A model migration method and device, equipment, program product, and storage medium

Abstract

The application discloses a model migration method and device, equipment, program product and storage medium. The method comprises the following steps: obtaining a clustering result of a crowd portrait clustering model; obtaining a first data set corresponding to the clustering result according to the clustering result; the data in the first data set is characteristic data corresponding to different motion types under the clustering result; determining an output result of a first network model according to the first data set based on the first network model, determining a first conclusion according to the output result, and the first conclusion represents the contribution of sensor information to the identification of different motion types.

Description

Technical Field

[0001] This application relates to the field of data service technology, and in particular to a model migration method, apparatus, device, program product, and storage medium. Background Technology

[0002] While providing an important means to compensate for the lack of real-time data in internet user profiling, it is not only able to analyze the traditional static profiles of people (relatively stable age, height and other behavioral habits over a period of time), but also to more accurately analyze the real-time dynamic profiles of people (subtle physical signs under different movement states). At the same time, the reuse of information data on the same limb position from different hospitals, different manufacturers and different terminals has become an urgent need. Summary of the Invention

[0003] To address the aforementioned technical problems, this application provides a model migration method, apparatus, device, program product, and storage medium.

[0004] The model transfer method provided in this application includes:

[0005] Obtain the clustering results from the user profile clustering model;

[0006] Based on the clustering results, obtain the first data set corresponding to the clustering results; the data in the first data set are the feature data corresponding to different motion types under the clustering results;

[0007] Based on the first network model, the output of the first network model is determined according to the first data set, and the first conclusion is determined based on the output. The first conclusion represents the contribution of sensor information to the identification of different motion types.

[0008] The model transfer apparatus provided in this application includes:

[0009] The acquisition unit is used to acquire the clustering results of the crowd profiling clustering model; and to acquire the first data set corresponding to the clustering results based on the clustering results; the data in the first data set are feature data corresponding to different motion types under the clustering results.

[0010] The determining unit is used to determine the output result of the first network model based on the first network model and the first data set, and to determine a first conclusion based on the output result. The first conclusion represents the contribution of sensor information to the identification of different motion types.

[0011] The model migration device provided in this application includes a processor and a memory, the memory being used to store computer programs, and the processor being used to call and run the computer programs stored in the memory to perform the above-described method.

[0012] This application provides a computer program product, comprising: a computer program that implements the above-described method when executed by a processor.

[0013] The computer-readable storage medium provided in this application is used to store a computer program that causes a computer to perform the above-described method.

[0014] In the technical solution of this application, the clustering results of a crowd profile clustering model are obtained; a first data set corresponding to the clustering results is obtained based on the clustering results; the data in the first data set are feature data corresponding to different motion types under the clustering results; the output result of the first network model is determined based on the first data set, and a first conclusion is determined based on the output result, the first conclusion representing the contribution of sensor information to the identification of different motion types. Thus, by obtaining the sensor contribution corresponding to each clustering result, it is possible to facilitate the reuse of industry capabilities and obtain user data information more efficiently. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the model transfer method provided in an embodiment of this application;

[0016] Figure 2 This is a schematic diagram of the structural composition of the model transfer device provided in the embodiments of this application;

[0017] Figure 3 This is a schematic structural diagram of a model transfer device provided in an embodiment of this application;

[0018] Figure 4 This is a schematic structural diagram of the chip according to an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will now be 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. 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.

[0020] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] It should also be noted that the terms "first," "second," and "third" used in the embodiments of this application are only used to distinguish similar objects and do not represent a specific order of objects. It is understood that "first," "second," and "third" can be interchanged in a specific order or sequence where permissible, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship. It should also be understood that the "instruction" mentioned in the embodiments of this application can be a direct instruction, an indirect instruction, or an indication of an association relationship. For example, A instructing B can mean that A directly instructs B, for example, B can be obtained through A; it can also mean that A indirectly instructs B, for example, A instructs C, and B can be obtained through C; or it can mean that there is an association relationship between A and B. It should also be understood that the term "correspondence" mentioned in the embodiments of this application may indicate a direct or indirect correspondence between the two, or an association between the two, or a relationship of instruction and being instructed, configuration and being configured, etc.

[0022] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and they all fall within the protection scope of the embodiments of this application.

[0023] For sensor data at the same limb location in a body area network, consistency judgment is a prerequisite for whether the data can be transferred. Taking the location of the foot sensor as an example, the industry has several solutions where the inertial measurement unit (IMU) is located in the lower leg, foot surface, foot sole, and foot side. Some solutions are fused with foot pressure sensors.

[0024] The need for reliable migration of limb location information collected by hospitals from different hospitals, manufacturers, and terminals provides an important means to compensate for the lack of real-time data in internet user profiling. In addition to analyzing traditional static profiles of people (relatively stable age, height, and other behavioral habits over a period of time), it also enables more accurate analysis of real-time dynamic profiles of people (subtle physical signs under different movement states). At the same time, the reuse of information data on the same limb location from different hospitals, manufacturers, and terminals has become an urgent need.

[0025] However, there is a lack of methods for reusing and reliably transferring data model capabilities for the same limb location, and for reusing experience on the contribution weights of different limb locations and different types of sensors to behavior recognition. This helps users choose different combinations of sensors for different limb locations as needed, reducing the number and dimensionality of sensors. But can the same limb location be directly reused? The same limb location, such as the foot, including the sole, instep, and ankle, may be integrated with shoes, insoles, or trouser legs when it comes to productization, resulting in different levels of comfort and cost. If the performance and effect can be consistent, capability transfer is possible; then only comfort and cost need to be considered. If there is a universal method to complete the above consistency proof, then manufacturers do not need to invest in repeated R&D and can directly reuse the experimental results that others can reuse. Therefore, how to achieve the transfer of this model becomes a problem that needs to be considered. To this end, the following technical solutions of the embodiments of this application are proposed.

[0026] To facilitate understanding of the technical solutions of the embodiments of this application, the technical solutions of this application are described in detail below through specific embodiments. The above-mentioned related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. The embodiments of this application include at least some of the following contents.

[0027] Figure 1 This is a flowchart illustrating the model transfer method provided in an embodiment of this application, as shown below. Figure 1 As shown, the model transfer method includes the following steps:

[0028] Step 101: Obtain the clustering results of the crowd profile clustering model.

[0029] In some implementations, to determine whether a model is transferable, it is first necessary to cluster the population based on its static characteristics. Different models can be used for different populations to determine applicability. This involves clustering populations with similar static characteristics. Specifically, a population profile clustering model is used to group gait data of healthy individuals with similar static human characteristics into separate classes. It can be understood that clustering results for similar static human characteristics constitutes a larger class, while clustering results for similar gait characteristics under different movement types within that larger class constitute smaller classes.

[0030] In some implementations, obtaining the clustering results of the crowd profile clustering model includes: obtaining multiple static feature parameters and multiple gait feature parameters; calculating the correlation between each static feature parameter and each gait feature parameter to obtain multiple correlation coefficients between the multiple static feature parameters and the multiple gait feature parameters; selecting target static feature parameters from the multiple static feature parameters based on the multiple correlation coefficients; selecting target gait feature parameters from the multiple gait feature parameters based on the target static feature parameters and the multiple correlation coefficients; and determining the clustering results of the crowd profile clustering model based on the selected target static feature parameters and target gait feature parameters.

[0031] In some implementations, it is first necessary to acquire static profile features of healthy individuals, including age, gender, height, and BMI, as well as multiple gait feature parameters, including gait feature parameters under different movement states, such as stride length, stride width, and stride frequency in walking, and stride length, stride width, and stride frequency in running. Then, these static feature parameter data and gait feature parameter data are acquired, and correlation calculations are performed on each static feature parameter and each gait feature parameter to obtain multiple correlation coefficients between the multiple static feature parameters and the multiple gait feature parameters. Based on the multiple correlation coefficients, target static feature parameters are selected from the multiple static feature parameters; based on the target static feature parameters and the multiple correlation coefficients, target gait feature parameters are selected from the multiple gait feature parameters; finally, the clustering result of the population profile clustering model is determined based on the selected target static feature parameters and target gait feature parameters.

[0032] In some implementations, the correlation coefficient is the Person correlation coefficient. The correlation coefficient between static feature parameters and gait feature parameters can be calculated with reference to formula (1).

[0033]

[0034] Among them, X i,k For the k-th data of the i-th static feature parameter, Y j,k For the k-th data of the j-th gait feature parameter, Let be the average value of the i-th static feature parameter data. Let r be the average value of the j-th gait feature parameter, N be the number of data in the database, and r be the average value of the j-th gait feature parameter. i,j Let be the Pearson correlation coefficient between the i-th static feature parameter and the j-th gait feature parameter.

[0035] In some implementations, the mean of the correlation coefficient for each static feature parameter is calculated, and the static feature parameter corresponding to the largest mean correlation coefficient is selected as the target static feature parameter. The specific method for calculating the mean of the correlation coefficient is shown in formula (2):

[0036]

[0037] Where L is the number of gait feature parameters. Let be the mean correlation coefficient of the i-th static feature parameter.

[0038] In some implementations, after selecting the target static features, gait feature parameters corresponding to correlation coefficients greater than a first threshold are selected from multiple correlation coefficients under the target static feature parameters as target gait feature parameters. For example, the first threshold is 0.3.

[0039] In some implementations, the clustering results of the crowd profile clustering model are obtained based on the target static feature parameters and the target gait feature parameters.

[0040] In some implementations, the method further includes: obtaining cluster boundaries of clustering results, the cluster boundaries being used to determine the cluster to which a user belongs.

[0041] In some implementations, test subjects are classified according to the K-Means gait profile clustering model and cluster boundaries, and an individual twin static profile corresponding to each user is calculated; the mean of the gait feature parameters of each group is calculated within a 95% confidence interval, and the calculation result is used as the group twin static profile boundary of that cluster category; users are assigned to the group twin static profile boundary according to the cluster boundary, and the group twin static profile is used as the individual twin static profile.

[0042] For example, based on the age of healthy individuals, a gait profile category is determined. All data under that category are retrieved from the healthy population clustered gait feature database. Then, the data within the 95% confidence interval are saved, and the mean of the gait feature data under that category is calculated, as shown in formula (3):

[0043]

[0044] Among them, Y i This represents the i-th data point under the selected gait profile category, N is the number of data points under the selected gait profile category, and μ is the mean of the gait feature data under this category.

[0045] Then, the variance of the gait feature data under this category is calculated according to formula (4).

[0046]

[0047] Where, σ 2 This represents the variance of the gait feature data for this category, where σ is the standard error.

[0048] The 95% confidence interval for specific gait characteristic data of healthy individuals is shown in formula (5):

[0049]

[0050] Where 1.96 is the coefficient of the standard error corresponding to the 95% confidence interval, and Confidence_Interval 0.95 This represents the 95% confidence interval for the calculated gait characteristic data of healthy individuals. The center value of the data within the 95% confidence interval is μ, and the boundary values ​​are the upper and lower limits of the confidence interval.

[0051] Step 102: Obtain the first data set corresponding to the clustering results; the data in the first data set are the feature data corresponding to different motion types under the clustering results.

[0052] To determine whether sensor data for the same limb range can be transferred, in addition to static feature parameters, dynamic movement types may also affect the performance of different sensors. For example, a pressure sensor may be effective in detecting walking, but may not be effective in detecting running.

[0053] In some implementations, a first data set corresponding to each clustering result is obtained. The data in the first data set consists of feature data corresponding to different movement types within the clustering result. This feature data describes / measures / determines the characteristics of the movement type; therefore, if the feature can be identified, the movement type can be confirmed. Specifically, the first data set includes movement paradigm data for different movement types of the population corresponding to each clustering result. This movement paradigm data includes gait feature data; that is, different movement types will have different movement paradigms, including gait features under different movement types. For example, by identifying gait features, the corresponding movement type can be identified. It is understood that the obtained data are feature data for different movement types within a broad category.

[0054] In some implementations, it is also necessary to obtain a static profile of the population corresponding to each clustering result.

[0055] In some implementations, a corresponding profile download request is sent to the analysis server, which calls the access interface from its aggregation middleware to download motion paradigm data and static profile models of different motion types for each population group to the cloud-based population application server. The motion paradigm data includes gait feature data.

[0056] In some embodiments, the method further includes: converting time-domain data in the first data set into frequency-domain data; and displaying the converted data in the first data set.

[0057] In some implementations, a Short-Time Fourier Transform (STFT) is applied sequentially to each channel of the processed data to analyze the time-varying characteristics of the signal in the frequency domain. The STFT obtains the spectral information within each time window, thereby generating a time-frequency spectrum, which is then used to further extract and analyze the spectral features within different time windows. The specific STFT transformation formula is shown in (6):

[0058]

[0059] Where x(n) is the input time series data, w(n) is the window function, mH is the left coordinate of the window, and N is the length of the sequence.

[0060] In some implementations, the data in the transformed first dataset is displayed.

[0061] For example, the original nine-axis data is divided into three rows, representing the time series of three data types: accelerometer, gyroscope, and compass. The data after STFT processing corresponds one-to-one with the data before processing. After STFT processing, the time series data format changes from the original location_num×type_num×channel_num×sequence_length to location_num×type_num×channel_num×time_steps×frequence_steps.

[0062] Where location_num=3 is the number of sensor locations, type_num=3 is the number of sensor data types, channel_num=3 is the number of channels for each data type of the sensor, sequence_length=500 is the length of the data time sequence, time_step=21 is the number of time steps of the data after STFT processing, and frequency_steps=26 is the number of frequency steps.

[0063] Step 103: Based on the first network model, determine the output result of the first network model according to the first data set, and determine the first conclusion based on the output result. The first conclusion represents the contribution of sensor information to the identification of different motion types.

[0064] In some implementations, the first network model is a residual convolutional neural network (CNN) model. Based on the first network model and the data in the first dataset, the output of the first network model is determined, and a first conclusion is determined based on this output. The first conclusion represents the contribution of the sensors to the recognition of different movement types. Specifically, the first network model processes movement paradigm data under different movement types, where the movement paradigm data includes data on different descriptive features collected by the sensors. In this way, the contribution of the sensors to the recognition of movement types can be obtained. For example, when the descriptive features of the movement type include gait features (gait patterns) data, the contribution of the number, type, and quantity of sensors to gait recognition can be obtained, and thus the contribution of the sensors to the recognition of different movement types can be obtained. That is, the contribution of different sensors to gait recognition under different movement types can be obtained through the calculation of the first network model. For example, the contribution of the type, quantity, and location of sensors to the recognition of gait features during running can be obtained, and it can be determined that if a user needs to measure their running performance, the location, quantity, and type of sensors worn by the user can be determined according to this result. It is understandable that for different types of motion, the descriptive features include not only gait features, but also other descriptive features of the motion type can be identified through network models to determine the corresponding motion type and to obtain the contribution of the sensor to motion type identification.

[0065] In some implementations, determining the output of the first network model based on the first data set includes: determining the output of the first network model based on the first network model and the transformed first data set.

[0066] In some implementations, the input to the first network model is data processed by the STFT, and the contribution levels of different sensors to gait recognition can be obtained through calculations by the first network model.

[0067] In some implementations, after obtaining the contribution levels of different sensors to gait recognition, a contribution level feature map can be plotted to enhance the interpretability of the model. The residual mechanism multiplies the output attention contribution data with the corresponding STFT data to highlight the effective data.

[0068] In some implementations, the residual convolutional neural network model contains a total of 4 convolutional blocks. Each convolutional block consists of a two-dimensional convolutional layer, an activation function layer, and a layer normalization layer. Due to the requirements of the residual structure, the output of the model has the same data format as the input data and is multiplied by the corresponding input data.

[0069] In some implementations, the residual convolutional neural network model channels are set. The number of channels in the four convolutional blocks are set to 64, 128, 64, and 27, respectively, the kernel size is set to 3×3, the activation function is LeakyReLU, and the final output contribution weights are multiplied by the corresponding STFT data to obtain the data after contribution weight processing.

[0070] In some implementations, the training settings for the residual convolutional neural network model are as follows: the parameters during model training are set to: learning rate = 10 -4 ,batch size=20,epoch size=100.

[0071] In some implementations, a dataset is used for validation. With the assistance of gait action category recognition task, a trained residual convolutional neural network module is finally obtained. The output data of this module is multiplied with the corresponding STFT data, and the data that is useful to the downstream classification model module is highlighted with a large weight. That is, this data is the attention contribution data, which contains the contribution data of sensor data of different positions, types and numbers.

[0072] For example, calculations and analyses based on this output result yielded the following first conclusions: the contribution weight of the sensor data located at the ankle was found to be 0.9880, which is much higher than that of the sensor data at other locations. Furthermore, the analysis also revealed that the contribution weight of the acceleration data and gyroscope data was 0.8370, which is much higher than that of the magnetometer data.

[0073] For example, the first conclusion also includes: Regarding sensor location and number: the sensor located on the foot contributes a weight of 0.8506 to gait recognition, which aligns with relevant experience and the needs of PD home systems; thus, it enables the reduction of multiple sensors originally distributed throughout the body to a single sensor located only on the foot. Among the locations for gait recognition, the foot is the most direct, while waist and leg sensors can provide some assistance; the foot includes the sole, instep, and ankle. Regarding sensor type: three-axis acceleration data and three-axis gyroscope data contribute a weight of 0.8923 to motion type (gait) recognition. Therefore, the original nine-axis IMU sensor can be reduced to a six-axis IMU.

[0074] In some embodiments, the method further includes: acquiring first information about the user, the first information including first static feature parameter data of the user and disease information of the user; determining an individual static profile of the user based on the first static feature parameter data and clustering results; and collecting a second set of data about the user based on the first conclusion and disease information, the data in the second set being used to monitor the user's health status.

[0075] In some implementations, first information about the user is acquired, including the user's first static feature parameter data and the user's disease information. The user's disease information includes the disease the user needs to be examined for, the corresponding department, the affected limb location, the relevant movement type for auxiliary examinations required for the disease, and the movement paradigm corresponding to that movement type. Based on the user's first static feature parameters, a clustering result is determined, and the group static profile corresponding to this clustering result serves as the user's individual static profile. Based on the disease information, the movement type the user needs to observe is determined. Based on the first conclusion, the type, number, and location of the sensors collecting user data corresponding to this movement type are determined, and then a second data set of the user is collected. The data in the second data set is used to monitor the user's health status.

[0076] For example, when the user is a patient, the doctor logs into the application server and, based on the patient's department, disease, affected limb location, and relevant movement types and paradigms for the auxiliary examinations required for the disease, inputs the static characteristic parameters of the patient population (age, gender, height, BMI) and corresponding dynamic profile characteristic parameters (movement state and corresponding gait characteristic parameters) to be retrieved. Based on the patient's static characteristic parameters, an individual twin static profile of the patient is determined. The relevant requirements are then used to create a patient file according to the doctor's department and disease requirements. A corresponding profile download request is sent to the analysis server, which calls the access interface from its aggregation middleware to download the movement paradigm data and individual static profile model of the corresponding population to the cloud-based population application server. Based on the first conclusion, the required data collection locations and sensor settings are obtained, and the initial data set of the patient is collected. Data meeting the collection conditions is uploaded, thus obtaining the user's second data set. For example, data meeting the criteria of more than 30 steps in straight walking and turning movements are uploaded to the cloud. For example, when a patient's condition requires observation of their walking to determine their recovery, the contribution of the sensors during walking is obtained based on the first conclusion, and the type, number, and location of the sensors used are determined.

[0077] In some implementations, a second dataset is compared with an individual static profile to monitor the patient's health status in real time.

[0078] The technical solution of this application embodiment obtains the clustering results of a crowd profile clustering model; obtains a first data set corresponding to the clustering results; the data in the first data set is feature data corresponding to different motion types under the clustering results; determines the output result of the first network model based on the first data set, and determines a first conclusion based on the output result, the first conclusion representing the contribution of sensor information to the identification of different motion types. Thus, by obtaining the sensor contribution corresponding to each clustering result, it can facilitate the reuse of industry capabilities and more efficiently obtain user data information.

[0079] The technical solutions of the embodiments of this application are illustrated below with specific examples.

[0080] For sensor data at the same limb location within a body area network (BAN), consistency is a prerequisite for data transfer. Taking foot sensor locations as an example, the industry offers several solutions, including IMUs located on the lower leg, instep, sole, and side of the foot. Some solutions integrate with foot pressure sensors. Hospitals require reliable transfer of limb location information collected from different hospitals, manufacturers, and terminals. This provides a crucial means to address the real-time limitations of internet user profiling. Besides analyzing traditional static profiles (relatively stable age, height, and other behavioral habits over a period of time), it also enables more accurate analysis of real-time dynamic profiles (subtle vital signs under different movement states). Furthermore, reusing information from the same limb location across different hospitals, manufacturers, and terminals is urgently needed.

[0081] However, currently, there is a lack of methods for reusing and reliably transferring data model capabilities for the same limb location, and for reusing experience on the contribution weights of different limb locations and different types of sensors to behavior recognition. This would help users choose different combinations of sensors for different limb locations as needed, reducing the number and dimensionality of sensors. But can the same limb location be directly reused? The same limb location, such as the foot, including the sole, instep, and ankle, may be integrated with shoes, insoles, or trouser legs when it comes to productization, resulting in different levels of comfort and cost. If the performance and effects can be consistent and capabilities can be transferred, then only comfort and cost need to be considered. If there is a universal method to complete the above consistency proof, then manufacturers would not need to invest in repeated R&D and could directly reuse the experimental results that others could reuse.

[0082] This application aims to provide a method that integrates with the information service capabilities of a carrier's body area network, assisting medical service institutions and related wearable system service providers in confirming data consistency and reusing and migrating models for the same limb position among different groups in post-diagnosis health management scenarios.

[0083] This application proposes a method for evaluating the consistency of sensor data fusion at the same limb position in behavior recognition based on body area networks.

[0084] Step 1: Import static feature parameters (age, gender, height, BMI) and dynamic feature parameters (exercise status) of healthy individuals, and iterate through the static feature parameters (demographic static features) respectively.

[0085] Step 2: Calculate the Pearson correlation coefficient between each demographic indicator and gait feature parameter according to formula (1); calculate the mean of the correlation coefficients according to formula (2), select the static feature parameter corresponding to the largest mean of the correlation coefficients as the target static feature parameter, and after selecting the target static feature, select the gait feature parameter corresponding to the correlation coefficient greater than 0.3 from multiple correlation coefficients as the target gait feature parameter under the target static feature parameter.

[0086] Step 3: Based on the K-Means gait profile clustering model, the clustering results are determined according to the target static feature parameters and the target gait feature parameters. The cluster boundaries are then obtained, and users are classified according to these boundaries. A corresponding individual twin static profile is calculated for each user. Specifically, the mean of the gait feature parameters for each group is calculated within a 95% confidence interval. This calculated result is used as the boundary of the individual twin static profile for each gait cluster category. Users are then assigned to individual twin static profiles based on these cluster boundaries. The specific method for calculating the cluster boundary is as described above and will not be repeated here.

[0087] To determine whether sensor data for the same limb range can be transferred, in addition to static crowd characteristics, dynamic movement types may also affect the performance of different sensors. For example, a pressure sensor may be effective in detecting walking, but may not be effective in detecting running.

[0088] Step 4: The doctor logs into the application server and, based on the department, the disease requiring the request, the affected limb position, the relevant movement type and movement paradigm for the auxiliary examinations, inputs the static characteristic parameters (age, gender, height, BMI) and corresponding dynamic characteristic parameters (movement state and gait characteristic parameters corresponding to each movement state) of the patient population to be retrieved; the relevant requirements are then used to create patient files according to the doctor's department and disease needs. Based on the patient's information and the results obtained from the following model, the sensor information for the patient's measurement data is determined.

[0089] Step 5: Send a corresponding profile download request to the analysis server, call the access interface from its aggregation middleware, and download the motion paradigm data and static profile model for each population group to the cloud-based population application server. The motion paradigm data includes gait feature data. When training the residual convolutional neural network model described below, download the motion paradigm data and static profile model corresponding to each clustering result to the cloud-based population server for subsequent data processing and model training. For each patient, download the motion paradigm data and corresponding static profile model for the patient's population to the cloud-based population application server to compare subsequently collected patient data with the corresponding motion paradigm data and monitor the patient's health status.

[0090] Step 6: Apply the Short Time Fourier Transform (STFT) sequentially to each channel of each prepared data point to analyze the time-varying characteristics of the signal in the frequency domain. The specific calculation formula is shown in formula (6).

[0091] Step 7: Obtain the spectral information within each time window through STFT, thereby generating a time-frequency spectrum, and further extract and analyze the spectral characteristics within different time windows.

[0092] Step 8: Identify and differentiate gait patterns for different activity types.

[0093] Step Nine: Visualization and Comparison Before and After Processing. The original nine-axis data is divided into three rows, representing the time series data of three types: accelerometer, gyroscope, and compass. The data positions correspond one-to-one with the data before and after STFT processing. After STFT processing, the time series data format changes from the original location_num×type_num×channel_num×sequence_length to location_num×type_num×channel_num×time_steps×frequence_steps.

[0094] Where location_num=3 is the number of sensor locations, type_num=3 is the number of sensor data types, channel_num=3 is the number of channels for each data type of the sensor, sequence_length=500 is the length of the data time sequence, time_step=21 is the number of time steps of the data after STFT processing, and frequency_steps=26 is the number of frequency steps.

[0095] Step 10: Upload data that meets the measurement criteria. For example, upload data that meets the criteria of more than 30 steps in both straight-line walking and turning movements to the cloud.

[0096] Design a residual convolutional neural network (CNN) model. The input of this model is data processed by STFT. The model can calculate the contribution level of different sensors to gait recognition and draw a feature map of the contribution level to enhance the interpretability of the model. The residual mechanism multiplies the output attention contribution data with the corresponding STFT data to highlight the effective data.

[0097] Step 11: Generate a residual convolutional neural network model, which contains 4 convolutional blocks. Each convolutional block consists of a two-dimensional convolutional layer, an activation function layer, and a normalization layer. Due to the requirements of the residual structure, the output of the model has the same data format as the input data and is multiplied by the corresponding input data.

[0098] Step 12: Channel Setup. The number of channels for the four convolutional blocks are set to 64, 128, 64, and 27 respectively. The kernel size is set to 3×3 for all convolutional blocks. The LeakyReLU activation function is used. The final output contribution weights are multiplied by the corresponding STFT data to obtain the data after contribution weight processing.

[0099] Step 13: Model training settings. Set the parameters for training the model to: learning rate = 10 -4 The batch size is 20, and the epoch size is 100. The model is trained using the data processed by the aforementioned STFT. The specific training process will not be described in detail here.

[0100] Step Fourteen: Validation using the dataset. With the assistance of the gait action category recognition task, the trained residual convolutional neural network module is finally obtained. The output data of this module is multiplied with the corresponding STFT data, highlighting the data that is useful to the downstream classification model module with a large weight. That is, this data is the attention contribution data, which contains contribution data from different locations and different types of sensors.

[0101] Step 15: Based on this output, calculations and analysis were performed. It was found that the contribution weight of the sensor data located at the ankle was 0.9880, which was much higher than that of the sensor data at other locations. In addition, the analysis also found that the contribution weight of the acceleration data and gyroscope data was 0.8370, which was much higher than that of the magnetometer data.

[0102] The research results show that: 1. Regarding sensor location and number: The sensor located on the foot contributes a weight of 0.8506 to gait recognition, which aligns with relevant experience and the needs of PD home systems; thus, it allows for the reduction of multiple sensors originally distributed throughout the body to a single sensor located only on the foot. Among gait recognition sites, the foot is the most directly affected, while waist and leg sensors can provide some assistance; foot sensors are located on the sole, instep, and ankle. 2. Regarding sensor type: Three-axis accelerometer data and three-axis gyroscope data contribute a weight of 0.8923 to gait recognition. Therefore, we can reduce the original nine-axis IMU sensor to a six-axis IMU.

[0103] The key point of this application's embodiments is that, within the framework of body area network (BNB) and cloud collaboration, it assists medical service institutions and related wearable system service providers in achieving data consistency verification and model reuse migration for the same limb position across different populations in post-diagnosis health management scenarios. It combines the dynamics of the BNB with the information service capabilities of the operator's BNB, providing an end-to-end process for medical service institutions and related wearable system service providers to verify data consistency and reuse migration for the same limb position across different populations. Furthermore, by combining different steps and the most suitable network elements, it offers an open method for behavior recognition that allows for the reuse of industry experience. This facilitates the reuse of industry capabilities, helps operators better leverage their network capabilities, and contributes to the development of the wearable industry.

[0104] Figure 2 This is a schematic diagram of the structural composition of the model transfer device provided in the embodiments of this application, as shown below. Figure 2 As shown, the model transfer device includes:

[0105] The acquisition unit 201 is used to acquire the clustering results of the crowd profile clustering model; and to acquire a first data set corresponding to the clustering results based on the clustering results; the data in the first data set is feature data corresponding to different motion types under the clustering results.

[0106] The determining unit 202 is used to determine the output result of the first network model based on the first network model and the first data set, and to determine a first conclusion based on the output result. The first conclusion represents the contribution of sensor information to the identification of different motion types.

[0107] In some embodiments, the acquisition unit 201 is used to acquire multiple static feature parameters and multiple gait feature parameters; perform correlation calculations on each static feature parameter and each gait feature parameter to obtain multiple correlation coefficients between the multiple static feature parameters and the multiple gait feature parameters; based on the multiple correlation coefficients, select target static feature parameters from the multiple static feature parameters; based on the target static feature parameters and the multiple correlation coefficients, select target gait feature parameters from the multiple gait feature parameters; and determine the clustering result of the crowd profile clustering model based on the selected target static feature parameters and target gait feature parameters.

[0108] In some implementations, the acquisition unit 201 is used to acquire the cluster boundaries of the clustering results, and the cluster boundaries are used to determine the cluster results to which the user belongs.

[0109] In some embodiments, the device further includes a conversion unit 203 and a display unit 204; the conversion unit 203 is used to convert time-domain data in the first data set into frequency-domain data; the display unit 204 is used to display the converted data in the first data set.

[0110] In some implementations, the determining unit 202 is used to determine the output result of the first network model based on the transformed first data set.

[0111] In some embodiments, the acquisition unit 201 is used to acquire first information of the user, the first information including first static feature parameter data of the user and disease information of the user; the determination unit 202 is used to determine the individual static profile of the user based on the first static feature parameter data and clustering results; and to collect a second data set of the user based on the first conclusion and disease information, the data in the second data set being used to monitor the user's health status.

[0112] Those skilled in the art should understand that Figure 2 The functions of each unit in the model transfer device shown can be understood by referring to the relevant descriptions of the aforementioned methods. Figure 2 The functions of each unit in the model transfer device shown can be implemented by a program running on a processor or by specific logic circuits.

[0113] Figure 3 This is a schematic structural diagram of a model transfer device 300 provided in an embodiment of this application. Figure 3 The model transfer device 300 shown includes a processor 310, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0114] Optionally, such as Figure 3As shown, the model transfer device 300 may further include a memory 320. The processor 310 can retrieve and run computer programs from the memory 320 to implement the methods described in this embodiment.

[0115] The memory 320 can be a separate device independent of the processor 310, or it can be integrated into the processor 310.

[0116] Optionally, such as Figure 3 As shown, the model transfer device 300 may also include a transceiver 330, which the processor 310 can control to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0117] The transceiver 330 may include a transmitter and a receiver. The transceiver 330 may further include an antenna, and the number of antennas may be one or more.

[0118] The model transfer device 300 can implement the corresponding processes implemented by the model transfer apparatus in the various methods of the embodiments of this application, and for the sake of brevity, it will not be described in detail here.

[0119] Figure 4 This is a schematic structural diagram of the chip according to an embodiment of this application. Figure 4 The chip 400 shown includes a processor 410, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0120] Optionally, such as Figure 4 As shown, chip 400 may further include memory 420. Processor 410 can retrieve and run computer programs from memory 420 to implement the methods described in this embodiment.

[0121] The memory 420 can be a separate device independent of the processor 410, or it can be integrated into the processor 410.

[0122] Optionally, the chip 400 may also include an input interface 430. The processor 410 can control the input interface 430 to communicate with other devices or chips; specifically, it can acquire information or data sent by other devices or chips.

[0123] Optionally, the chip 400 may also include an output interface 440. The processor 410 can control the output interface 440 to communicate with other devices or chips, specifically, to output information or data to other devices or chips.

[0124] This chip can implement the corresponding processes implemented by the model transfer device in the various methods of the embodiments of this application, which will not be described in detail here for the sake of brevity.

[0125] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0126] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0127] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0128] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0129] This application also provides a computer program product, including a computer program.

[0130] When executed by the processor, the computer program implements the corresponding processes implemented by the model transfer device in the various methods of the embodiments of this application, which will not be described in detail here for the sake of brevity.

[0131] This application also provides a computer-readable storage medium for storing computer programs.

[0132] The computer program causes the computer to execute the corresponding processes implemented by the model transfer apparatus in the various methods of the embodiments of this application, which will not be described in detail here for the sake of brevity.

[0133] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0134] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0135] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0138] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0139] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model transfer method, characterized in that, The method includes: Obtain the clustering results from the user profile clustering model; A first data set corresponding to the clustering results is obtained based on the clustering results; the data in the first data set is feature data corresponding to different motion types under the clustering results; Based on the first network model, the output result of the first network model is determined according to the first data set, and a first conclusion is determined based on the output result. The first conclusion represents the contribution of sensor information to the identification of different motion types.

2. The method according to claim 1, characterized in that, The clustering results obtained from the population profile clustering model include: Obtain multiple static feature parameters and multiple gait feature parameters; Correlation calculations are performed on each static feature parameter and each gait feature parameter to obtain multiple correlation coefficients between the multiple static feature parameters and the multiple gait feature parameters; based on the multiple correlation coefficients, target static feature parameters are selected from the multiple static feature parameters. Based on the target static feature parameters and the plurality of correlation coefficients, target gait feature parameters are selected from the plurality of gait feature parameters; The clustering results of the crowd profile clustering model are determined based on the selected target static feature parameters and target gait feature parameters.

3. The method according to claim 2, characterized in that, The method further includes: Obtain the cluster boundaries of the clustering results, which are used to determine the cluster to which the user belongs.

4. The method according to claim 1, characterized in that, The method further includes: Convert the time-domain data in the first dataset into frequency-domain data; Display the data in the first dataset after the transformation.

5. The method according to claim 4, characterized in that, The step of determining the output result of the first network model based on the first data set includes: Based on the first network model, the output result of the first network model is determined according to the transformed first data set.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain the user's first information, which includes the user's first static characteristic parameter data and the user's disease information; The user's individual static profile is determined based on the first static feature parameter data and the clustering results; and a second data set of the user is collected based on the first conclusion and the disease information, the data in the second data set being used to monitor the user's health status.

7. A model transfer device, characterized in that, The device includes: The acquisition unit is used to acquire the clustering results of the crowd profile clustering model; acquire a first data set corresponding to the clustering results based on the clustering results; the data in the first data set is feature data corresponding to different motion types under the clustering results; The determining unit is configured to determine the output result of the first network model based on the first data set, and determine a first conclusion based on the output result, wherein the first conclusion represents the contribution of sensor information to the identification of different motion types.

8. A model transfer device, characterized in that, include: A processor and a memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the method as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, include: A computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the method as described in any one of claims 1 to 6.

Images