermal field distribution at Δt=5s;

[0049] The reinforcement learning controller generates the optimal moxibustion trajectory and minimizes the energy function:

[0050] J=\int_0^t\keft(\|T_{ref}-T\|^2+λ\|\nable(T_{ref}-T)\|^2\right)dt.

[0051] As a further step in the present invention, the thermal field distribution prediction model training method includes:

[0052] Network architecture: Physically guided spatiotemporal graph convolutional network (PG-STGCN);

[0053] Input features: historical 10s thermal field sequence + real-time body posture parameters;

[0054] Loss function: weighted mean square error + thermodynamic constraint (Fourier's law).

[0055] As a further aspect of the present invention, a computer-readable storage medium stores program code for executing an algorithm.

[0056] As a further aspect of the present invention, the computer-readable storage medium includes:

[0057] Multi-threaded architecture: data acquisition (5ms), calculation (10ms), and control (2ms) threads run in parallel;

[0058] Supports ROS and TCM knowledge base API docking, calling the acupoint database to optimize path planning.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] Temperature drift of 120 hours of continuous operation is ≤±0.28℃ (4 times higher than traditional temperature control);

[0061] The microporous structure increases the skin's breathability by 60% and reduces discomfort by 75%;

[0062] The tracking error of thermal points in dynamic body positions is ≤0.8mm (traditional method ≥5mm);

[0063] Response time to sudden changes in body position ≤45ms (shortened by 85%);

[0064] Treatment efficiency increased by 2.3 times (single treatment time from 25 minutes to 11 minutes);

[0065] The scalding accident rate is reduced to below 0.2% (6.8% for traditional systems). DETAILED DESCRIPTION

[0066] The present invention may be more readily understood by referring to the following detailed description of preferred embodiments of the present invention and the included Examples. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention pertains. In the event of a conflict, the definitions in this specification shall prevail.

[0067] When amount, concentration or other value or parameter is represented with range, preferred range or the range that a series of upper preferred value and lower preferred value limit are expressed, this should be understood as specifically disclosing all ranges formed by any pairing of any range upper limit or preferred value and any range lower limit or preferred value, no matter whether this range is disclosed separately.For example, when disclosing scope "1 to 5", described scope should be interpreted as including scope "1 to 4", "1 to 3", "1 to 2", "1 to 2 and 4 to 5", "1 to 3 and 5" etc.When numerical range is described in this article, unless otherwise stated, otherwise this scope is intended to include its end value and all integers and fractions within this range.

[0068] The singular includes plural references unless the context clearly dictates otherwise. "Optional" or "either" means that the subsequently described event or incident can or cannot occur, and that the description includes instances where the event occurs and instances where it does not.

[0069] Approximating terms in the specification and claims are used to modify a quantity, indicating that the invention is not limited to that specific quantity and includes acceptable modifications close to that quantity without altering the basic function. Accordingly, the use of "approximately" or "about" to modify a numerical value indicates that the invention is not limited to that exact value. In some cases, approximate terms may refer to the precision of the instrument used to measure the value. In the specification and claims of this application, range definitions may be combined and / or interchanged, and unless otherwise indicated, such ranges include all subranges contained therein.

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0071] The thermal moxibustion tracking algorithm and system based on long-term moxibustion therapy sensing and three-dimensional body posture perception consists of five core modules:

[0072] 1. Long-lasting flexible thermal sensing subsystem (hardware innovation)

[0073] (1) Multilayer heterostructure

[0074] ①Surface layer: microporous breathable polyurethane membrane (pore size 80-200μm, porosity ≥45%), allowing continuous attachment for ≥120 hours;

[0075] ②Middle layer: serpentine silver nanowire temperature sensor array (256 channels, accuracy ±0.1°C);

[0076] ③ Bottom layer: bioimpedance detection electrodes (32 channels, adjustable frequency 100kHz-1MHz);

[0077] (2) Self-calibration circuit

[0078] ① Temperature drift compensation: Zero point calibration is performed every 30 minutes, and the long-term stability error is ≤±0.25℃;

[0079] ② Dynamic impedance correction: real-time baseline correction algorithm based on reference impedance loop.

[0080] 2. 3D body perception module (multi-source sensor fusion)

[0081] (1) Depth Vision Unit

[0082] ①TOF camera (940nm VCSEL light source, depth resolution ±0.8mm, frame rate 60fps);

[0083] ② Real-time generation of body surface 3D point cloud (density ≥ 500 points / cm 2 );

[0084] (2) Biomechanical sensing

[0085] ① Flexible strain sensor array (liquid metal-silicone composite structure, stretching rate ≥180%);

[0086] ②9-axis IMU (three-axis gyroscope + accelerometer + magnetometer, angle error ≤ 0.3°);

[0087] (3) Dynamic modeling algorithm

[0088] ①Real-time reconstruction of body surface mesh based on the improved non-rigid ICP algorithm (error ≤ 0.6mm);

[0089] ②The spatiotemporal convolutional network (ST-CNN) is used to predict the body posture evolution trend in the next 0.5s.

[0090] 3. Multimodal data fusion processing unit (algorithm innovation)

[0091] (1) Spatiotemporal registration mechanism

[0092] ① Time synchronization: Extended Kalman filter aligns multi-source data streams (synchronization error ≤ 0.8ms);

[0093] ② Spatial mapping: constructing a dynamic association model of anatomical feature points (registration error ≤ 0.5 mm);

[0094] (2) Coupled modeling equations:

[0095] \frac{\partialT}{\partialt}=\nabla\cdot\left[D(\theta,\varepsilon)\nabl aT\right]+Q_{moxa}(x,y,z,t)\quad(1), where D(θ,ε)=D0+K1θ+K2ε' is the posture-dependent heat diffusion tensor; θ is the joint angle, and ε is the muscle strain rate.

[0096] 4. Dynamic thermal field modeling algorithm (computational innovation)

[0097] (1) Surface thermal field reconstruction

[0098] ① Improved Kriging interpolation algorithm, which introduces anisotropic kernel function to compensate for human body surface distortion;

[0099] ②Spatial resolution increased to 0.5mm 2 (4 times higher than traditional methods);

[0100] (2) Subcutaneous heat conduction inversion

[0101] ① Solve equation (1) based on the finite element method, with a meshing accuracy of 0.2 mm 3 ;

[0102] ②Dynamically correct tissue thermal conductivity parameters based on bioimpedance data.

[0103] 5. Adaptive moxibustion control subsystem (control innovation)

[0104] (1) Reinforcement Learning Controller:

[0105] ① State space: {thermal gradient distribution, body parameters, historical trajectory, safety mark};

[0106] ②Action space: {manipulator arm position (x, y, z, α, β, γ), moxibustion pressure (2-5N), movement speed (0.1-3cm / s)};

[0107] ③Reward function:

[0108] R=e^{-\|T_{targ}-T_{real}\|^2}-λ_1\|\nabla(T_{targ}-T_{real})\|^2-λ_2\|u_{change}\|^2\quad(2);

[0109] (2) Security protection mechanism

[0110] ① Three levels of warning: Level I (epidermis>43°C), Level II (subcutaneous 3mm>40°C), Level III (dT / dt>1.2°C / s);

[0111] ② Dynamic safety fence: Construct a 3D restricted area based on real-time point cloud data with a 3mm expansion.

[0112] Based on the above five core modules, the more specific solutions of the present invention are as follows:

[0113] The thermal moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception includes:

[0114] Long-lasting flexible thermal sensing array, integrating temperature and bioimpedance sensors, supporting ≥72 hours of continuous monitoring;

[0115] A three-dimensional posture perception device, including a depth camera, a flexible strain sensor, and an inertial measurement unit;

[0116] Multimodal data fusion processor to achieve spatiotemporal registration and coupled modeling of body posture and thermal data;

[0117] Dynamic thermal field modeling unit, solving the unsteady heat conduction equation related to body shape;

[0118] The six-degree-of-freedom moxibustion robotic arm has sub-millimeter trajectory tracking capabilities;

[0119] Edge computing platform, deploys reinforcement learning control algorithms and security protection mechanisms.

[0120] In its system, the long-lasting flexible thermal sensor array adopts a microporous breathable structure with a pore size of 80-200μm and a porosity of ≥45%; an integrated temperature drift self-calibration circuit with a long-term stability error of ≤±0.25℃; and a serpentine silver nanowire routing layout that can withstand a tensile deformation of ≥20%.

[0121] In its system, the three-dimensional posture perception device is configured as follows: TOF depth camera with a resolution of 1280×720@60fps and a depth accuracy of ±0.8mm; liquid metal flexible strain sensor with a stretching rate ≥180% and a resistance change rate ≤0.5% / time; 9-axis IMU unit with an angle measurement error ≤0.3° and a sampling frequency ≥200Hz.

[0122] In its system, the multimodal data fusion processor has the following processing methods: extended Kalman filtering to align the timestamps of multi-source data, with a synchronization error of ≤0.8ms; establishing an anatomical coordinate system mapping relationship based on a feature point propagation algorithm; and constructing a four-dimensional space-time tensor T(x, y, z, t, θ, ε).

[0123] In the system, the dynamic thermal field modeling unit has the following modeling method:

[0124] Surface layer: Improved Kriging interpolation generates 0.5mm 2 Resolution temperature distribution;

[0125] Subcutaneous layer: Finite element method is used to solve the body-related heat conduction equation, with a grid size of 0.2 mm 3 ;

[0126] Real-time inversion of thermal conductivity parameters K(T,Z)=K0+αT+βZ.

[0127] In its system, the six-degree-of-freedom moxibustion robotic arm has the following control strategies: optimizing the moxibustion path based on model predictive control (MPC); impedance control to maintain a constant moxibustion pressure (3N±0.2N); emergency braking response time ≤20ms, and braking distance accuracy ±0.3mm.

[0128] In its system, the safety protection module has: dual redundant temperature detection channels, which trigger shutdown when the difference is >0.5℃; dynamic safety fence is generated based on real-time point cloud expansion of 3mm; three-level gradient cooling mechanism, with adjustable cooling rate (1-3℃ / s).

[0129] The thermal moxibustion tracking algorithm based on long-term moxibustion therapy sensing and three-dimensional body posture perception includes:

[0130] Obtain the temperature matrix T(x,y,t) and impedance matrix Z(x,y,t) through the long-term sensor array;

[0131] The 3D posture sensing device captures joint angles θ, muscle strains ε, and body surface point clouds;

[0132] Multimodal data spatiotemporal registration to generate coupled model input tensors;

[0133] Solve the body-related heat conduction equation and predict the future thermal field distribution at Δt=5s;

[0134] The reinforcement learning controller generates the optimal moxibustion trajectory and minimizes the energy function:

[0135] J=\int_0^t\left(\|T_{ref}-T\|^2+λ\|\nabla(T_{ref}-T)\|^2\right)dt.

[0136] In its algorithm, the thermal field distribution prediction model training method includes:

[0137] Network architecture: Physically guided spatiotemporal graph convolutional network (PG-STGCN);

[0138] Input features: historical 10s thermal field sequence + real-time body parameters;

[0139] Loss function: weighted mean square error + thermodynamic constraint (Fourier's law).

[0140] A computer-readable storage medium storing program code for executing an algorithm, the computer-readable storage medium comprising:

[0141] Multi-threaded architecture: data acquisition (5ms), calculation (10ms), and control (2ms) threads run in parallel;

[0142] Supports ROS and TCM knowledge base API docking, calling the acupoint database to optimize path planning.

[0143] Based on the above specific solution content, the present invention includes the following embodiment content:

[0144] Example 1: Hardware system implementation

[0145] 1. Hardware Configuration

[0146] (1) Sensing layer

[0147] ①Long-lasting thermal array: 8×8cm 2 Region, 256 temperature + 32 impedance channels, thickness 0.3mm;

[0148] ②Body perception: Intel RealSense D455 depth camera + 128-channel liquid metal strain suit;

[0149] (2) Computational layer:

[0150] ①NVIDIA Jetson AGX Xavier (GPU 512 CUDA cores, 64TOPS computing power)

[0151] ②Real-time operating system: Ubuntu 20.04 + ROS2Foxy

[0152] (3) Execution layer:

[0153] ①UR5e collaborative robot arm (repeatability accuracy ±0.03mm);

[0154] ② Constant force moxibustion head module (pressure range 2-5N, resolution 0.1N);

[0155] 2. Communication Protocol

[0156] (1) Sensor data transmission: TDMA time division multiple access protocol, period 5ms;

[0157] (2) Control command transmission: EtherCAT bus, delay ≤1ms.

[0158] Example 2: Algorithm Optimization Solution

[0159] 1. Thermal field prediction model training

[0160] (1) Dataset: 100,000 sets of clinical data (including 8 typical body position change scenarios);

[0161] (2) Network structure:

[0162] ① Encoder: 4 layers of spatiotemporal graph convolution (64-128-256-512 channels per layer);

[0163] ②Decoder: 3 layers of deconvolution + physical constraint layer;

[0164] (3) Training parameters:

[0165] ① Initial learning rate 0.001, cosine annealing decay;

[0166] ②Batch size 32, training epochs 200;

[0167] 2. Control strategy optimization

[0168] (1) Reinforcement learning parameters:

[0169] ① State space dimension: 256 (thermal field) + 18 (body) + 20 (history) = 294;

[0170] ② Action space dimensions: 6 (posture) + 1 (pressure) + 1 (speed) = 8;

[0171] ③ Discount factor γ = 0.99, exploration rate ε = 0.1 → 0.01 decay.

[0172] Example 3: Clinical application scenario

[0173] 1. Scenario 1: Treatment with turning over in supine position

[0174] (1) Initial state: The patient lies prone and moxibustion is applied to the Yaoyangguan acupoint;

[0175] (2) Posture change: hip joint angle change Δθ = 15° (lasting 2 s) was detected;

[0176] (3) System response:

[0177] ①Predict the thermal field deviation direction in the next 3 seconds;

[0178] ② Adjust the movement trajectory of the moxibustion head (curvature radius from 5mm to 8mm);

[0179] ③ Reduce the moxibustion power from 18W to 12W;

[0180] 2. Scenario 2: Sudden muscle tremor

[0181] (1) Trigger condition: IMU detects 4-6Hz high-frequency vibration (Parkinson's disease patient scenario)

[0182] (2) System response:

[0183] ① Start the vibration compensation algorithm and update the control period to 1ms;

[0184] ②Switch to impedance control mode, contact force fluctuation ≤±0.2N;

[0185] ③The safety module is activated, limiting the acceleration of the robot arm to ≤2m / s 2 .

[0186] In summary, this technical solution overcomes the performance stability problem (drift error ≤±0.3℃) of flexible sensors under long-term operation (≥72 hours) through full-chain innovation of long-term sensing-three-dimensional perception-coupling modeling-intelligent control; establishes a four-dimensional coupling model of body posture-thermal field-time to achieve submillimeter-level thermal sensitive point tracking under dynamic body posture; constructs a closed-loop control system with millisecond-level response (≤50ms) to improve adaptability to complex clinical scenarios; and systematically solves key technical problems in dynamic moxibustion scenarios.

[0187] The examples referred to herein are merely illustrative and serve to illustrate some features of the method of the present invention, and the appended claims are intended to claim the widest possible scope that can be imagined, and the embodiments presented herein are merely illustrative of selected implementation methods according to the combination of all possible embodiments. Therefore, it is the applicant's intention that the appended claims are not limited by the selection of examples illustrating the features of the present invention. Some numerical ranges used in the claims also include subranges therein, and changes in these ranges should also be interpreted as being covered by the appended claims where possible.

Claims

1. The thermal moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception is characterized by: include: Long-lasting flexible thermal sensing array, integrating temperature and bioimpedance sensors, supporting ≥72 hours of continuous monitoring; A three-dimensional posture perception device, including a depth camera, a flexible strain sensor, and an inertial measurement unit; Multimodal data fusion processor to achieve spatiotemporal registration and coupled modeling of body posture and thermal data; Dynamic thermal field modeling unit, solving the unsteady heat conduction equation related to body shape; The six-degree-of-freedom moxibustion robotic arm has sub-millimeter trajectory tracking capabilities; Edge computing platform, deploys reinforcement learning control algorithms and security protection mechanisms.

2. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: Long-lasting flexible thermal sensing array: Adopt microporous breathable structure, pore size 80-200μm, porosity ≥45%; Integrated temperature drift self-calibration circuit, long-term stability error ≤±0.25℃; The serpentine silver nanowire routing layout can withstand ≥20% tensile deformation.

3. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: A three-dimensional posture sensing device, comprising: TOF depth camera, resolution 1280×720@60fps, depth accuracy ±0.8mm; Liquid metal flexible strain sensor, stretching rate ≥ 180%, resistance change rate ≤ 0.5% / time; 9-axis IMU unit, angle measurement error ≤ 0.3°, sampling frequency ≥ 200Hz.

4. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: A multimodal data fusion processor, wherein the processing method includes: Extended Kalman filter aligns multi-source data timestamps, with synchronization error ≤ 0.8ms; Establish anatomical coordinate system mapping relationship based on feature point propagation algorithm; Construct a four-dimensional space-time tensor T(x,y,z,t,θ,ε).

5. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: Dynamic thermal field modeling unit, its modeling method is: Surface layer: Improved Kriging interpolation generates 0.5mm 2 Resolution temperature distribution; Subcutaneous layer: Finite element method is used to solve the body-related heat conduction equation, with a grid size of 0.2 mm 3 ; Real-time inversion of thermal conductivity parameters K(T,Z)=K0+αT+βZ.

6. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: The control strategy of the six-degree-of-freedom moxibustion robot arm includes: Optimize the moxibustion path based on model predictive control (MPC); Impedance control maintains a constant moxibustion pressure (3N±0.2N); Emergency braking response time ≤ 20ms, braking distance accuracy ± 0.3mm.

7. The thermosensitive moxibustion tracking system based on long-term moxibustion therapy sensing and three-dimensional body posture perception according to claim 1 is characterized in that: Security protection module: Dual redundant temperature detection channels, triggering shutdown when the difference is >0.5℃; Dynamic safety fences are generated based on real-time point cloud expansion of 3mm; Three-stage gradient cooling mechanism with adjustable cooling rate (1-3℃ / s).

8. The thermosensitive moxibustion tracking algorithm based on long-acting moxibustion therapy sensing and three-dimensional body posture perception according to any one of claims 1 to 7, characterized in that: include: Obtain the temperature matrix T(x,y,t) and impedance matrix Z(x,y,t) through the long-term sensor array; The 3D posture sensing device captures joint angles θ, muscle strains ε, and body surface point clouds; Multimodal data spatiotemporal registration to generate coupled model input tensors; Solve the body-related heat conduction equation and predict the future thermal field distribution at Δt=5s; The reinforcement learning controller generates the optimal moxibustion trajectory and minimizes the energy function: J=\int_0^t\left(\|T_{ref}-T\|^2+λ\|\nabla(T_{ref}-T)\|^2\right)dt.

9. The thermosensitive moxibustion tracking algorithm based on long-acting moxibustion therapy sensing and three-dimensional body posture perception according to claim 8 is characterized in that: The thermal field distribution prediction model training method includes: Network architecture: Physically guided spatiotemporal graph convolutional network (PG-STGCN); Input features: historical 10s thermal field sequence + real-time body posture parameters; Loss function: weighted mean square error + thermodynamic constraint term.

10. The computer-readable storage medium according to any one of claims 8 to 9, wherein: The computer-readable storage medium stores program code for executing an algorithm, which includes: Multi-threaded architecture: data acquisition (5ms), calculation (10ms), and control (2ms) threads run in parallel; Supports ROS and TCM knowledge base API docking, calling the acupoint database to optimize path planning.