Healthcare recommendation method and system using rehabilitation device

Abstract

A health care recommendation method and system utilizing rehabilitation medical devices, wherein the health care recommendation method is executed by a processing device and includes: acquiring multiple sets of raw sensing data from multiple sensing devices; labeling one of the multiple sets of raw sensing data according to multiple personalized features to generate labeled data; inputting one of the multiple sets of raw sensing data into a pre-trained prediction model to generate predicted data; fusing the labeled data and the predicted data to generate fused data; extracting multi-dimensional feature data from the multiple sets of raw sensing data; inputting the fused data and the multi-dimensional feature data into a pre-trained inference model to generate exercise performance indicators; and generating a health care plan based on the exercise performance indicators.

Description

Technical Field

[0001] This invention relates to a health care recommendation method and system. Prior Technology

[0002] Population aging has become a global trend and is progressing rapidly. Therefore, how to effectively care for the elderly will be a major challenge for society in the future. Faced with the arrival of an aging society, care institutions need to integrate and utilize information and communication technology products and services to address manpower shortages while improving the quality of care for the elderly.

[0003] In current rehabilitation care, the assessment of rehabilitation equipment use mainly relies on the number of times and the duration of use, or uses general standard norms for quantitative evaluation. However, these methods often overlook individual physiological conditions, the process of use, and personal health history, failing to provide personalized effectiveness assessments based on multi-dimensional data, thus limiting the accuracy and targeted nature of rehabilitation results. Summary of the Invention

[0004] In view of the above, the present invention provides a health care recommendation method and system for solving the above problems.

[0005] A health care recommendation method according to an embodiment of the present invention, executed by a processing device, includes: acquiring multiple sets of raw sensing data from multiple sensing devices; labeling one of the multiple sets of raw sensing data according to multiple personalized features to generate labeled data; inputting one of the multiple sets of raw sensing data into a pre-trained prediction model to generate predicted data; fusing the labeled data and the predicted data to generate fused data; extracting multi-dimensional feature data from the multiple sets of raw sensing data; inputting the fused data and the multi-dimensional feature data into a pre-trained inference model to generate an exercise performance index; and generating a health care plan based on the exercise performance index.

[0006] A health care recommendation system according to an embodiment of the present invention includes multiple sensing devices and a processing device. The multiple sensing devices are used to acquire multiple sets of raw sensing data. The processing device is connected to the multiple sensing devices and is used to label one of the multiple sets of raw sensing data according to multiple personalized features to generate labeled data; input one of the multiple sets of raw sensing data into a pre-trained prediction model to generate predicted data; fuse the labeled data and the predicted data to generate fused data; extract multi-dimensional feature data from the multiple sets of raw sensing data; input the fused data and the multi-dimensional feature data into a pre-trained inference model to generate exercise performance indicators; and generate a health care plan based on the exercise performance indicators.

[0007] In summary, the health care recommendation method and system according to one or more embodiments of the present invention can perform personalized labeling, multi-dimensional feature extraction, and multivariate model data processing on the sensed data, use the above data to evaluate personalized exercise performance indicators, and provide users with health care solutions accordingly, thereby solving the problem of manpower shortage in care institutions and improving the quality of care.

[0008] The foregoing description of the disclosure and the following description of the embodiments are intended to demonstrate and explain the spirit and principles of the present invention, and to provide a further explanation of the scope of the patent application of the present invention. Simple Explanation of the Diagram

[0009] Figure 1 is a block diagram illustrating a health care recommendation system according to an embodiment of the present invention. Figure 2 is a flowchart illustrating a health care recommendation method according to an embodiment of the present invention. Figure 3 is a flowchart illustrating motion fluency analysis in a health care recommendation method according to an embodiment of the present invention. Figure 4 is a flowchart illustrating the establishment of personalized features in a health care recommendation method according to an embodiment of the present invention. Figure 5 is a flowchart illustrating the process of obtaining a pre-trained inference model in a health care recommendation method according to an embodiment of the present invention. Figure 6 is a flowchart illustrating the process of obtaining a pre-trained inference model in a health care recommendation method according to another embodiment of the present invention. Implementation

[0010] The following detailed description of the features and advantages of the present invention is sufficient to enable anyone skilled in the art to understand the technical content of the present invention and implement it accordingly. Based on the disclosure, patent claims, and drawings in this specification, anyone skilled in the art can easily understand the relevant objectives and advantages of the present invention. The following embodiments further illustrate the points of the present invention, but are not intended to limit the scope of the present invention in any way.

[0011] Please refer to Figure 1, which is a block diagram illustrating a health care recommendation system according to an embodiment of the present invention. As shown in Figure 1, the health care recommendation system 1 includes multiple sensing devices 11, a processing device 12, and an input / output device 13. The processing device 12 is connected to the sensing devices 11 and the input / output device 13 via wired or wireless means. The input / output device 13 is used to provide health care solutions to the user. It should be noted that Figure 1 schematically shows two sensing devices 11, but the number of sensing devices may be greater than two; the present invention does not limit the number of sensing devices.

[0012] The sensing devices 11 may be motion sensing devices, physiological sensing devices, other sensing devices related to the user's movement performance or physiological state, or a combination of the above sensing devices. The sensing devices 11 are used to sense the user's movements and / or physiological state to generate raw sensing data. Motion sensing devices may include one or more of distance sensors, speed sensors, and angle sensors, and the corresponding raw sensing data may include one or more of movement time, movement trajectory, and motion-sensing interaction data. Furthermore, the movement trajectory may be obtained based on one or more of the movement distance, movement speed, and rotation angle. The motion sensing devices may be installed on the user or on rehabilitation medical equipment (e.g., rehabilitation exercise equipment) operated by the user to obtain the user's movement trajectory and corresponding time points. Physiological sensing devices may include one or more of the following: blood pressure sensor, blood oxygen sensor, blood glucose sensor, body temperature sensor, and body fat / weight sensor, including but not limited to the physiological information measurement sensors listed above. The corresponding raw sensing data may include one or more of the following: blood pressure, blood oxygen concentration, blood glucose concentration, body temperature, body fat, and weight, including but not limited to the raw sensing data items listed above. Physiological sensing devices may be installed on the user's body.

[0013] The processing device 12 is used to determine the user's exercise effectiveness based on the sensing results of the sensing device 11, and recommend corresponding health care plans to the user accordingly. The processing device 12 may include one or more processors, such as a central processing unit, graphics processing unit, microcontroller, programmable logic controller, or other processors with signal processing capabilities.

[0014] Please refer to Figures 1 and 2 together, where Figure 2 is a flowchart illustrating a health care recommendation method according to an embodiment of the present invention. As shown in Figure 2, the health care recommendation method includes steps S111 to S121.

[0015] In step S111, the processing device 12 obtains raw sensing data from the sensing device 11, which indicates the user's physiological state and / or movement state. The raw sensing data can be used to indicate the user's movement state and / or physiological state during movement.

[0016] In step S112, the processing device 12 extracts multi-dimensional feature data from the raw sensing data. Multi-dimensional feature data can indicate data from multiple raw sensing data sets within the same time interval. In an embodiment where one sensing device generates one type of raw sensing data, the number of dimensions can correspond to the number of sensing devices. The processing device 12 can extract portions corresponding to the same time interval from different types of raw sensing data generated by different types of sensing devices 11 as multi-dimensional feature data. For example, the raw sensing data can include, but is not limited to, motion sensing data, physiological sensing data, fitness data, action recognition data, and sound recognition data. In step S112, the processing device 12 simultaneously considers the characteristics of diverse raw sensing data to establish personalized multi-dimensional feature data.

[0017] In step S113, the processing device 12 uses personalized feature tags to label one of the original sensed data to generate labeled data. In one embodiment, the processing device 12 may use a threshold-based auto-labeling (TBAL) algorithm to perform feature labeling based on a given threshold and waveform features corresponding to various motion features. The threshold can be used to indicate that the original sensed data can be determined as valid data, and the waveform features can be the waveform features of the original sensed data. Personalized feature tags may include the part of the body moving and the corresponding motion features, such as abduction of the upper limb, pulling backward of the upper limb, squatting of the lower limb, raising the knee of the lower limb, etc. The present invention does not limit the part of the body moving or the motion features.

[0018] In step S115, the processing device 12 inputs one of the original sensing data into the pre-trained prediction model to generate prediction data. The pre-trained prediction model is a model pre-trained to predict motion-related parameters, and may include, but is not limited to, one or more of the following: a calorie prediction model, a motion intensity prediction model, and a muscle endurance prediction model. The prediction data is the prediction result of the pre-trained prediction model, and may include, but is not limited to, one or more of the following: calories burned, intensity used, and muscle endurance level. It should be noted that the original sensing data marked in step S113 and the original sensing data input into the pre-trained prediction model in step S115 may be the same data or different data.

[0019] In step S117, the processing device 12 performs data fusion on the labeled data and the predicted data to generate fused data, wherein the fused data may be metadata. Data fusion can be achieved through observation-level fusion, feature-level fusion, and / or decision-level fusion, and the present invention is not limited thereto.

[0020] In step S119, the processing device 12 inputs the fused data and multi-dimensional feature data into the pre-trained inference model to generate exercise performance indicators. The exercise performance indicators can be presented in the form of scores, percentages, etc., and can serve as the analysis results of the user's physiological and exercise status.

[0021] In step S121, the processing device 12 generates and outputs a health care plan based on exercise performance indicators. For example, when the health care recommendation method and system according to one or more embodiments of the present invention are applied to a rehabilitation context, the health care plan can be generated through health promotion effectiveness analysis. Health promotion effectiveness analysis can combine various data obtained in previous steps (e.g., pre- and post-fitness assessment data, somatosensory interaction data, movement fluency analysis data, etc.) to analyze the user's physiological state, providing rehabilitation indicators and suggestions for health promotion. The health care plan may include a dynamic prescription and / or a personalized health care plan. The dynamic prescription is a rehabilitation prescription provided to the user that is dynamically adjusted by the processing device 12 based on the results of the health promotion effectiveness analysis. The personalized health care plan may include the integration of the results of the health promotion effectiveness analysis and the dynamic prescription to provide a complete training plan and suggestions suitable for the user.

[0022] The health care recommendation method and system according to one or more embodiments of the present invention can perform personalized labeling, multi-dimensional feature extraction and multivariate model data processing on the sensed data, use the above data to evaluate personalized exercise performance indicators, and provide users with health care plans accordingly, thereby solving the problem of manpower shortage in care institutions and improving the quality of care.

[0023] In one embodiment, the processing device 12 may be further configured to acquire user feedback data through the input / output device 13, and adjust the threshold setting used for marking in step S113 based on the user feedback data. User feedback data may include, but is not limited to, the user's rate of perceived exertion (RPE), the user's physiological signals, and fitness test results. The threshold setting may indicate the action goal to be achieved, including, but not limited to, waveform amplitude thresholds, frequency thresholds, slope thresholds, and time interval thresholds. For example, when user feedback data shows that the current exercise intensity exceeds the user's tolerance level, the processing device 12 may lower the threshold setting; conversely, when user feedback data shows that the current exercise intensity is lower than the user's tolerance level, the processing device 12 may raise the threshold setting. Accordingly, even if the user's tolerance level is low, lowering the threshold allows the user to complete exercise that matches their physical condition; conversely, when the user's tolerance level is high, raising the threshold allows the user to challenge themselves with exercise that matches their physical condition.

[0024] Please refer to Figures 1 and 3 together, where Figure 3 is a flowchart illustrating motion fluency analysis in a health care recommendation method according to an embodiment of the present invention. In the embodiment of Figure 3, the raw sensing data may include motion time and motion trajectory. As shown in Figure 3, motion fluency analysis includes steps S211 and S213.

[0025] In step S211, the processing device 12 can calculate the acceleration change using the motion time and motion trajectory. The processing device 12 can calculate the change in acceleration over time (the derivative of acceleration) corresponding to the motion trajectory.

[0026] In step S213, the processing device 12 can perform motion smoothness analysis using acceleration changes to generate smoothness data. The smoothness data can be represented as a score; for example, the smaller the acceleration change, the higher the score of the smoothness data, and vice versa. Specifically, when a movement is fast, consistent, and smooth, it indicates that the user has good motor skills. Generally, younger people have higher smoothness than older people, and therefore lower acceleration changes; conversely, older people have lower motion smoothness and higher acceleration changes.

[0027] Accordingly, the movement time of rehabilitation medical equipment and the amount of movement of the user's movement trajectory can be collected through the sensing device during the user's rehabilitation activities. The smoothness of the user's operation of rehabilitation medical equipment can be obtained through the variable acceleration calculation mechanism.

[0028] Processing device 12 can utilize fluency data as one of the data sources for pre-training inference models to generate motion performance indicators. Steps S211 and S213 can be performed, but are not limited to, steps S111 and S117 in Figure 2. Specifically, step S117 in Figure 2 may include fusing labeled data, predicted data, and fluency data to generate fused data.

[0029] Furthermore, step S213 in Figure 3 may include obtaining the motion type corresponding to the motion trajectory, and selecting one of multiple motion smoothness analysis algorithms based on the motion type for motion smoothness analysis. Specifically, the processing device 12 may determine whether the motion type is linear motion or rotational motion based on the motion trajectory, and select the corresponding motion smoothness analysis algorithm based on the motion type for motion smoothness analysis. The motion smoothness analysis algorithm may include the log dimensionless jerk (LDLJ) algorithm and the spectral arc length (SPARC) algorithm. When the motion type is linear motion, the processing device 12 may use either the log dimensionless jerk algorithm or the spectral arc length (SPARC) algorithm, with the SPARC algorithm being preferred; when the motion type is rotational motion, the processing device 12 may use the log dimensionless jerk algorithm.

[0030] In one embodiment, the processing device 12 using the spectral arc length motion smoothness algorithm can obtain smoothness by calculating the arc length of the Fourier spectrum of a given velocity curve in the frequency range of 0 Hz to 20 Hz. Furthermore, the processing device 12 can change the cutoff frequency by adjusting the threshold. For waveforms with more noise, using a higher threshold can improve measurement accuracy, while using a lower threshold can obtain more motion details. Therefore, through pre-testing, automatic / semi-automatic threshold adjustment can find a balance between accuracy and stability, thereby obtaining appropriate smoothness data.

[0031] Please refer to Figures 1 and 4 together, where Figure 4 is a flowchart illustrating the establishment of personalized features in a health care recommendation method according to an embodiment of the present invention. As shown in Figure 4, establishing a normal model includes steps S311 to S319. The steps shown in Figure 4 can be performed before the original sensing data is labeled according to personalized features from the sensing device 11 (step S115 in Figure 2), but the present invention is not limited thereto. The steps in Figure 4 can be performed by the processing device 12 in Figure 1, or by another processing device, and after the other processing device obtains the normal model, it outputs the normal model to the processing device 12.

[0032] In step S311, the first historical sensing data obtained by the processing device 12 can be data from the sensing device 11, but the generation time of the first historical sensing data can be earlier than the original sensing data. The first historical sensing data can be unlabeled and unprocessed data. The user feedback data of the first historical sensing data can be the same type of data as the aforementioned user feedback data, but the user feedback data of the first historical sensing data is the user's feedback on the motion corresponding to the first historical sensing data.

[0033] In step S313, the processing device 12 marks the first historical sensing data according to the threshold setting and preset motion features to generate first historical marked data. The preset motion features may include the moving part and the corresponding motion type, which are the same as the moving part and motion type included in the aforementioned personalized feature marking. The threshold setting corresponding to the preset motion features may be an initial setting, while the threshold setting corresponding to the aforementioned personalized feature marking may be a setting adjusted from the initial setting. The first historical marked data may include the first historical sensing data and the corresponding feature marking. The processing device 12 may use a threshold-based automatic marking (TBAL) algorithm to mark the first historical sensing data.

[0034] In step S315, the processing device 12 can adjust the threshold setting based on user feedback data and the first historical marker data. In other words, the processing device 12 can adjust the threshold setting for marking using preset action features based on user feedback data. The implementation of step S315 is the same as the above-described method for adjusting the threshold setting for marking based on personalized features based on user feedback data, and will not be described again here.

[0035] In step S317, the processing device 12 can mark multiple second historical sensing data according to the adjusted threshold setting and the multiple preset action features to generate multiple second historical marked data. The second historical sensing data can be of the same type as the first historical sensing data. Each preset action feature can correspond to multiple second historical sensing data, and the multiple second historical sensing data for each preset action feature can have different degrees of completion. The second historical sensing data and the first historical sensing data can have the same or different generation time points. The difference between the first historical sensing data and the second historical sensing data is that the first historical sensing data is used to adjust the threshold setting, while the second historical sensing data is used to establish the normal model described later. The implementation of step S317 can be the same as step S313, and will not be repeated here.

[0036] In step S319, the processing device 12 can generate a normal model using the plurality of second historical labeled data, wherein the normal model indicates the plurality of personalized features. The normal model can be a classification model, such as based on logistic regression, decision numbers, random forests, etc., but is not limited thereto. Specifically, the normal model can include multiple preset action features, and waveform features of multiple completion levels corresponding to each preset action feature. For example, the normal model can include waveform features of 0% to 100% completion level of a specific action. In one embodiment, the processing device 12 can directly establish a normal model using the second historical labeled data and classification algorithms. In another embodiment, the processing device 12 can first generate one or more initial models using the second historical labeled data and one or more classification algorithms, and then verify the initial models (e.g., cross-validation, bootstrap resampling, etc.). When the verification result indicates that the performance of the initial model does not meet the expected performance, the threshold setting is adjusted again and the process is repeated from step S313 until the verification result of the initial model indicates that the performance of the initial model meets the expected performance, at which point the initial model is used as the normal model. After generating the normal model, the processing device 12 can use the normal model to correct the labeled data in order to automatically adjust the threshold.

[0037] Please refer to Figures 1 and 5 together, where Figure 5 is a flowchart illustrating the process of obtaining a pre-trained inference model in a health care recommendation method according to an embodiment of the present invention. As shown in Figure 5, obtaining the pre-trained inference model includes steps S411 to S417. The steps shown in Figure 5 can be performed before inputting the fused data and multi-dimensional feature data into the pre-trained inference model to generate exercise performance indicators (step S119 in Figure 2), but the present invention is not limited thereto. It should also be noted that the steps in Figure 5 can be performed by the processing device 12 in Figure 1, or by another processing device, and after the other processing device obtains the pre-trained inference model, the pre-trained inference model is output to the processing device 12. For ease of explanation, the steps in Figure 5 are described below using the processing device 12.

[0038] The implementation method of step S411 to obtain historical sensing data is the same as the method of obtaining original sensing data in step S111 of Figure 2. The difference is that the time point of generation of historical sensing data can be earlier than that of original sensing data.

[0039] In step S413, the processing device 12 can perform data fusion on the historical sensing data and extract multiple historical multi-dimensional feature data corresponding to multiple time intervals, and assign the corresponding motion performance index to each historical multi-dimensional feature data as training data.

[0040] In step S415, the processing device 12 can use the multiple training data to train and generate a multi-dimensional inference model.

[0041] In step S417, the processing device 12 can approximate and compress the multi-dimensional inference model to generate a pre-trained inference model. The processing device 12 can convert the logic of the multi-dimensional inference model into a numerical analysis process, optimize the parameters using numerical analysis methods (such as Lait, Least Squares, Lagrange multiplication, etc.), evaluate the influence of constraints in optimization based on the optimization results, and finally perform predictive analysis. Furthermore, the processing device 12 can calculate the weights of a layer of the multi-dimensional inference model into smaller bits to represent each weight, and then store the corresponding weights using a clustering algorithm, where more frequently occurring symbols can be represented by shorter bits. Accordingly, through approximate estimation and compression, important connections can be preserved to obtain a sparse network structure.

[0042] Accordingly, the pre-trained inference model generated through the above method can be implemented on edge devices, performing efficient and low-cost computations. Furthermore, personalized healthcare solutions can be generated through model inference on edge devices. During the inference process of the pre-trained inference model, the model can group similar values ​​by weight and remove features that contribute less to the inference result, further reducing computational and storage requirements.

[0043] Please refer to Figures 1 and 6 together, where Figure 6 is a flowchart illustrating the process of obtaining a pre-trained inference model in a health care recommendation method according to another embodiment of the present invention. As shown in Figure 6, obtaining the pre-trained inference model includes steps S511 to S521. The steps shown in Figure 6 can be performed before inputting the fused data and multi-dimensional feature data into the pre-trained inference model to generate exercise performance indicators (step S119 in Figure 2), but the present invention is not limited thereto. Step S511 can be the same as step S411 in Figure 5; the implementation of steps S512, S513, S515 and S517 can be the same as S112, S113, S115, S117 in Figure 2, respectively, and will not be described again here. It should also be noted that the steps in Figure 6 can be executed by the processing device 12 in Figure 1, or by another processing device, and after the other processing device obtains the pre-trained inference model, the pre-trained inference model is output to the processing device 12. For ease of explanation, the steps in Figure 6 are described below using the processing device 12.

[0044] In step S519, the processing device 12 can use historical fused data and historical multi-dimensional feature data as training data. In step S521, the processing device 12 can use the multiple training data to train and generate a multi-dimensional inference model. The multi-dimensional inference model can be used to generate the aforementioned health care plan. In one embodiment, step S521 of FIG6 can be implemented by steps S415 and S417 of FIG5.

[0045] The health care recommendation method and system described in one or more of the above embodiments can solve the current problems in quantifying the effectiveness of rehabilitation exercises. It can use sensor data related to rehabilitation medical devices to generate a personal routine model for multi-faceted and multi-dimensional feature extraction, thereby generating a personalized health care plan. In addition, it can also effectively reduce the burden on caregivers.

[0046] In summary, the health care recommendation method and system according to one or more embodiments of the present invention can perform personalized labeling, multi-dimensional feature extraction, and multivariate model data processing on sensed data. Using this data, personalized exercise performance indicators can be evaluated, and health care plans can be provided to users accordingly. This can solve the problem of manpower shortage in care institutions and improve the quality of care. Furthermore, by adjusting threshold settings based on user feedback data, even if a user's tolerance for exercise intensity is low, the threshold can be lowered to allow the user to complete exercises suitable for their physical condition; conversely, when a user's tolerance for exercise intensity is high, the threshold can be raised to allow the user to challenge exercises suitable for their physical condition. The pre-trained inference model generated through approximate estimation and compression can be implemented on edge devices, performing efficient and low-cost computations. Personalized health care plans can be generated through model inference on edge devices. During the inference process of the pre-trained inference model, the pre-trained inference model can group similar values ​​by weight and remove features that contribute less to the inference result, further reducing computational and storage requirements.

[0047] While the present invention has been disclosed above with reference to the foregoing embodiments, it is not intended to limit the invention. Any modifications and refinements made without departing from the spirit and scope of the invention are within the scope of patent protection of the present invention. For details regarding the scope of protection defined in the present invention, please refer to the appended claims.

[0048] 1: Health Care Recommendation System 11: Sensing Device 12: Processing device 13: Input / Output Devices S111,S112,S113,S115,S117,S119,S121,S211,S213,S311,S313,S315,S317,S319,S411,S413,S415,S417,S511,S512,S513,S515,S517,S519,S521: Steps

Claims

1. A health care recommendation method, executed by a processing device, comprising: acquiring multiple sets of raw sensing data from multiple sensing devices; labeling one of the raw sensing data according to multiple personalized features to generate labeled data; inputting one of the raw sensing data into a pre-trained prediction model to generate prediction data; fusing the labeled data and the prediction data to generate fused data; extracting multi-dimensional feature data from the raw sensing data; inputting the fused data and the multi-dimensional feature data into a pre-trained inference model to generate an exercise performance index; and generating a health care plan based on the exercise performance index.

2. The health care recommendation method as described in claim 1, wherein the raw sensing data includes a movement time and a movement trajectory, and the health care recommendation method further includes: calculating an acceleration change using the movement time and the movement trajectory; and performing motion smoothness analysis using the acceleration change to generate smoothness data.

3. The health care recommendation method as described in claim 2, wherein fusing the labeled data and the predicted data to generate the fused data further comprises: fusing the labeled data, the predicted data and the fluency data to generate the fused data.

4. The health care recommendation method as described in claim 2, wherein using the acceleration change to perform motion smoothness analysis to generate the smoothness data comprises: obtaining a motion type corresponding to the motion trajectory; and selecting one of a plurality of motion smoothness analysis algorithms based on the motion type to perform the motion smoothness analysis.

5. The health care recommendation method as described in claim 1 further includes: obtaining user feedback data; and adjusting a threshold setting for marking based on the user feedback data according to the personalized characteristics.

6. The health care recommendation method as described in claim 1 further comprises: obtaining a first historical sensing data and user feedback data corresponding to the first historical sensing data; marking the first historical sensing data according to a threshold setting and multiple preset action features to generate a first historical marked data; adjusting the threshold setting according to the user feedback data and the first historical marked data; marking multiple second historical sensing data according to the adjusted threshold setting and the preset action features to generate multiple second historical marked data; and generating a normality model using the second historical marked data, wherein the normality model indicates the personalized features.

7. The health care recommendation method as described in claim 1 further comprises: acquiring multiple historical sensing data from the sensing devices; generating multiple training data using the historical sensing data; training the model using the training data to generate a multidimensional inference model; and approximating and compressing the multidimensional inference model to generate the pre-trained inference model.

8. The health care recommendation method as described in claim 1 further comprises: acquiring multiple historical sensing data from the sensing devices; labeling one of the historical sensing data according to the personalized features to generate historical labeled data; inputting one of the historical sensing data into the pre-trained prediction model to generate historical predicted data; fusing the historical labeled data and the historical predicted data to generate historical fused data; extracting historical multi-dimensional feature data from the historical sensing data; using the historical fused data and the historical multi-dimensional feature data as one of multiple training data; and using the training data to generate the pre-trained inference model.

9. A health care recommendation system, comprising: a plurality of sensing devices for acquiring a plurality of raw sensing data; and a processing device connected to the sensing devices for labeling one of the raw sensing data according to a plurality of personalized features to generate labeled data, inputting one of the raw sensing data into a pre-trained prediction model to generate predicted data, fusing the labeled data and the predicted data to generate fused data, extracting multi-dimensional feature data from the raw sensing data, inputting the fused data and the multi-dimensional feature data into a pre-trained inference model to generate an exercise performance index, and generating a health care plan based on the exercise performance index.

10. The health care recommendation system as claimed in claim 9, wherein the raw sensing data includes a motion time and a motion trajectory, and the processing device is further configured to calculate an acceleration change using the motion time and the motion trajectory, perform motion smoothness analysis using the acceleration change to generate smoothness data, and fuse the labeled data, the predicted data and the smoothness data to generate the fused data.

11. The health care recommendation system as claimed in claim 10, wherein the processing device is further configured to fuse the labeled data, the predicted data and the fluency data to generate the fused data.

12. The health care recommendation system as claimed in claim 10, wherein the processing device is configured to obtain a type of motion corresponding to the motion trajectory, and select one of a plurality of motion smoothness analysis algorithms based on the motion type for performing motion smoothness analysis.

13. The health care recommendation system as claimed in claim 9, wherein the processing device is further configured to obtain user feedback data and adjust a threshold setting for marking based on the personalized characteristics based on the user feedback data.

14. The health care recommendation system of claim 9, wherein the processing device is further configured to: acquire a first historical sensing data and user feedback data corresponding to the first historical sensing data; mark the first historical sensing data according to a threshold setting and a plurality of preset action features to generate a first historical marked data; adjust the threshold setting according to the user feedback data and the first historical marked data; mark a plurality of second historical sensing data according to the adjusted threshold setting and the preset action features to generate a plurality of second historical marked data; and generate a normality model using the second historical marked data, wherein the normality model indicates the personalized features.

15. The health care recommendation system as claimed in claim 9, wherein the processing device is further configured to: acquire multiple historical sensing data from the sensing devices; generate multiple training data using the historical sensing data; train using the training data to generate a multidimensional inference model; and approximate and compress the multidimensional inference model to generate the pre-trained inference model.

16. The health care recommendation system as claimed in claim 9, wherein the processing device is further configured to: acquire multiple historical sensing data from the sensing devices; label one of the historical sensing data according to the personalized features to generate historical labeled data; input one of the historical sensing data into the pre-trained prediction model to generate historical predicted data; fuse the historical labeled data and the historical predicted data to generate historical fused data; extract historical multi-dimensional feature data from the historical sensing data; use the historical fused data and the historical multi-dimensional feature data as one of multiple training data; and generate the pre-trained inference model using the training data.

Images