Multi-dimensional exercise prescription system based on cloud computing

Abstract

The present invention discloses a multi-dimensional exercise prescription formulation system based on cloud computing. The order of multiple exercise mode categories that meet user information is found to obtain multiple two-dimensional exercise connectivity graphs. Among the multiple exercise connectivity graphs, a exercise connectivity graph with a similar and frequently occurring order of exercise mode categories is found. The exercise connectivity graph is used to represent multiple paths in the graph structure, thereby representing the connection methods of multiple sequences of multiple exercise mode categories. The exercise connectivity graph is traversed to obtain multiple exercise mode routes after considering user information. A three-dimensional exercise parameter point graph is constructed according to the exercise mode category as the horizontal coordinate and the corresponding human body information of a user before and after exercise as the vertical coordinate. A 2*2*2 three-dimensional convolution kernel is used to detect the three-dimensional change characteristics of the exercise parameter point graph. The second exercise convolution network is used to detect the characteristics of the change of the human body information corresponding to a movement mode category.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a multi-dimensional exercise prescription formulation system based on cloud computing. Background Art

[0002] At present, the correct way of exercising can change the user's physical condition, making the human body healthier during exercise. However, the current method of specifying exercise prescriptions generally directly uses multi-dimensional information to divide users and build user portraits to give the corresponding ranking of exercise categories, or directly uses deep learning to make judgments.

[0003] Smart wearable devices collect the user's heart rate data during exercise, and the obtained GPS data is converted into the user's speed information. However, the data obtained are all unsupervised, multimodal time series data. It is very necessary to analyze the characteristics of time series data to obtain the user's exercise ability and recommend exercise prescriptions suitable for the user's own ability. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-dimensional exercise prescription formulation system based on cloud computing to solve the above-mentioned problems existing in the prior art.

[0005] The embodiment of the present invention provides a multi-dimensional exercise prescription formulation system based on cloud computing, including:

[0006] A monitoring module is configured to obtain multi-dimensional user information, exercise body information, and exercise mode categories; the user information represents body information that does not change with exercise; the exercise body information includes a plurality of post-exercise body sign values ​​and corresponding pre-exercise body sign values; the post-exercise body sign values ​​represent the values ​​of the body sign before exercise; and the post-exercise body sign values ​​represent the values ​​of the body sign after exercise;

[0007] The processor is used to process the following method:

[0008] Based on the user information and the exercise mode category, finding a sequence that matches the user's exercise mode category to obtain a exercise connectivity graph;

[0009] Based on the motion connectivity graph, a basic motion mode set is obtained by traversing the graph; the basic motion mode set includes a plurality of motion mode routes; the motion mode routes represent the order of motion mode categories that meet the user information;

[0010] Through the movement mode prediction network, based on the moving human body information, basic movement mode set and movement mode category, the changes in the user's physical signs during exercise are judged to obtain an exercise prescription; the exercise prescription represents the ranking of the movement mode categories suitable for the user's physical signs.

[0011] Optionally, the step of using an exercise pattern prediction network to discern changes in the user's physical signs during exercise based on the moving human body information, the basic exercise pattern set, and the exercise pattern category to obtain an exercise prescription includes:

[0012] Based on the information of the moving human body and the movement mode category, a relationship between multiple movement mode categories is constructed to obtain a movement parameter dot graph; the movement parameter dot graph represents the changes in the physical signs of the human body before and after the movement corresponding to the movement mode category;

[0013] Based on the motion parameter point graph, a motion mode prediction network is used to determine the relationship between motion mode categories and obtain multiple motion impact orders;

[0014] Find the exercise route corresponding to the order of exercise impact in the basic exercise method set and obtain the exercise prescription.

[0015] Optionally, the motion mode prediction network includes a first motion convolutional network, a second motion convolutional network and a third motion network.

[0016] Optionally, the determining, based on the motion parameter point graph, through a motion mode prediction network, the relationship between motion mode categories to obtain the motion impact order includes:

[0017] Inputting the motion parameter point map into a first motion convolutional network to determine the user's motion state and obtain a first feature map;

[0018] The number of the exercise mode categories is used as the number of exercise mode categories;

[0019] The segmentation length is obtained by dividing the horizontal coordinate corresponding to the lower right corner of the motion parameter point graph by the number of motion mode categories;

[0020] On the horizontal axis, the point with the distance to the dividing length is used as the dividing point;

[0021] Segmenting the motion parameter point map according to the segmentation points to obtain a plurality of first motion parameter segmentation maps;

[0022] Keeping the lengths of the plurality of first motion parameter segmentation maps unchanged, the plurality of first motion parameter segmentation maps are superimposed in width to obtain a superimposed image;

[0023] Inputting the superimposed image into a second motion convolutional network to obtain a second feature map;

[0024] The first feature map and the second feature map are input into a third motion network to obtain a motion influence sequence.

[0025] Optionally, the convolution kernel of the first motion convolution network is a three-dimensional convolution kernel of 2*2*2, and the convolution kernel of the second motion convolution network is a three-dimensional convolution kernel with a width equal to the length of the motion parameter point map, a height of 2, and a depth equal to the number of pages of the motion parameter point map.

[0026] Optionally, the finding, based on the user information and the exercise type category, a sequence matching the user's exercise type category to obtain an exercise connectivity graph includes:

[0027] Based on the user information, a plurality of user motion connectivity graphs are obtained through a user prediction network; the user motion connectivity graphs have motion mode categories as vertices;

[0028] Based on multiple user motion connectivity graphs, similarities are identified and a motion connectivity graph is obtained.

[0029] Optionally, determining similarities based on multiple user motion connectivity graphs to obtain a motion connectivity graph includes:

[0030] Input the two user motion connectivity graphs into the similarity discrimination network to judge the similarity of the two user motion connectivity graphs and obtain the motion similarity value; the p user motion connectivity graphs correspond to (p-1)*p / 2 motion similarity values;

[0031] Obtain the motion similarity value that is smaller than other motion similarity values ​​among the (p-1)*p / 2 motion similarity values ​​as the first similarity cluster center;

[0032] Obtaining a similarity gap value; the similarity gap value represents a fixed value from the first similarity cluster center;

[0033] The sum of the first similar cluster center and the similarity gap value is used as the gap threshold;

[0034] Constructing a motion similarity set from motion similarity values ​​smaller than a gap threshold among the multiple motion similarity values;

[0035] The user motion connectivity graph corresponding to the motion similarity set is used as a similar motion connectivity graph;

[0036] A similar motion connectivity graph is randomly obtained from multiple similar motion connectivity graphs as a motion connectivity graph.

[0037] Optionally, the basic motion mode set is obtained by traversing the motion connectivity graph, including:

[0038] Obtaining m initial motion values; wherein the initial motion values ​​represent m motion mode categories in the motion connectivity graph;

[0039] According to the initial movement value, the movement order of the movement mode category is traversed to obtain n movement mode routes; m initial movement values ​​correspond to m*n movement mode routes;

[0040] The m*n movement mode routes constitute a basic movement mode set.

[0041] Optionally, the step of constructing a relationship between multiple movement types based on the moving human body information and the movement type types to obtain a movement parameter point graph includes:

[0042] Acquire a blank image; the blank image is a binary image;

[0043] The horizontal axis of the blank image represents multiple categories of movement patterns; the vertical axis of the blank image represents a value that can reflect the change value of all movement signs;

[0044] Construct a coordinate axis based on the blank image; the lower left corner of the blank image is used as the origin, the length of the blank image is used as the horizontal coordinate, and the width of the blank image is used as the vertical coordinate;

[0045] Subtract the value of the body sign before exercise from the value of the body sign after exercise to obtain the value of the change of the body sign;

[0046] The motion sign change values ​​corresponding to the multiple motion mode categories are punctuated in a blank image to obtain a two-dimensional motion parameter point map; multiple post-exercise human body sign values ​​and the corresponding pre-exercise human body sign values ​​are correspondingly obtained to obtain multiple two-dimensional motion parameter point maps;

[0047] Multiple two-dimensional motion parameter point maps are superimposed to construct a three-dimensional image and obtain a motion parameter point map.

[0048] Optional methods for training user prediction networks include:

[0049] Acquire multiple training user information and corresponding multiple annotated user motion connectivity graphs; one training user information corresponds to multiple annotated user motion connectivity graphs;

[0050] Inputting the training user information into a user prediction network to obtain a predicted user motion connectivity graph;

[0051] The loss is calculated by comparing the predicted user motion connectivity graph with the labeled user motion connectivity graph to obtain a trained user prediction network.

[0052] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0053] An embodiment of the present invention also provides a multi-dimensional exercise prescription formulation system based on cloud computing.

[0054] In the present invention, because different user information has different motion mode categories and corresponding motion sequences of motion mode categories, the order of multiple motion mode categories that meet the user information is found and represented by a graph structure to obtain multiple two-dimensional motion connectivity graphs. A motion connectivity graph with a similar and frequently occurring order of motion mode categories is found in the multiple motion connectivity graphs. The motion connectivity graph is used to represent multiple paths in the graph structure, thereby representing the connection modes of multiple motion mode categories in multiple sequences. The connection modes of complex motion mode categories are represented by a connectivity graph constructed by 0 and 1, which is convenient for storage and calculation. The motion connectivity graph is traversed to obtain multiple motion mode routes after considering the user information. A three-dimensional motion parameter point graph is constructed according to the motion mode category as the horizontal coordinate and the corresponding motion human body information of a user before and after exercise as the vertical coordinate. The three-dimensional change features of the motion parameter point graph are detected using a 2*2*2 three-dimensional convolution kernel. The second motion convolution network is used to detect the change features of the motion human body information corresponding to a motion mode category. The technical effect of being able to more accurately obtain the changes in the user's human body information before and after exercise, thereby more accurately finding an exercise prescription that suits the user's physique and can meet the user's human body information before and after exercise is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a method flow chart of a multi-dimensional exercise prescription formulation system based on cloud computing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The present invention will be described in detail below with reference to the accompanying drawings. Example

[0057] like Figure 1 As shown, an embodiment of the present invention provides a multi-dimensional exercise prescription formulation system based on cloud computing, the system including a monitoring module and a processor:

[0058] A monitoring module is configured to obtain multi-dimensional user information, exercise body information, and exercise mode categories; the user information represents body information that does not change with exercise; the exercise body information includes a plurality of post-exercise body sign values ​​and corresponding pre-exercise body sign values; the post-exercise body sign values ​​represent the values ​​of the body sign before exercise; and the post-exercise body sign values ​​represent the values ​​of the body sign after exercise;

[0059] The post-exercise body sign value and the corresponding pre-exercise body sign value are separated by a fixed time length.

[0060] In this embodiment, the moving body category is used as the key, and the moving body information is used as the value to construct a key-value pair.

[0061] In this embodiment, the user information includes age and gender.

[0062] Among them, in this embodiment, the human body information includes blood pressure, body temperature, pulse-heart rate, cardiac output, blood sugar, body temperature, body fat percentage, and skeletal muscle content.

[0063] For example, if a user is obese, it means that their body fat percentage is too high, and they need to increase their aerobic exercise training to reduce fat accumulation. If the user's skeletal muscle mass is low, they need to increase strength training to improve muscle strength and endurance.

[0064] The processor is used to process the following method:

[0065] S101: Based on the user information, determine the order of the exercise mode categories adapted by the user and obtain a movement connectivity graph;

[0066] The graph structure in the data structure is constructed by graph theory using multiple optional motion modes as points and the order of motion mode categories as edges. The motion connectivity graph is used to represent the edges in the graph structure, that is, the order of motion mode categories.

[0067] The exercise mode categories represent the categories of exercise modes that can be performed, and in this embodiment, include swimming, jogging, and other exercise mode categories.

[0068] S102: traversing the image based on the connection graph to obtain a basic movement mode set; the basic movement mode set includes multiple movement mode routes; the movement mode routes represent the order of movement mode categories that meet user information.

[0069] S103: Detecting the user's exercise state through an exercise mode prediction network based on the exercise user information and a basic exercise mode set, and obtaining an exercise prescription; the exercise prescription represents a ranking of exercise mode categories suitable for the user's physical signs.

[0070] Optionally, detecting the user's motion state and obtaining an exercise prescription by using a motion mode prediction network based on the moving human body information and a basic motion mode set may include:

[0071] Based on the moving human body information and the movement mode category, the relationship between multiple movement mode categories is constructed to obtain a movement parameter point diagram; the movement parameter point diagram represents the changes in the physical signs of the human body before and after exercise corresponding to the movement mode category.

[0072] Based on the motion parameter point graph, a motion mode prediction network is used to determine the relationship between motion mode categories and obtain multiple motion impact sequences.

[0073] Find the exercise route corresponding to the order of exercise impact in the basic exercise method set and obtain the exercise prescription.

[0074] Optionally, the motion mode prediction network includes a first motion convolutional network, a second motion convolutional network and a third motion network.

[0075] In this embodiment, the first motion convolutional network is a yolov5 model, the second motion convolutional network is a convolutional neural network (CNN), and the third motion network is a fully-connected neural network (FCNN).

[0076] Optionally, the determining, based on the motion parameter point graph, through a motion mode prediction network, the relationship between motion mode categories to obtain the motion impact order includes:

[0077] The motion parameter point map is input into a first motion convolutional network to determine the user's motion state and obtain a first feature map.

[0078] Among them, the trained first motion convolutional network represents the ranking of the user's corresponding motion categories under multiple historical data.

[0079] The number of the movement pattern categories is used as the number of movement pattern categories.

[0080] In this embodiment, there are 16 types of exercise categories.

[0081] The segmentation length is obtained by dividing the horizontal coordinate corresponding to the lower right corner of the motion parameter point graph by the number of motion mode categories;

[0082] On the horizontal axis, the point with the distance to the dividing length is used as the dividing point.

[0083] In this embodiment, if the segmentation length is 32, the positions corresponding to 32, 64, 96, 128 ... 480, 512 in the horizontal coordinate are used as segmentation points.

[0084] The motion parameter point map is segmented according to the segmentation points to obtain a plurality of first motion parameter segmentation maps.

[0085] In this embodiment, the length of the first motion parameter segmentation map is 32, the width is equal to the width of the motion parameter point map, and the number of pages is equal to the number of pages of the motion parameter point map.

[0086] The plurality of first motion parameter segmentation maps are kept unchanged in length and are superimposed in width to obtain a superimposed image.

[0087] The plurality of first motion parameter segmentation maps are sequentially superimposed according to the segmentation points from small to large.

[0088] In this embodiment, the total number of the first motion parameter segmentation maps is used as the number of segmentations. The length of the superimposed image is equal to 32, the width is equal to the product of the width of the first motion parameter segmentation map and the number of segmentations, and the number of pages is equal to the number of pages of the first motion parameter segmentation map.

[0089] Inputting the superimposed image into a second motion convolutional network to obtain a second feature map;

[0090] The first feature map and the second feature map are input into a third motion network to obtain a motion influence sequence.

[0091] Optionally, the convolution kernel of the first motion convolution network is a three-dimensional convolution kernel of 2*2*2, and the convolution kernel of the second motion convolution network is a three-dimensional convolution kernel with a width equal to the length of the motion parameter point map, a height of 2, and a depth equal to the number of pages of the motion parameter point map.

[0092] Among them, the first motion convolution network is used to detect the three-dimensional change characteristics of the motion parameter point map, and the second motion convolution network is used to detect the change characteristics of the moving human body information corresponding to a motion mode category.

[0093] Optionally, the finding, based on the user information and the exercise type category, a sequence matching the user's exercise type category to obtain an exercise connectivity graph includes:

[0094] Based on the user information, a plurality of user motion connectivity graphs are obtained through a user prediction network; the user motion connectivity graphs have motion mode categories as vertices;

[0095] The user movement connectivity graph is a representation of a graph structure.

[0096] Based on multiple user motion connectivity graphs, similarities are identified and a motion connectivity graph is obtained.

[0097] Optionally, determining similarities based on multiple user motion connectivity graphs to obtain a motion connectivity graph includes:

[0098] The two user motion connectivity graphs are input into the similarity discrimination network to determine the similarity between the two user motion connectivity graphs and obtain the motion similarity value. For each p user motion connectivity graph, (p-1)*p / 2 motion similarity values ​​are obtained.

[0099] In this embodiment, the motion similarity value ranges from 0 to 1. p is a natural number greater than 0.

[0100] Obtain the motion similarity value that is smaller than other motion similarity values ​​among the (p-1)*p / 2 motion similarity values ​​as the first similarity cluster center;

[0101] A similarity gap value is obtained; the similarity gap value represents a fixed value from the first similarity cluster center.

[0102] In this embodiment, the similarity gap value is 0.15.

[0103] The sum of the first similar cluster center and the similarity gap value is used as the gap threshold;

[0104] A motion similarity set is constructed by using motion similarity values ​​that are smaller than a gap threshold among multiple motion similarity values.

[0105] The motion similarity set includes multiple motion similarity values ​​that are smaller than a gap threshold.

[0106] The user motion connectivity graph corresponding to the motion similarity set is used as a similar motion connectivity graph;

[0107] A similar motion connectivity graph is randomly obtained from multiple similar motion connectivity graphs as a motion connectivity graph.

[0108] Optionally, the basic motion mode set is obtained by traversing the motion connectivity graph, including:

[0109] Obtain m motion initial values; the motion initial values ​​represent m motion mode categories in the motion connectivity graph.

[0110] For example, the first row and first column of the motion connectivity graph represent swimming, while the second row and second column represent jogging. The first row and second column are 1, indicating that the partial sorting of the motion categories is such that swimming is the first value and jogging is the second value.

[0111] According to the movement initial value, the movement sequence of the movement mode category is traversed to obtain n movement mode routes; m movement initial values ​​correspond to obtaining m*n movement mode routes.

[0112] Wherein, m and n are natural numbers greater than 0.

[0113] The m*n movement mode routes constitute a basic movement mode set.

[0114] Optionally, the step of constructing a relationship between multiple movement types based on the moving human body information and the movement type types to obtain a movement parameter point graph includes:

[0115] Acquire a blank image; the blank image is a binary image.

[0116] In this embodiment, the blank image is 512*512.

[0117] In this embodiment, all values ​​in the blank image are 255, indicating white.

[0118] The horizontal axis of the blank image represents multiple categories of motion patterns; the vertical axis of the blank image represents a numerical value that can reflect the change values ​​of all motion signs.

[0119] Construct a coordinate axis based on the blank image; the lower left corner of the blank image is used as the origin, the length of the blank image is used as the horizontal coordinate, and the width of the blank image is used as the vertical coordinate;

[0120] Subtract the value of the body sign before exercise from the value of the body sign after exercise to obtain the value of the change of the body sign;

[0121] The horizontal axis of the blank image represents a plurality of categories of motion modes; the vertical axis of the blank image represents a numerical value that can reflect the change values ​​of all motion signs.

[0122] In this embodiment, the position with a horizontal coordinate of 16 is used as the first exercise type category, and 16 + 32 is used as the second exercise type category. For example, in a two-dimensional exercise parameter point graph corresponding to body temperature, using 512 / 32 = 16, the position with a vertical coordinate of 16 is used as the value corresponding to a body temperature of 15 degrees Celsius, the position with a vertical coordinate of 32 is used as the value corresponding to a body temperature of 16 degrees Celsius, and so on, until the position with a vertical coordinate of 512 is used as the value corresponding to a body temperature of 46 degrees Celsius.

[0123] The motion sign change values ​​corresponding to the multiple motion mode categories are punctuated in a blank image to obtain a two-dimensional motion parameter point map; multiple post-exercise human body sign values ​​and the corresponding pre-exercise human body sign values ​​are correspondingly obtained to obtain multiple two-dimensional motion parameter point maps;

[0124] In this embodiment, a value of 0 is used for punctuation, that is, it is represented as a black dot in a blank image.

[0125] Multiple two-dimensional motion parameter point maps are superimposed to construct a three-dimensional image and obtain a motion parameter point map.

[0126] Among them, the length of the motion parameter point graph is equal to the length of the two-dimensional motion parameter point graph, the width of the motion parameter point graph is equal to the width of the two-dimensional motion parameter point graph, and the number of pages of the motion parameter point graph is equal to the number of post-exercise human body sign values ​​of the two-dimensional motion parameter point graph.

[0127] Optional methods for training user prediction networks include:

[0128] Acquire multiple training user information and corresponding multiple annotated user motion connectivity graphs; one training user information corresponds to multiple annotated user motion connectivity graphs;

[0129] The training user information is input into a user prediction network to obtain a predicted user motion connectivity graph.

[0130] The training user information is one-dimensional data, the user prediction network is a convolutional neural network that performs deconvolution, and the convolution kernel corresponding to the user prediction network is 2*2*2.

[0131] The loss is calculated by comparing the predicted user motion connectivity graph with the labeled user motion connectivity graph to obtain a trained user prediction network.

[0132] Among them, the cross entropy loss function is used to calculate the loss.

[0133] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0134] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0135] The various component embodiments of the present invention may be implemented in hardware, as software modules running on one or more processors, or as a combination thereof. Those skilled in the art will appreciate that, in practice, a microprocessor or digital signal processor (DSP) may be used to implement some or all of the functionality of some or all of the components of the apparatus according to the embodiments of the present invention. The present invention may also be implemented as an apparatus or device program (e.g., a computer program or computer program product) for performing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium or in the form of one or more signals. Such signals may be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A multi-dimensional exercise prescription system based on cloud computing, characterized by: Including monitoring module and processor: A monitoring module is configured to obtain multi-dimensional user information, exercise body information, and exercise mode categories; the user information represents body information that does not change with exercise; the exercise body information includes a plurality of post-exercise body sign values ​​and corresponding pre-exercise body sign values; the post-exercise body sign values ​​represent the values ​​of the body sign before exercise; and the post-exercise body sign values ​​represent the values ​​of the body sign after exercise; The processor is used to process the following method: Based on the user information and the exercise mode category, finding a sequence that matches the user's exercise mode category to obtain a exercise connectivity graph; Based on the motion connectivity graph, a basic motion mode set is obtained by traversing the graph; the basic motion mode set includes a plurality of motion mode routes; the motion mode routes represent the order of motion mode categories that meet the user information; Using an exercise pattern prediction network, based on the moving human body information, the basic exercise pattern set, and the exercise pattern categories, the changes in the user's physical signs during exercise are determined to obtain an exercise prescription; the exercise prescription represents a ranking of the exercise pattern categories suitable for the user's physical signs; The exercise mode prediction network determines changes in the user's physical signs during exercise based on the moving human body information, the basic exercise mode set, and the exercise mode category, and obtains an exercise prescription, including: Acquire a blank image; the blank image is a binary image; The horizontal axis of the blank image represents multiple categories of movement patterns; the vertical axis of the blank image represents a value that can reflect the change value of all movement signs; Construct a coordinate axis based on the blank image; the lower left corner of the blank image is used as the origin, the length of the blank image is used as the horizontal coordinate, and the width of the blank image is used as the vertical coordinate; Subtract the value of the body sign before exercise from the value of the body sign after exercise to obtain the value of the change of the body sign; The motion sign change values ​​corresponding to the multiple motion mode categories are punctuated in a blank image to obtain a two-dimensional motion parameter point map; multiple post-exercise human body sign values ​​and the corresponding pre-exercise human body sign values ​​are correspondingly obtained to obtain multiple two-dimensional motion parameter point maps; Superimposing multiple two-dimensional motion parameter point maps to construct a three-dimensional image and obtain a motion parameter point map; The exercise parameter dot graph represents the changes in the physical signs of the human body before and after exercise corresponding to the exercise mode category; Based on the motion parameter point graph, a motion mode prediction network is used to determine the relationship between motion mode categories and obtain multiple motion impact orders; Find the movement pattern route corresponding to the movement impact sequence in the basic movement pattern set and obtain the movement prescription; The motion mode prediction network includes a first motion convolutional network, a second motion convolutional network and a third motion network; The method of determining the relationship between the motion mode categories based on the motion parameter point graph and obtaining the motion impact sequence through a motion mode prediction network includes: Inputting the motion parameter point map into a first motion convolutional network to determine the user's motion state and obtain a first feature map; The number of the exercise mode categories is used as the number of exercise mode categories; The segmentation length is obtained by dividing the horizontal coordinate corresponding to the lower right corner of the motion parameter point graph by the number of motion mode categories; On the horizontal axis, the point with the distance to the dividing length is used as the dividing point; Segmenting the motion parameter point map according to the segmentation points to obtain a plurality of first motion parameter segmentation maps; Keeping the lengths of the plurality of first motion parameter segmentation maps unchanged, the plurality of first motion parameter segmentation maps are superimposed in width to obtain a superimposed image; Inputting the superimposed image into a second motion convolutional network to obtain a second feature map; The first feature map and the second feature map are input into a third motion network to obtain a motion influence sequence.

2. The multi-dimensional exercise prescription system based on cloud computing according to claim 1, characterized in that: The convolution kernel of the first motion convolution network is a three-dimensional convolution kernel of 2*2*2, and the convolution kernel of the second motion convolution network is a three-dimensional convolution kernel with a width equal to the length of the motion parameter point map, a height of 2, and a depth equal to the number of pages of the motion parameter point map.

3. The multi-dimensional exercise prescription system based on cloud computing according to claim 1, characterized in that: The step of finding a sequence of exercise categories that matches the user's exercise category based on the user information and the exercise category to obtain an exercise connectivity graph includes: Based on the user information, a plurality of user motion connectivity graphs are obtained through a user prediction network; the user motion connectivity graphs have motion mode categories as vertices; Based on multiple user motion connectivity graphs, similarities are identified and a motion connectivity graph is obtained.

4. The multi-dimensional exercise prescription system based on cloud computing according to claim 3, characterized in that: The method of determining similarities based on multiple user motion connectivity graphs to obtain a motion connectivity graph includes: Input the two user motion connectivity graphs into the similarity discrimination network to judge the similarity of the two user motion connectivity graphs and obtain the motion similarity value; p user motion connectivity graphs corresponding to (p-1)*p / 2 motion similarity values ​​are obtained; Obtain the motion similarity value that is smaller than other motion similarity values ​​among the (p-1)*p / 2 motion similarity values ​​as the first similarity cluster center; Obtaining a similarity gap value; the similarity gap value represents a fixed value from the first similarity cluster center; The sum of the first similar cluster center and the similarity gap value is used as the gap threshold; Constructing a motion similarity set from motion similarity values ​​smaller than a gap threshold among the multiple motion similarity values; The user motion connectivity graph corresponding to the motion similarity set is used as a similar motion connectivity graph; A similar motion connectivity graph is randomly obtained from multiple similar motion connectivity graphs as a motion connectivity graph.

5. The multi-dimensional exercise prescription system based on cloud computing according to claim 1, characterized in that: The basic motion mode set is obtained by traversing the motion connectivity graph, including: Obtaining m initial motion values; wherein the initial motion values ​​represent m motion mode categories in the motion connectivity graph; According to the initial movement value, the movement order of the movement mode category is traversed to obtain n movement mode routes; m initial movement values ​​correspond to m*n movement mode routes; The m*n movement mode routes constitute a basic movement mode set.

6. The multi-dimensional exercise prescription system based on cloud computing according to claim 1, characterized in that: Methods for training user prediction networks include: Acquire multiple training user information and corresponding multiple annotated user motion connectivity graphs; one training user information corresponds to multiple annotated user motion connectivity graphs; Inputting the training user information into a user prediction network to obtain a predicted user motion connectivity graph; The loss is calculated by comparing the predicted user motion connectivity graph with the labeled user motion connectivity graph to obtain a trained user prediction network.

Images