Multi-sensor fusion-based warm traction dynamic optimization method and system

Abstract

The invention relates to the related technical field of traction physiotherapy, in particular to a warm traction dynamic optimization method and system based on multi-sensor fusion, and the method comprises the steps: obtaining multi-modal physiological feature parameters, determining a first traction sensitive feature set and a second traction sensitive feature set, carrying out the correlation verification through combining historical treatment records with the subjective comfort score of a patient, and obtaining a first traction sensitive feature set and a second traction sensitive feature set; and correcting the deviation of the traction force regulation and control function to obtain an optimal traction force output parameter, and driving an execution mechanism to adjust the thermal therapy temperature and the mechanical traction force. The technical problem that most traction regulation and control functions are set in a fixed mode and cannot meet muscle group characteristics, tolerance thresholds and treatment process differences of different patients is solved, multi-modal parameters such as surface myoelectricity and infrared thermal imaging are fused, first and second traction sensitive characteristic sets under heat stress dominance and muscle fatigue dominance are constructed, and the traction sensitivity of the patient is improved. Heat stress and muscle fatigue are accurately recognized, the stability and reliability of traction force regulation and control are guaranteed, and the technical effects that the risk in the warm traction therapy process is effectively reduced are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the related technical field of traction physiotherapy, in particular to a warm traction force dynamic optimization method and system based on multi-sensor fusion. BACKGROUND

[0002] Warm traction therapy is a rehabilitation method for musculoskeletal system diseases. Common musculoskeletal system diseases include cervical spondylosis and lumbar disc herniation. Warm traction therapy is usually in a static treatment mode. However, the static treatment mode cannot meet the differences in muscle group characteristics, tolerance threshold and treatment process of different patients. Single monitoring of mechanical traction force or body surface temperature cannot comprehensively reflect the dual physiological states of thermal stress and muscle fatigue. The traction force regulation function is mostly a fixed formula, which is prone to over-treatment problems and insufficient treatment problems. In addition, physiological signals are easily affected by motion artifacts and environmental noise, and lack effective noise filtering mechanisms, resulting in low feature recognition accuracy and further affecting the accuracy of traction force regulation, which seriously restricts the intelligent and personalized level of warm traction therapy.

[0003] In summary, the prior art has the technical problem that the traction force regulation function is mostly fixed, which cannot meet the differences in muscle group characteristics, tolerance threshold and treatment process of different patients, and is prone to feature recognition deviation caused by motion artifacts and environmental noise, which further affects the accuracy of traction force regulation. SUMMARY

[0004] The present application provides a warm traction force dynamic optimization method and system based on multi-sensor fusion, which aims to solve the technical problem in the prior art that the traction force regulation function is mostly fixed, which cannot meet the differences in muscle group characteristics, tolerance threshold and treatment process of different patients, and is prone to feature recognition deviation caused by motion artifacts and environmental noise, which further affects the accuracy of traction force regulation.

[0005] In view of the above problems, the technical solution of the present application is as follows: The first aspect of the present application provides a warm traction force dynamic optimization method based on multi-sensor fusion, wherein the method comprises: acquiring multi-modal physiological characteristic parameters including surface electromyogram, infrared thermal imaging data and skin impedance data according to a target muscle group region; determining a first traction sensitive feature set under thermal stress dominance and a second traction sensitive feature set under muscle fatigue dominance based on the multi-modal physiological characteristic parameters; verifying the relevance based on the first traction sensitive feature set and the second traction sensitive feature set, combining historical treatment records and patient subjective comfort scores, correcting the deviation of the traction force regulation function, and obtaining optimal traction force output parameters; generating a warm traction force dynamic optimization instruction according to the optimal traction force output parameters, and driving an execution mechanism to adjust the thermal therapy temperature and the mechanical traction force through the warm traction force dynamic optimization instruction.

[0006] Preferably, by combining the infrared thermal imaging data and skin impedance data in the multimodal physiological characteristic parameters with the current treatment stage for coupling analysis, the first traction sensitive feature set under heat stress is determined; at the same time, in the heat conduction field corresponding to the target muscle group area, the partial differential relationship between heat energy penetration depth and temperature gradient is analyzed to determine the dynamic thermal response coefficient as the thermal diffusion kinetic index.

[0007] Preferably, by combining the surface electromyography (EMG) signal and skin impedance data from the multimodal physiological characteristic parameters with the current traction intensity, a second traction-sensitive feature set under muscle fatigue is determined; simultaneously, the ratio of the integral EMG value of the surface EMG signal during the traction cycle to the phase angle change rate in the skin impedance data is obtained, and the ratio is normalized and used as a muscle-thermal synergistic stability factor.

[0008] Preferably, a traction-sensitive feature matrix is ​​constructed based on the first traction-sensitive feature set and the second traction-sensitive feature set; the row vectors of the traction-sensitive feature matrix correspond to physiological feedback samples under multiple treatment time windows, and the column vectors include the thermal distribution entropy and thermal imaging feature region intensity in the first traction-sensitive feature set, and the electromyographic fluctuation coefficient and impedance phase change amplitude in the second traction-sensitive feature set.

[0009] Preferably, the thermal diffusion dynamics index is used as the query vector of the attention network, the muscle-thermal synergistic stability factor is used as the key vector of the attention network, and the traction-sensitive feature matrix is ​​used as the value vector of the attention network; the key traction-sensitive features under the weight distribution are dynamically focused through the attention mechanism.

[0010] Preferably, the attention mechanism uses a similarity metric based on kernel density estimation instead of inner product operation; the similarity index is obtained by obtaining the kernel density overlap area between the query vector and the key vector: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area between the kernel density curve corresponding to the query vector and the kernel density curve corresponding to the key vector is calculated.

[0011] Preferably, the similarity index includes a first similarity threshold and a second similarity threshold; if the first similarity threshold is met, the attention mechanism tilts the weight distribution towards the low-frequency temporal variation components of thermal distribution entropy and electromyographic fluctuation coefficient; if the second similarity threshold is met, the attention mechanism tilts the weight distribution towards the high-frequency transient change components of the intensity and impedance phase change amplitude of the thermal imaging feature region.

[0012] Preferably, the low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the traction-sensitive feature matrix through empirical mode decomposition, which respectively correspond to the low-order and high-order components in multiple intrinsic mode functions, and the instantaneous amplitude and frequency features are extracted by combining Hilbert transform, and differential weighted fusion is performed.

[0013] Preferably, the key traction-sensitive features after differential weighting are input into the gated circulation unit; the gated circulation unit suppresses physiological noise interference, and a traction force regulation function is constructed based on the output features of the gated circulation unit to dynamically output the thermotherapy temperature setpoint and the mechanical traction force amplitude.

[0014] In a second aspect, this application provides a dynamic optimization system for thermo-traction force based on multi-sensor fusion, wherein the system comprises: a multimodal physiological feature parameter acquisition module: acquiring multimodal physiological feature parameters including surface electromyography signals, infrared thermal imaging data, and skin impedance data based on the target muscle group region; a first and second traction-sensitive feature set determination module: determining a first traction-sensitive feature set under heat stress and a second traction-sensitive feature set under muscle fatigue based on the multimodal physiological feature parameters; a correlation verification module: performing correlation verification based on the first and second traction-sensitive feature sets, combined with historical treatment records and patient subjective comfort scores, correcting deviations in the traction force control function, and obtaining optimal traction force output parameters; and an actuator driving module: generating a dynamic optimization command for thermo-traction force based on the optimal traction force output parameters, and driving the actuator to adjust the thermotherapy temperature and mechanical traction force through the dynamic optimization command for thermo-traction force.

[0015] In summary, one or more technical solutions provided in this application achieve the following: They integrate multimodal parameters such as surface electromyography and infrared thermography to construct a first traction-sensitive feature set dominated by heat stress and a second traction-sensitive feature set dominated by muscle fatigue. This enables accurate identification of heat stress and muscle fatigue, ensuring the stability and reliability of traction force regulation. By combining historical treatment records and subjective comfort scores to correct the regulation function, the accuracy of thermo-traction force regulation is improved, and treatment parameters are dynamically adapted to individual patient needs, effectively reducing the technical risks of thermo-traction therapy. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1A flowchart illustrating the dynamic optimization method for thermal traction force based on multi-sensor fusion is provided for this application.

[0018] Figure 2 This application provides a structural schematic diagram of a dynamic optimization system for thermal traction force based on multi-sensor fusion.

[0019] Figure labeling: Multimodal physiological feature parameter acquisition module M100, first and second traction sensitive feature set determination module M200, correlation verification module M300, actuator drive module M400. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0021] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a method for dynamic optimization of thermal traction force based on multi-sensor fusion, wherein the method includes: S1: Based on the target muscle group region, acquire multimodal physiological characteristic parameters including surface electromyography signals, infrared thermal imaging data, and skin impedance data; S2: Based on the multimodal physiological characteristic parameters, determine the first traction-sensitive feature set under heat stress and the second traction-sensitive feature set under muscle fatigue.

[0022] Specifically, multimodal physiological characteristic parameters refer to different types of data that reflect the physiological state of the human body, acquired through various sensors. These include surface electromyography (EMG) signals, infrared thermography data, and skin impedance data. Specifically, EMG signals are electrical signals generated during muscle contraction, acquired through electrode patches, reflecting the intensity of muscle activity and the degree of fatigue. Infrared thermography data, acquired through a thermal imager, reflects the temperature distribution on the human body surface and can be used to assess heat stress. Skin impedance data, measured through electrophysiological sensors, reflects the electrical properties of the skin and is related to skin moisture, temperature, and metabolic state. The traction-sensitive feature set refers to the set of key features extracted from the multimodal physiological characteristic parameters that are closely related to the effect of traction therapy. Among them, the first traction-sensitive feature set dominated by heat stress mainly reflects the human body's response to heat during thermotherapy; the second traction-sensitive feature set dominated by muscle fatigue mainly reflects the characteristics of muscle fatigue during traction.

[0023] Execution steps: Based on the target muscle group area, surface electromyography (EMG) signals, infrared thermography data, and skin impedance data are acquired through sensors. The acquisition of multimodal physiological characteristic parameters is real-time, which can comprehensively reflect the physiological state of the target muscle group during thermotherapy and traction. Furthermore, the frequency and amplitude of the surface EMG signal can reflect the activity intensity and fatigue level of the muscle. When the frequency and amplitude of the EMG signal decrease, it indicates that the muscle may be in a state of fatigue. The infrared thermography data visually displays the heat stress area through a temperature distribution map. When the temperature gradient change in the heat stress area is significant, it indicates that the thermotherapy effect is good. The skin impedance data reflects the skin's moisture and metabolic state by monitoring changes in the skin's electrical properties. When the skin impedance decreases, it indicates that the skin moisture increases and the heat stress effect is good.

[0024] By coupling and analyzing these multimodal physiological characteristic parameters, a first traction-sensitive feature set dominated by heat stress and a second traction-sensitive feature set dominated by muscle fatigue were determined, accurately identifying the physiological characteristics of heat stress and muscle fatigue. Furthermore, by analyzing the phase angle change rate of the integrated electromyography (EMG) value and skin impedance data, it was found that when the integrated EMG value increases and the phase angle change rate exceeds a certain range, the degree of muscle fatigue increases significantly. At this point, this feature can be identified as part of the second traction-sensitive feature set. By analyzing the temperature gradient change of infrared thermal imaging data, if the temperature gradient reaches a threshold value, it indicates a significant heat stress effect, which can be used as an important feature of the first traction-sensitive feature set. Through this comprehensive analysis of multimodal data, the limitations of single-parameter monitoring can be effectively avoided, the accuracy of feature identification can be improved, and thus a more reliable basis for traction force regulation can be provided.

[0025] S3: Based on the first traction-sensitive feature set and the second traction-sensitive feature set, and combined with historical treatment records and patient subjective comfort scores, a correlation verification is performed to correct the deviation of the traction force control function and obtain the optimal traction force output parameters; S4: According to the optimal traction force output parameters, a dynamic optimization command for thermal traction force is generated, and the actuator is driven to adjust the thermal therapy temperature and mechanical traction force through the dynamic optimization command for thermal traction force.

[0026] Specifically, correlation verification refers to comprehensively analyzing the extracted first and second traction-sensitive feature sets with historical treatment records and patient subjective comfort scores to verify the correlation between these feature sets and treatment effects. Furthermore, historical treatment records include data such as parameter settings, treatment duration, and treatment effect assessments from previous treatments; patient subjective comfort scores are the patient's subjective evaluation of comfort and pain during treatment. This multi-dimensional correlation verification allows for a more comprehensive assessment of the effectiveness of the traction-sensitive feature sets. The traction force control function dynamically adjusts the traction force and thermotherapy temperature based on the input physiological characteristic parameters. The deviation correction of the traction force control function refers to adjusting the parameters in the model based on the results of correlation verification to improve the model's accuracy and adaptability. The obtained optimal traction force output parameters refer to the traction force and thermotherapy temperature parameters that, after correction, achieve the best treatment effect. The dynamic optimization command for thermal traction force refers to the control command generated based on the optimal traction force output parameters, used to drive the actuator to adjust the thermotherapy temperature and mechanical traction force to achieve dynamic optimization of the treatment process. The actuator includes a motor and a heating element.

[0027] Execution steps: During thermo-traction therapy, the correlation between the first and second traction-sensitive feature sets, historical treatment records, and patient subjective comfort scores is verified. Specifically, the currently extracted traction-sensitive feature set is compared and analyzed with the feature sets in historical treatment records to assess their consistency in performance at different treatment stages. Furthermore, historical data shows that the treatment effect is optimal when the heat distribution entropy in the first traction-sensitive feature set under heat stress reaches a certain threshold. At the same time, the patient's subjective comfort score is analyzed to analyze changes in patient comfort under different parameter settings. If the patient's comfort score is high under a certain parameter setting and is consistent with the optimal treatment parameters in historical data, the verification is successful.

[0028] Through correlation verification, the deviation of the traction force control function is effectively corrected. Furthermore, if the correlation verification reveals a low patient comfort score, the temperature parameter in the control function can be adjusted. After multiple iterations, the optimal traction force output parameter is obtained, ensuring the personalization and scientific nature of the treatment parameters. Based on the optimal traction force output parameter, a dynamic optimization command for thermotherapy traction is generated, driving the actuator to adjust the thermotherapy temperature and mechanical traction force, realizing a closed-loop linkage of perception-analysis-control-execution, ensuring dynamic optimization of the treatment process. The dynamic optimization mechanism improves the accuracy and safety of treatment and effectively reduces the risks during the treatment process.

[0029] Furthermore, based on the aforementioned multimodal physiological characteristic parameters, the method of this application determines the first traction-sensitive feature set under heat stress dominance, including: By combining the infrared thermal imaging data and skin impedance data from the multimodal physiological characteristic parameters with the current treatment stage, a coupled analysis is performed to determine the first traction-sensitive feature set under the dominance of heat stress. At the same time, in the heat conduction field corresponding to the target muscle group region, the partial differential relationship between heat energy penetration depth and temperature gradient is analyzed to determine the dynamic thermal response coefficient as the thermal diffusion kinetic index.

[0030] Specifically, coupling analysis refers to the comprehensive analysis of various related data, including infrared thermal imaging data and skin impedance data, to extract more comprehensive physiological characteristic information. By coupling analysis of infrared thermal imaging data and skin impedance data, the physiological changes of the target muscle group under heat stress can be more accurately reflected. The first traction sensitive feature set under heat stress refers to the set of key features extracted from multimodal physiological characteristic parameters under heat stress conditions, which can reflect the impact of heat stress on the effect of traction therapy. The heat conduction field refers to the physical field formed by the distribution and transfer of heat energy in the target muscle group area. The heat penetration depth in the heat conduction field refers to the depth to which heat energy can be effectively transferred into the muscle group. The temperature gradient refers to the rate of temperature change per unit distance, reflecting the intensity and direction of heat energy transfer. The dynamic thermal response coefficient and the thermal diffusion kinetic index are parameters obtained by analyzing the partial differential relationship between the heat penetration depth and the temperature gradient. They are used to quantify the dynamic characteristics of heat energy transfer and diffusion under heat stress, serving as important indicators for evaluating the effect of thermotherapy.

[0031] Execution steps: In thermo-traction therapy, by coupling and analyzing infrared thermal imaging data and skin impedance data, combined with the current treatment stage, the first set of traction-sensitive features under heat stress is determined. Furthermore, the temperature distribution map of the target muscle group area is obtained using an infrared thermal imager, and key features of temperature change, such as temperature gradient and temperature change rate in the heat stress area, are extracted. At the same time, skin impedance data is obtained through a skin impedance sensor, and its correlation with temperature change is analyzed. Through this coupling analysis, the first set of traction-sensitive features under heat stress is extracted, including key features such as heat distribution entropy and intensity of thermal imaging feature areas.

[0032] In the heat conduction field corresponding to the target muscle group region, the partial differential relationship between heat penetration depth and temperature gradient is further analyzed. Specifically, by establishing a heat conduction model, the partial differential equation between heat penetration depth and temperature gradient is calculated to obtain the dynamic thermal response coefficient, and the heat penetration depth in the heat conduction model is analyzed. Temperature gradient The relationship is ,in, The dynamic thermal response coefficient is obtained through fitting experimental data. The value of is used as a thermal diffusion dynamic index to reflect the efficiency and dynamic changes of heat energy transfer in muscle groups, providing an important reference for traction force regulation. By monitoring the dynamic thermal response coefficient, coupling analysis of multimodal data and dynamic analysis of the heat conduction field are carried out to accurately identify key physiological characteristics under heat stress. The thermotherapy temperature can be adjusted in real time to ensure that heat energy is effectively transferred to the target muscle groups and improve the treatment effect.

[0033] Furthermore, based on the aforementioned multimodal physiological characteristic parameters, a second traction-sensitive feature set under muscle fatigue-dominated conditions is determined. The method of this application includes: By combining the surface electromyography (EMG) signal and skin impedance data from the multimodal physiological characteristic parameters with the current traction intensity, a coupling analysis is performed to determine the second traction-sensitive feature set under muscle fatigue dominance. At the same time, the ratio of the integral EMG value of the surface EMG signal during the traction cycle to the phase angle change rate in the skin impedance data is obtained, and the ratio is normalized and used as a muscle-thermal synergistic stability factor.

[0034] Specifically, coupling analysis refers to combining surface electromyography (EMG) signals and skin impedance data with the current traction intensity to comprehensively analyze and extract features related to muscle fatigue. Surface EMG signals reflect the activity state of muscles, including contraction strength and fatigue level; skin impedance data is related to the electrical properties of the skin and can reflect the skin's moisture content and metabolic state. Coupling analysis comprehensively assesses muscle fatigue status. The second traction-sensitive feature set dominated by muscle fatigue refers to a set of features closely related to muscle fatigue extracted from multimodal physiological characteristic parameters, reflecting the degree of muscle fatigue during traction and providing a basis for traction force regulation. The integrated EMG value refers to the amplitude integral of the surface EMG signal within a certain time window, used to quantify the overall muscle activity level; a higher integrated EMG value indicates stronger muscle activity. The phase angle change rate refers to the rate of change of the phase angle over time in the skin impedance data, reflecting the dynamic changes in skin electrical properties. The muscle-thermal synergistic stability factor is a parameter obtained by calculating the ratio of the integrated EMG value to the phase angle change rate and performing normalization; it is used to assess the synergistic stability between muscle activity and thermotherapy effects.

[0035] Execution steps: By coupling analysis of surface electromyography (EMG) signals and skin impedance data, combined with the current traction intensity, a second traction-sensitive feature set dominated by muscle fatigue is determined. Specifically, surface EMG signals are collected and their integrated EMG values ​​are calculated during the traction cycle. Simultaneously, skin impedance data is collected and the phase angle change rate is calculated. Through coupling analysis, the integrated EMG values ​​are correlated with the traction intensity. It is found that when the traction intensity increases, the integrated EMG values ​​rise significantly, indicating enhanced muscle activity and possible fatigue. Combined with the phase angle change rate of the skin impedance data, the degree of muscle fatigue is further confirmed.

[0036] Simultaneously, the ratio of the integrated electromyographic value of the surface electromyography signal to the phase angle change rate in the skin impedance data was calculated and normalized to obtain the muscle-thermal synergistic stability factor, which reflects the synergistic stability between muscle activity and thermotherapy effect. The closer it is to 1, the better the synergy between muscle activity and thermotherapy effect, and the more stable the treatment effect. Through the coupling analysis of multimodal data, the muscle fatigue state is accurately identified, and the synergistic effect of treatment is evaluated by the muscle-thermal synergistic stability factor. Furthermore, by monitoring the muscle-thermal synergistic stability factor, the traction force and thermotherapy parameters can be dynamically adjusted, and precise control can be made within the optimal range of thermotherapy effect and muscle activity synergistic adaptation, ensuring that muscle fatigue is effectively controlled during treatment and improving the accuracy and personalization of thermotherapy traction treatment.

[0037] Furthermore, the method of this application also includes: Based on the first traction-sensitive feature set and the second traction-sensitive feature set, a traction-sensitive feature matrix is ​​constructed. The row vectors of the traction-sensitive feature matrix correspond to physiological feedback samples under multiple treatment time windows, and the column vectors include the thermal distribution entropy and thermal imaging feature region intensity in the first traction-sensitive feature set, and the electromyographic fluctuation coefficient and impedance phase change amplitude in the second traction-sensitive feature set.

[0038] Specifically, the traction-sensitive feature matrix is ​​a mathematical structure used to integrate and organize traction-sensitive features. Each row represents a physiological feedback sample within a specific time window, and each column represents a specific traction-sensitive feature. The construction of the traction-sensitive feature matrix aims to provide a systematic data framework for subsequent analysis and processing. Thermal distribution entropy refers to the uniformity of heat energy distribution, usually calculated from thermal imaging data. The higher the thermal distribution entropy, the more uniform the heat energy distribution and the better the thermotherapy effect. Thermal imaging feature region intensity refers to the temperature intensity of a specific area in a thermal imaging image, reflecting the concentration of heat stress. Electromyographic fluctuation coefficient refers to the fluctuation characteristics of surface electromyographic signals, used to assess the stability of muscle activity. The higher the fluctuation coefficient, the more unstable the muscle activity, possibly approaching a state of fatigue. Impedance phase change amplitude refers to the amplitude of the phase angle change in skin impedance data, reflecting rapid changes in skin electrical properties.

[0039] Execution steps: Based on the first and second traction-sensitive feature sets, a traction-sensitive feature matrix is ​​constructed. Specifically, the physiological feedback samples within each treatment time window are used as row vectors of the matrix. The treatment process is divided into multiple time windows. The physiological feedback samples within each time window include features such as thermal distribution entropy, thermal imaging feature area intensity, electromyographic fluctuation coefficient, and impedance phase change amplitude. Furthermore, by analyzing the data in the matrix, the changing trends of heat stress and muscle fatigue characteristics over time are observed. As time progresses, the thermal distribution entropy gradually decreases, indicating that the heat energy distribution becomes uneven; the thermal imaging feature area intensity gradually increases, indicating enhanced heat stress; and both the electromyographic fluctuation coefficient and the impedance phase change amplitude gradually increase, indicating a deepening of muscle fatigue. In the above steps, the traction-sensitive feature matrix is ​​used to accurately assess physiological changes during the treatment process, providing data support for dynamically adjusting traction force and thermotherapy parameters.

[0040] Furthermore, the method of this application also includes: The thermal diffusion dynamics index is used as the query vector of the attention network, the muscle-thermal synergistic stability factor is used as the key vector of the attention network, and the traction-sensitive feature matrix is ​​used as the value vector of the attention network; the key traction-sensitive features under the weight distribution are dynamically focused through the attention mechanism.

[0041] Specifically, attention networks are deep learning-based mechanisms used to dynamically focus on the most relevant parts of a large amount of data for the current task. Attention networks achieve dynamic focusing on traction-sensitive features by defining query vectors, key vectors, and value vectors. Furthermore, the thermal diffusion dynamics index, as the query vector, represents the dynamic characteristics of heat transfer and diffusion, guiding the search direction of the attention mechanism; the muscle-thermal synergistic stability factor, as the key vector, reflects the synergistic stability between muscle activity and the effect of thermotherapy, and the key vector is matched with the query vector; the traction-sensitive feature matrix, as the value vector, contains all feature data related to traction sensitivity and serves as the input data source for the attention mechanism. Dynamic focusing refers to the attention mechanism dynamically adjusting the weights of each feature in the value vector based on the matching degree between the query and key, thereby extracting the most relevant traction-sensitive features for the current treatment stage. The attention mechanism can dynamically adjust the focus based on real-time physiological states, improving the flexibility and accuracy of feature extraction.

[0042] Execution steps: During thermo-traction therapy, an attention mechanism is used to dynamically focus on key traction-sensitive features. Specifically, the thermal diffusion dynamics index is used as the query vector, the muscle-thermal synergistic stability factor is used as the key vector, and the traction-sensitive feature matrix is ​​used as the value vector, all input into the attention network. The attention mechanism calculates the similarity between the query vector and the key vector and assigns weights to each feature in the value vector. The similarity between the query vector and the key vector is either the dot product similarity or the kernel function-based similarity. Furthermore, the similarity between the query vector and the key vector is obtained to determine the correlation between the current thermal diffusion dynamics index and the muscle-thermal synergistic stability factor. The attention mechanism assigns weights to each feature in the traction-sensitive feature matrix, including thermal distribution entropy weights, thermal imaging feature region intensity weights, electromyographic fluctuation coefficient weights, and impedance phase mutation amplitude weights.

[0043] In the above steps, through this weighting allocation, the attention mechanism dynamically focuses on the key traction-sensitive features of the current treatment stage. Furthermore, if the weights of thermal distribution entropy and thermal imaging feature region intensity are high, it indicates that thermal stress features are more important at this time; if the weights of electromyographic fluctuation coefficient and impedance phase change amplitude are high, it indicates that muscle fatigue features are more critical. The attention mechanism dynamically adjusts the degree of attention to different traction-sensitive features to ensure that the features most relevant to the treatment effect can be accurately extracted at different treatment stages. For example, in the early stage of treatment, muscle fatigue features may be more critical; in the later stage of treatment, thermal stress features may be more important. By dynamically focusing and adjusting the traction force and thermotherapy parameters in real time, it can adapt to the individual differences of patients and the treatment process.

[0044] Furthermore, the method of this application includes: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation; it obtains the kernel density overlap area between the query vector and the key vector as a similarity index: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area between the kernel density curve corresponding to the query vector and the kernel density curve corresponding to the key vector is calculated.

[0045] Specifically, kernel density estimation is used to estimate the probability distributions of query and key vectors, estimating the probability density function of random variables, thus replacing traditional inner product operations to measure their similarity; the Gaussian kernel function is used for kernel density estimation, smoothing data points through a Gaussian distribution, making the estimated probability density function smoother and more continuous; the Silverman rule is used to adaptively determine the bandwidth in kernel density estimation, as bandwidth determines the smoothness of the kernel function. The Silverman rule automatically adjusts the bandwidth according to the data distribution to obtain a more accurate density estimate; the kernel density overlap area refers to the area of ​​the overlapping part between two kernel density curves, used to measure the similarity between two vectors. The larger the overlap area, the more similar the two vectors are.

[0046] Execution steps: The attention mechanism replaces the traditional inner product operation with a similarity metric based on kernel density estimation to more accurately measure the similarity between the query vector and the key vector. Specifically, kernel density estimation is performed on both the query vector and the key vector. If the query vector is [ , , ..., ], the key vector is [ , , ..., The kernel density is estimated using a Gaussian kernel function, and the bandwidth is adaptively determined using the Silverman rule. The kernel density curves of the query vector and key vector are calculated using the bandwidth obtained by the Silverman rule, and the kernel density overlap area between the two kernel density curves is determined. The similarity index will be used in the attention mechanism to assign weights to the features in the traction-sensitive feature matrix.

[0047] Preferably, compared with inner product operation, kernel density estimation can better capture the complex relationships between vectors, especially when the data distribution is uneven or there is noise. Through the similarity measurement of kernel density estimation, the characteristic changes of heat stress and muscle fatigue can be identified more accurately, improving the accuracy of feature recognition and the stability of treatment effect, thereby dynamically adjusting traction force and thermotherapy parameters.

[0048] Furthermore, the method of this application includes: The similarity index includes a first similarity threshold and a second similarity threshold. If the first similarity threshold is met, the attention mechanism will distribute the weights towards the low-frequency temporal variation components of thermal distribution entropy and electromyographic fluctuation coefficient. If the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity and impedance phase change amplitude of the thermal imaging feature region.

[0049] Specifically, the similarity threshold is a set threshold used to determine whether the similarity represented by the kernel density overlap area has reached a certain state. Two similarity thresholds are set: a first similarity threshold and a second similarity threshold. When the similarity index reaches or exceeds the first similarity threshold, it indicates that the similarity between the thermal diffusion dynamics index and the muscle-thermal synergistic stability factor is high, and the low-frequency temporal variation characteristics of heat stress and muscle fatigue during traction therapy are more critical. When the similarity index reaches or exceeds the second similarity threshold, it indicates that the similarity between the thermal diffusion dynamics index and the muscle-thermal synergistic stability factor is even higher, and the high-frequency transient mutation characteristics of the intensity of thermal imaging feature regions and the amplitude of impedance phase mutations during traction therapy are more critical. The low-frequency temporal variation component refers to the low-frequency signal part in the traction sensitive feature matrix that is related to the thermal distribution entropy and electromyographic fluctuation coefficient, which usually reflects the long-term trend and stability of physiological characteristics. The high-frequency transient mutation component refers to the high-frequency signal part in the traction sensitive feature matrix that is related to the intensity of thermal imaging feature regions and the amplitude of impedance phase mutations, which usually reflects the short-term fluctuations and mutations of physiological characteristics.

[0050] Execution steps: In thermo-traction therapy, the weight distribution of the attention mechanism is dynamically adjusted by setting a similarity threshold to adapt to different treatment stages and physiological states. Furthermore, the similarity index calculated based on the kernel density overlap area is compared with a preset first and second similarity threshold. If the similarity index reaches or exceeds the first similarity threshold but is below the second similarity threshold, the attention mechanism tilts the weight distribution towards the low-frequency temporal variation components of thermal distribution entropy and electromyographic fluctuation coefficient. Specifically, in the traction-sensitive feature matrix, the low-frequency temporal variation components are extracted as slowly varying feature sub-vectors. When the similarity index reaches or exceeds the second similarity threshold, the attention mechanism tilts the weight distribution towards the high-frequency transient mutation components of the intensity of the thermal imaging feature region and the amplitude of impedance phase mutations.

[0051] By dynamically adjusting the weight distribution, different traction-sensitive characteristics can be flexibly focused on based on the current physiological state and similarity index. Specifically, in the early stage of treatment, when the similarity index is low, more attention is paid to low-frequency temporal variation components to assess the long-term trend of muscle fatigue; in the later stage of treatment, when the similarity index is high, more attention is paid to high-frequency transient mutation components to capture short-term fluctuations in heat stress in a timely manner. This dynamic adjustment mechanism can improve the accuracy and adaptability of treatment and ensure real-time optimization of traction force and thermotherapy parameters.

[0052] Furthermore, the method of this application includes: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the traction-sensitive feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentiated weighted fusion is performed.

[0053] Specifically, Empirical Mode Decomposition (EMD) is an adaptive signal processing method used to decompose complex signals into several Intrinsic Mode Functions (IMFs). Each IMF represents a characteristic component of the signal at different frequencies and time scales. Low-frequency time-series variation components correspond to the lower-order IMFs after EMD, reflecting the long-term trend and low-frequency characteristics of the signal. High-frequency transient change components correspond to the higher-order IMFs after EMD, reflecting the short-term fluctuations and high-frequency characteristics of the signal. Hilbert Transform is a signal processing technique used to convert real signals into analytic signals, extracting the instantaneous amplitude and instantaneous frequency. The instantaneous amplitude represents the amplitude of the signal at each time point, and the instantaneous frequency represents the frequency change of the signal at each time point. Differential weighted fusion refers to assigning different weights to the signal based on its different characteristics, including low-frequency and high-frequency components, and fusing these weighted features to obtain a more comprehensive feature representation.

[0054] Execution steps: The traction-sensitive feature matrix is ​​decomposed into columns using Empirical Mode Decomposition (EMD) to extract low-frequency temporal variation components and high-frequency transient change components. Specifically, EMD is performed on each column of the traction-sensitive feature matrix, which includes thermal distribution entropy, electromyographic fluctuation coefficient, thermal imaging feature region intensity, and impedance phase change amplitude. Multiple intrinsic mode functions (IMFs) are obtained through EMD. Furthermore, lower-order IMFs primarily reflect low-frequency temporal variations, while higher-order IMFs primarily reflect high-frequency transient changes. Hilbert transform is applied to each IMF to extract instantaneous amplitude and instantaneous frequency features. Specifically, Hilbert transform is performed on lower-order IMFs to obtain instantaneous amplitude and instantaneous frequency; Hilbert transform is also performed on higher-order IMFs to obtain instantaneous amplitude and instantaneous frequency.

[0055] Differential weighted fusion is performed based on the importance of low-frequency and high-frequency features, fusing the weighted instantaneous amplitude and frequency features. Through empirical mode decomposition and Hilbert transform, low-frequency and high-frequency features are extracted from the traction-sensitive feature matrix, and then weighted fusion is performed according to the importance of these features. Low-frequency features reflect long-term trends in physiological states, such as the accumulation of muscle fatigue; high-frequency features reflect short-term fluctuations and abrupt changes, such as transient changes in heat stress. Through differential weighted fusion, the accumulation of muscle fatigue and transient changes in heat stress are accurately captured, comprehensively assessing physiological changes during treatment and improving the accuracy of dynamic adjustments to traction force and thermotherapy parameters.

[0056] Furthermore, the method of this application includes: The key traction-sensitive features, after differential weighting, are input into the gated circulation unit; the gated circulation unit suppresses physiological noise interference, and a traction force regulation function is constructed based on the output features of the gated circulation unit to dynamically output the thermotherapy temperature setpoint and the mechanical traction force amplitude.

[0057] Specifically, the gated recurrent unit (ROU) is used to process sequential data and is suitable for processing time-dependent physiological signals. By introducing update and reset gates to control the flow of information, it can effectively solve the gradient vanishing problem of recurrent neural networks and performs well when processing long sequence data. Physiological noise interference refers to the interference signals introduced during the physiological signal acquisition process due to environmental noise, equipment errors, patient movement, and other factors. Physiological noise interference can mask the true physiological characteristics and affect the accuracy of signal processing and analysis. The traction control function is a dynamic model built based on the output characteristics of the gated recurrent unit. It is used to dynamically adjust the thermotherapy temperature and mechanical traction force according to real-time physiological characteristics. By learning from historical data and real-time feedback, it can adaptively output the optimal treatment parameters.

[0058] Execution steps: The differentially weighted key traction-sensitive features are input into a gated recurrent unit (GRU). The GRU suppresses physiological noise interference and constructs a traction force control function based on its output features, dynamically outputting the thermotherapy temperature setpoint and mechanical traction force amplitude. Specifically, the differentially weighted key traction-sensitive features, including low-frequency temporal variation components and high-frequency transient change components, are input as an input sequence into the GRU network. Based on the time step, the GRU updates the gated recurrent unit... and reset door Dynamically adjust the flow of information to suppress physiological noise interference, and further update the gate. Reset the door ,in, It is the hidden state of the previous time step. These are the input features at the current time step. and It is a weight matrix. and It is a bias term. It is the Sigmoid activation function.

[0059] By controlling the updating and resetting of the gates, the gated loop unit can effectively suppress noise interference and extract purer physiological features. In the feature sequence processed by the gated loop unit, noise interference is reduced and the signal-to-noise ratio of the features is improved. A traction control function is constructed based on the output features of the gated loop unit. Based on the output feature sequence of the gated loop unit, the traction control function dynamically outputs the thermotherapy temperature setpoint and mechanical traction amplitude by learning the relationship between these features and historical treatment data. The output features of the gated loop unit contain relevant information about thermotherapy temperature and traction force. The mapping function constructed based on the output features of the gated loop unit is learned through training data.

[0060] By leveraging the noise suppression and dynamic modeling capabilities of the gated loop unit, the accuracy and adaptability of traction force regulation are improved. The gated loop unit can effectively handle noise interference in time series data, extract purer physiological features, and provide more accurate input to the traction force regulation function. The traction force regulation function constructed based on the output features of the gated loop unit can dynamically adjust the thermotherapy temperature and mechanical traction force, ensuring the personalization and intelligence of the treatment process. The features processed by the gated loop unit are input into the traction force regulation function to adjust the thermotherapy temperature and traction force in real time.

[0061] In summary, the beneficial effects of the embodiments of this application are: This application employs a multimodal physiological characteristic parameter acquisition method and system based on multi-sensor fusion. This method acquires surface electromyography (EMG) signals, infrared thermography data, and skin impedance data for the target muscle group region. Based on these parameters, a first traction-sensitive feature set under heat stress and a second traction-sensitive feature set under muscle fatigue are determined. The correlation between these two sets, along with historical treatment records and patient subjective comfort scores, is validated to correct deviations in the traction force control function, resulting in optimal traction force output parameters. Based on these optimal parameters, a dynamic optimization command for thermo-traction force is generated, which drives the actuator to adjust the thermotherapy temperature and mechanical traction force. This technology integrates multimodal parameters such as surface electromyography and infrared thermography to construct a first traction sensitive feature set dominated by heat stress and a second traction sensitive feature set dominated by muscle fatigue. This enables precise identification of heat stress and muscle fatigue, ensuring the stability and reliability of traction force regulation. By combining historical treatment records and subjective comfort scores to correct the regulation function, the accuracy of thermo-traction force regulation is improved, dynamically adapting treatment parameters to individual patient needs and effectively reducing the technical risks of thermo-traction therapy.

[0062] Example 2, based on the same inventive concept as the multi-sensor fusion-based dynamic optimization method for thermal traction force in the preceding examples, such as... Figure 2As shown in the figure, this application provides a dynamic optimization system for thermal traction force based on multi-sensor fusion, wherein the system includes: Multimodal physiological feature parameter acquisition module M100: Based on the target muscle group region, it acquires multimodal physiological feature parameters including surface electromyography signals, infrared thermal imaging data, and skin impedance data.

[0063] First and Second Traction Sensitive Feature Set Determination Module M200: Based on the multimodal physiological feature parameters, determine the first traction sensitive feature set under heat stress and the second traction sensitive feature set under muscle fatigue.

[0064] The correlation verification module M300: Based on the first traction sensitive feature set and the second traction sensitive feature set, it performs correlation verification by combining historical treatment records and patient subjective comfort scores, corrects the deviation of the traction force control function, and obtains the optimal traction force output parameters.

[0065] Actuator drive module M400: Based on the optimal traction force output parameters, it generates a dynamic optimization command for thermal traction force, and drives the actuator to adjust the thermal therapy temperature and mechanical traction force through the dynamic optimization command for thermal traction force.

[0066] Furthermore, the first and second traction-sensitive feature set determination modules M200 are used to execute the following method: By combining the infrared thermal imaging data and skin impedance data from the multimodal physiological characteristic parameters with the current treatment stage, a coupled analysis is performed to determine the first traction-sensitive feature set under the dominance of heat stress. At the same time, in the heat conduction field corresponding to the target muscle group region, the partial differential relationship between heat energy penetration depth and temperature gradient is analyzed to determine the dynamic thermal response coefficient as the thermal diffusion kinetic index.

[0067] Furthermore, the first and second traction-sensitive feature set determination modules M200 are used to execute the following method: By combining the surface electromyography (EMG) signal and skin impedance data from the multimodal physiological characteristic parameters with the current traction intensity, a coupling analysis is performed to determine the second traction-sensitive feature set under muscle fatigue dominance. At the same time, the ratio of the integral EMG value of the surface EMG signal during the traction cycle to the phase angle change rate in the skin impedance data is obtained, and the ratio is normalized and used as a muscle-thermal synergistic stability factor.

[0068] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: Based on the first traction-sensitive feature set and the second traction-sensitive feature set, a traction-sensitive feature matrix is ​​constructed. The row vectors of the traction-sensitive feature matrix correspond to physiological feedback samples under multiple treatment time windows, and the column vectors include the thermal distribution entropy and thermal imaging feature region intensity in the first traction-sensitive feature set, and the electromyographic fluctuation coefficient and impedance phase change amplitude in the second traction-sensitive feature set.

[0069] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: The thermal diffusion dynamics index is used as the query vector of the attention network, the muscle-thermal synergistic stability factor is used as the key vector of the attention network, and the traction-sensitive feature matrix is ​​used as the value vector of the attention network; the key traction-sensitive features under the weight distribution are dynamically focused through the attention mechanism.

[0070] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation; it obtains the kernel density overlap area between the query vector and the key vector as a similarity index: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area between the kernel density curve corresponding to the query vector and the kernel density curve corresponding to the key vector is calculated.

[0071] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: The similarity index includes a first similarity threshold and a second similarity threshold. If the first similarity threshold is met, the attention mechanism will distribute the weights towards the low-frequency temporal variation components of thermal distribution entropy and electromyographic fluctuation coefficient. If the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity and impedance phase change amplitude of the thermal imaging feature region.

[0072] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the traction-sensitive feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentiated weighted fusion is performed.

[0073] Furthermore, the first and second traction-sensitive feature set determination modules M200 are also used to perform the following methods: The key traction-sensitive features, after differential weighting, are input into the gated circulation unit; the gated circulation unit suppresses physiological noise interference, and a traction force regulation function is constructed based on the output features of the gated circulation unit to dynamically output the thermotherapy temperature setpoint and the mechanical traction force amplitude.

[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The method and specific examples of dynamic optimization of temperature and thermal traction force based on multi-sensor fusion in Example 1 are also applicable to the dynamic optimization system of temperature and thermal traction force based on multi-sensor fusion in this embodiment. Through the foregoing detailed description of the dynamic optimization method of temperature and thermal traction force based on multi-sensor fusion, those skilled in the art can clearly understand the dynamic optimization system of temperature and thermal traction force based on multi-sensor fusion in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0075] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0076] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A dynamic optimization method for thermal traction force based on multi-sensor fusion, characterized in that, The method includes: Based on the target muscle group region, acquire multimodal physiological characteristic parameters including surface electromyography signals, infrared thermal imaging data, and skin impedance data; Based on the aforementioned multimodal physiological characteristic parameters, a first traction-sensitive feature set under heat stress and a second traction-sensitive feature set under muscle fatigue are determined. Based on the first traction-sensitive feature set and the second traction-sensitive feature set, the correlation between historical treatment records and patient subjective comfort scores is verified, the deviation of the traction force control function is corrected, and the optimal traction force output parameters are obtained. Based on the optimal traction output parameters, a dynamic optimization command for thermal traction is generated, which drives the actuator to adjust the thermotherapy temperature and mechanical traction force.

2. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 1, characterized in that, Based on the aforementioned multimodal physiological characteristic parameters, a first traction-sensitive feature set under heat stress is determined, the method comprising: By combining the infrared thermal imaging data and skin impedance data from the multimodal physiological characteristic parameters with the current treatment stage, a coupling analysis is performed to determine the first traction-sensitive feature set under the dominance of heat stress. Meanwhile, in the heat conduction field corresponding to the target muscle group region, the partial differential relationship between heat penetration depth and temperature gradient is analyzed to determine the dynamic thermal response coefficient as the thermal diffusion kinetic index.

3. The dynamic optimization method for thermal traction force based on multi-sensor fusion as described in claim 2, characterized in that, Based on the aforementioned multimodal physiological characteristic parameters, a second traction-sensitive feature set under muscle fatigue-dominated conditions is determined, the method comprising: By combining the surface electromyography signal and skin impedance data from the multimodal physiological characteristic parameters with the current traction intensity, a coupling analysis is performed to determine the second traction-sensitive feature set under muscle fatigue dominance. Simultaneously, the ratio of the integral electromyographic value of the surface electromyographic signal during the traction cycle to the phase angle change rate in the skin impedance data is obtained, and the ratio is normalized and used as a muscle-thermal synergistic stability factor.

4. The dynamic optimization method for thermal traction force based on multi-sensor fusion as described in claim 3, characterized in that, The method further includes: Based on the first traction-sensitive feature set and the second traction-sensitive feature set, a traction-sensitive feature matrix is ​​constructed; The row vectors of the traction-sensitive feature matrix correspond to physiological feedback samples under multiple treatment time windows, and the column vectors include the thermal distribution entropy and thermal imaging feature region intensity in the first traction-sensitive feature set, and the electromyographic fluctuation coefficient and impedance phase change amplitude in the second traction-sensitive feature set.

5. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 4, characterized in that, The method further includes: The thermal diffusion dynamics index is used as the query vector of the attention network, the muscle-thermal synergistic stability factor is used as the key vector of the attention network, and the traction-sensitive feature matrix is ​​used as the value vector of the attention network. Dynamically focus on key, sensitive features under weighted distribution through an attention mechanism.

6. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 5, characterized in that, The method includes: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation; The similarity index is obtained by using the kernel density overlap area between the query vector and the key vector: kernel density is estimated for both the query vector and the key vector using a Gaussian kernel function, and the bandwidth is adaptively determined using the Silverman rule; the kernel density overlap area between the kernel density curves corresponding to the query vector and the key vector is calculated.

7. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 6, characterized in that, The similarity index includes a first similarity threshold and a second similarity threshold; If the first similarity threshold is met, the attention mechanism will shift the weight distribution towards the low-frequency temporal variation components of thermal distribution entropy and electromyographic fluctuation coefficient. If the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity and impedance phase change amplitude of the thermal imaging feature region.

8. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 7, characterized in that, The method includes: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the traction-sensitive feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentiated weighted fusion is performed.

9. The method for dynamic optimization of thermal traction force based on multi-sensor fusion as described in claim 8, characterized in that, The method includes: The key traction-sensitive features, after differential weighting, are input into the gated loop unit; Physiological noise interference is suppressed by the gated circulation unit, and a traction force control function is constructed based on the output characteristics of the gated circulation unit to dynamically output the thermotherapy temperature setpoint and the mechanical traction force amplitude.

10. A dynamic optimization system for thermal traction force based on multi-sensor fusion, characterized in that, The system is used to implement the multi-sensor fusion-based dynamic optimization method for thermal traction force according to any one of claims 1-9, wherein the system comprises: Multimodal physiological feature parameter acquisition module: Based on the target muscle group region, acquire multimodal physiological feature parameters including surface electromyography signals, infrared thermal imaging data, and skin impedance data; First and second traction-sensitive feature set determination module: Based on the multimodal physiological feature parameters, determine the first traction-sensitive feature set under heat stress and the second traction-sensitive feature set under muscle fatigue. Correlation verification module: Based on the first traction sensitive feature set and the second traction sensitive feature set, and combined with historical treatment records and patient subjective comfort scores, correlation verification is performed to correct the deviation of the traction force control function and obtain the optimal traction force output parameters; Actuator drive module: Based on the optimal traction force output parameters, it generates a dynamic optimization command for thermal traction force, and drives the actuator to adjust the thermotherapy temperature and mechanical traction force through the dynamic optimization command for thermal traction force.

Images