Generative adversarial network-based sports injury risk simulation and intervention method

Abstract

The invention discloses an exercise injury risk simulation and intervention method based on a generative adversarial network. According to the method, the physical sign state and the motion state of the body of the user and the influence degree of the sport injury are represented through a hierarchical structure and are classified, and the characteristics with high influence degree are more considered when the sport injury condition is predicted, so that a network capable of extracting the characteristics with different degrees and a multi-head attention mechanism are established. Due to the fact that the network is large, a teacher network and a student network are adopted, and soft label training is used for replacing the light-weight student network to serve as a generator; training a generator through a discriminator; the technical effect that the sport injury risk of each part of the human body at the future time point can be more accurately detected and displayed is achieved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method for simulating and intervening in sports injury risks based on generative adversarial networks. Background Technology

[0002] Sports injuries refer to damage to body tissues that occurs during sports activities or daily physical activities. They are very broad, encompassing everything from minor discomfort to serious trauma. Risks of sports injuries include tendinitis, stress fractures, and tibial periostitis. Excessive exercise in sports such as running (especially long-distance running), basketball, volleyball, badminton, tennis, and rope skipping can cause injuries such as runner's knee (patellofemoral pain syndrome), medial tibial stress syndrome (calf pain), Achilles tendinitis, plantar fasciitis, and stress fractures. Sports such as tennis, badminton, baseball / softball (throwing), swimming (especially freestyle and butterfly), fitness (bench press, shoulder press), and volleyball (spiking) place extremely high demands on the stability and flexibility of the shoulder and elbow joints, placing heavy loads on the rotator cuff and elbow tendons. Excessive exercise in these sports can cause subacromial impingement syndrome, rotator cuff tendinitis or tears, and shoulder instability.

[0003] Such risks require proactive prevention, achieving the goal of "treating disease before it occurs." "Treating disease before it occurs" is a key concept in Traditional Chinese Medicine (TCM), emphasizing prevention and taking preventative measures, reflecting the wisdom of traditional Chinese medicine. This concept aligns closely with modern health concepts, reflecting a high level of importance placed on public health. I will develop TCM culture and integrate it with modern medicine. The challenge lies in predicting and simulating the state of injury at future points in time based on users' activity patterns. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for simulating and intervening sports injury risks based on generative adversarial networks, in order to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, embodiments of the present invention provide a method for simulating and intervening in sports injury risks based on generative adversarial networks, including: Acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time. Based on the user's vital signs and motion data, a teacher network is constructed through a hierarchical structure; Build a student network using a teacher network; use the student network as a generator. Based on the user's vital signs and motion data, a predicted injury image is obtained through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss. The generator is trained using a discriminator; Based on the predicted damage image, a risk intervention strategy is generated.

[0006] Optionally, the step of constructing a teacher network based on the user's vital sign data and motion data through a hierarchical structure includes: Based on the user's vital signs data and motion data, a set of vital signs levels and a set of motion levels are obtained through a hierarchical structure. The teacher network includes a vital signs network, a motor network, a fusion neural network, and a deconvolutional neural network; By using a vital sign network, based on multiple sets of vital sign levels, we obtain vital sign feature vectors. Motion feature vectors are obtained through a motion network based on a set of multiple motion levels; The vital sign feature vector and motion feature vector are input into a fusion neural network to obtain a fusion feature vector; By using a deconvolutional neural network, based on the fused feature vector, deconvolution is performed to obtain the image of the predicted damage location.

[0007] Optionally, the step of obtaining the vital sign feature vector through the vital sign network, based on multiple sets of vital sign levels, includes: The user's vital signs data corresponding to the first index in multiple sets of vital signs are input into the first neural network to detect the features of multiple user vital signs data and their correlation, and the first feature is obtained. The user's vital signs data corresponding to the second index in multiple sets of vital signs are input into the first preparation neural network, and the size is adjusted to obtain the first preparation feature. The first feature and the first prepared feature are input into a multi-head attention mechanism and fused to obtain the first fused feature; The user's vital signs data corresponding to the third index in multiple sets of vital signs are input into the second preparation neural network to extract features and obtain the second preparation features. The first fused feature and the second prepared feature are input into a multi-head attention mechanism and fused to obtain the second fused feature. By traversing the entire set of physical characteristic levels, we obtain the physical characteristic feature vector.

[0008] Optionally, the vital signs network includes a first neural network, multiple i-th preparation neural networks, and multiple multi-head attention mechanisms; The number of the i-th prepared neural network is equal to the number of multi-head attention mechanisms; The number of layers in the i-th preparatory neural network is less than the number of layers in the i-th neural network; the i-th preparatory neural network is used for feature extraction and size adjustment; Where i is a positive integer, and i corresponds to the set of physical characteristic levels.

[0009] Optionally, the training method for the teacher network includes: The system acquires historical user vital sign data, historical motion data, labeled data, and labeled images; the labeled data includes labeled injury locations and labeled injury degrees; the labeled images represent images marked with labeled injury locations and labeled injury degrees. Based on the historical user vital sign data and historical motion data, a training fusion feature vector is obtained through a vital sign network, a motion network, and a fusion neural network. The fused feature vectors are classified to obtain the training impairment locations and corresponding training impairment severity values; Loss is calculated using labeled data, training injury locations, and corresponding training injury severity values. The training includes a vital signs network, a motor network, and a fusion neural network. The training fusion feature vector is deconvolved using a deconvolutional neural network to obtain a training damage location image. The discriminator obtains the loss value based on the training damage location image and the labeled image.

[0010] The teacher network is trained based on the stated loss value.

[0011] Optionally, obtaining the loss value through a discriminator based on the training damage location image and the labeled image includes: Based on the training damage location image, extract the second training damage location and the corresponding second training damage degree value; The first loss value is obtained by calculating the loss between the second training damage location and the corresponding second training damage degree value and the labeled data; The training damage location image and the labeled image are input into the discriminator for discrimination to obtain a discrimination value; The first loss value and the discrimination value are added together to obtain the loss value.

[0012] Optionally, the construction of the student network through the teacher network includes: The output of the student network is a predicted image of the injury location; the input of the student network consists of multiple sets of physical signs and multiple sets of motor signs; one input neuron corresponds to one set of physical signs or one set of motor signs. The training method for the student network includes: Historical user vital sign data and historical motion data are input into the trained teacher network to obtain the teacher fusion feature vector; Historical user vital sign data and historical motion data are input into the student extraction network to obtain a training fusion feature vector. The loss is calculated based on the teacher's fused feature vector and the training fused feature vector, and the soft label is used to train the student extraction network. The output of the trained student extraction network is used as the input to the deconvolutional neural network in the teacher network to construct the student network; The output of the student network is compared with the labeled image input to the discriminator to obtain the loss, and the student network is trained.

[0013] Optionally, based on the user's vital sign data and motion data, the step of obtaining a set of vital sign levels and a set of motion levels through a hierarchical structure includes: Acquire vital sign data scores and exercise scores; the vital sign data represents the score of the user's vital sign data; the exercise score represents the score of the exercise data; User vital sign data with the same range of vital sign scores are treated as the same level. Through the hierarchical structure, multiple sets of vital sign levels are obtained. The multiple sets of vital sign levels are sorted according to the sequence number of the corresponding vital sign scores from largest to smallest. The motion data corresponding to the same range of motion scores are regarded as the same level. Through the hierarchical structure, multiple sets of motion levels are obtained. The multiple sets of motion levels are sorted according to the serial number of the corresponding vital sign scores from largest to smallest. Optionally, the method for constructing the motion network is the same as that for the vital signs network.

[0014] Secondly, embodiments of the present invention provide a sports injury risk simulation and intervention system based on generative adversarial networks, comprising: The acquisition module is used to acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time. The teacher network module is used to construct a teacher network based on the user's vital sign data and motion data, through a hierarchical structure. The student network module is used to build student networks using the teacher network; it uses the student network as a generator. The generation module is used to generate a predicted injury image based on the user's vital sign data and motion data, through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss; The generator is trained using a discriminator; An intervention strategy module is used to generate a risk intervention strategy based on the predicted damage image.

[0015] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: This invention also provides a method for simulating and intervening in sports injury risks based on generative adversarial networks.

[0016] In this invention, a hierarchical structure is used to represent and classify the influence of a user's physical and movement states on sports injuries. Features with higher influence are given more consideration when predicting sports injuries, so a network capable of extracting features of different degrees and a multi-head attention mechanism are established. Because the network is large, a teacher network and a student network are used, with soft labeling training replacing the previous method with a lightweight student network as the generator. The generator is then trained using a discriminator. This achieves the technical effect of more accurately detecting and displaying the risk of sports injuries to various parts of the body at future points in time. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for simulating and intervening sports injury risk based on generative adversarial networks, provided by an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the relationship between the student network and the teacher network in a method for simulating and intervening sports injury risk based on generative adversarial networks provided in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings.

[0020] Example 1 like Figure 1 As shown, this embodiment of the invention provides a method for simulating and intervening in sports injury risk based on generative adversarial networks, the method comprising: S101: Acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time; The user vital signs data refers to the measured vital signs data of the user. In this embodiment, the user vital signs data includes heart rate, heart rate variability, subjective fatigue sensation, sleep quality, and resting heart rate. The user vital signs data represent the body's physiological and psychological responses to external loads.

[0021] In this embodiment, the types of sports include football, basketball, running, volleyball, badminton, tennis, rope skipping, and swimming. Football and basketball can cause ligament injuries such as sprains (especially ankles and knees), muscle strains, fractures, dislocations, concussions, dental damage, and knee strain (such as patellar tendinitis and meniscus wear) and ankle instability due to repeated impacts and sudden stops and changes of direction. The exercise time refers to the duration of continuous exercise.

[0022] S102: Based on the user's vital signs and motion data, a teacher network is constructed through a hierarchical structure.

[0023] S103: Construct a student network through the teacher network; use the student network as a generator.

[0024] The structures of the teacher network and the student network are as follows: Figure 2 As shown.

[0025] This involves training a lightweight student network to learn the knowledge of a complex, high-performance teacher network.

[0026] S104: Based on the user's vital signs data and motion data, a predicted injury image is obtained through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss.

[0027] The predicted damage image represents the state of motion-related damage that will be detected at a future time point.

[0028] The generator is trained using a discriminator; S105: Generate a risk intervention strategy based on the predicted damage image.

[0029] In this embodiment, a damaged image is acquired; the damaged image represents an image that marks the location of damage to the user's body when no movement is performed. In this embodiment, a human body image is drawn from the damaged image, and various regions of the human body are classified, such as the knee position, the lower leg position, and the shoulder position.

[0030] Among them, it is possible to simulate the risk of sports injuries by detecting predicted injury images at multiple time points.

[0031] Among them, the risk intervention strategy is generated by finding the state of sports injury at future time points based on the predicted injury image, adjusting the corresponding sports type or reducing the sports time, and can also be adjusted manually.

[0032] Optionally, the step of constructing a teacher network based on the user's vital sign data and motion data through a hierarchical structure includes: Based on the user's vital signs data and motion data, a set of vital signs levels and a set of motion levels are obtained through a hierarchical structure.

[0033] The teacher network includes a vital signs network, a motor network, a fusion neural network, and a deconvolutional neural network.

[0034] By using a vital sign network, based on multiple sets of vital sign levels, we obtain vital sign feature vectors.

[0035] Motion feature vectors are obtained through a motion network based on a set of multiple motion levels.

[0036] The vital sign feature vector and motion feature vector are input into a fusion neural network to obtain a fusion feature vector.

[0037] The fused neural network is a fully connected neural network (FCN).

[0038] By using a deconvolutional neural network (DNN) based on the fused feature vector, a deconvolution is performed to obtain an image of the predicted damage location.

[0039] Optionally, the step of obtaining the vital sign feature vector through the vital sign network, based on multiple sets of vital sign levels, includes: The user's vital signs data corresponding to the first index in multiple sets of vital signs are input into the first neural network to detect the features of multiple user vital signs data and their correlation, and the first feature is obtained.

[0040] The user's vital sign data corresponding to the second index in multiple sets of vital sign levels is input into the first preparation neural network, and the size is adjusted to obtain the first preparation feature.

[0041] Using the above method, directly inputting the user's vital sign data corresponding to the second index in multiple vital sign level sets along with the first feature will result in the first feature overshadowing the information of the user's vital sign data corresponding to the second index, and a mismatch in form will occur. Therefore, the size of the first preparatory neural network is adjusted to obtain more feature information.

[0042] The first feature and the first preparatory feature are input into a multi-head attention mechanism and fused to obtain the first fused feature. Using this method, because the number of layers in the i-th preparatory neural network is less than that in the i-th neural network, the first feature contains more information than the first preparatory feature. Furthermore, the multi-head attention mechanism allows for greater attention to the feature information of the user's vital signs data corresponding to the first index during training. This indicates that more attention is paid to user vital signs data that are highly correlated with sports injuries.

[0043] The user's vital signs data corresponding to the third index in multiple sets of vital signs are input into the second preparation neural network to extract features and obtain the second preparation features. The first fused feature and the second prepared feature are input into the multi-head attention mechanism for fusion to obtain the second fused feature.

[0044] By traversing the entire set of physical characteristic levels, we obtain the physical characteristic feature vector.

[0045] Optionally, the vital signs network includes a first neural network, multiple i-th preparation neural networks, and multiple multi-head attention mechanisms.

[0046] In this embodiment, the i-th neural network and the i-th preparatory neural network are fully connected neural networks (FCN).

[0047] Where i is a positive integer, and i corresponds to the set of physical characteristic levels.

[0048] Where the maximum value of i is equal to the number of the symptom level set. In this embodiment, the symptom level set is 3, then the symptom network includes a first neural network, a first preparatory neural network, a second preparatory neural network, a third preparatory neural network, and 3 multi-head attention mechanisms.

[0049] Wherein, i is used to represent the label of the neural network.

[0050] Optionally, the training method for the teacher network includes: Acquire historical user vital sign data, historical motion data, and labeled data; the labeled data includes labeled injury location and labeled injury degree; the historical user vital sign data is user vital sign data at historical time points used for training; the historical motion data is motion data at historical time points used for training.

[0051] The marked damage location refers to the area of ​​injury, such as the knee, calf, or shoulder.

[0052] Based on the historical user vital sign data and historical motion data, a training fusion feature vector is obtained through a vital sign network, a motion network, and a fusion neural network.

[0053] The fused feature vectors are classified to obtain training categories and corresponding training impairment values.

[0054] In this embodiment, the softmax function is used for classification.

[0055] The loss is calculated using labeled data, training injury locations, and corresponding training injury severity values. The training is conducted using a vital sign network, a motor network, and a fusion neural network.

[0056] In this embodiment, the cross-entropy loss function is used to calculate the loss.

[0057] The training fusion feature vector is deconvolved using a deconvolutional neural network to obtain a training damage location image. The discriminator obtains the loss value based on the training damage location image and the labeled image.

[0058] The teacher network is trained based on the stated loss value.

[0059] Optionally, obtaining the loss value through a discriminator based on the training damage location image and the labeled image includes: Based on the training damage location image, extract the second training damage location and the corresponding second training damage degree value.

[0060] In this embodiment, a human body image is drawn in the labeled image, and various regions of the human body are classified, such as the knee position, the lower leg position, and the shoulder position. In the training injury location image, the average gray value of each location is found to determine whether a sports injury exists. If it is less than a sports loss threshold, no injury is determined; if it is greater than the sports loss threshold, sports loss is determined. In this embodiment, the sports loss threshold is 20. If a sports injury exists, the corresponding location is designated as the second training injury location, and the average gray value is divided by 256 (rounded up) to obtain the second training injury severity value.

[0061] In this case, the locations of various regions of the human body are the same in both the labeled image and the damaged image.

[0062] The first loss value is obtained by calculating the loss between the second training damage location and the corresponding second training damage degree value and the labeled data.

[0063] In this embodiment, the cross-entropy damage function is used to calculate the loss. The loss is calculated by combining the second training damage location and the labeled damage location, and the loss is calculated by combining the second training damage degree value and the labeled damage degree value. The sum of the two values ​​is used to obtain the first loss value.

[0064] The training damage location image and the labeled image are input into the discriminator for discrimination to obtain the discrimination value.

[0065] In this embodiment, the labeled image is a grayscale image. The labeled damage degree value is expressed as a percentage of the motion score. Multiplying the labeled damage degree value by 256 allows the grayscale value to represent the degree of motion damage; a larger grayscale value indicates a higher degree of motion damage at the corresponding damage location.

[0066] The first loss value and the discrimination value are added together to obtain the loss value.

[0067] Optionally, the construction of the student network through the teacher network includes: The output of the student network is a predicted image of the injury location; the input of the student network consists of multiple sets of physical signs and multiple sets of motor signs; one input neuron corresponds to one set of physical signs or one set of motor signs. The training method for the student network includes: Historical user vital sign data and historical motion data are input into the trained teacher network to obtain the teacher fusion feature vector; Historical user vital sign data and historical motion data are input into the student extraction network to obtain a training fusion feature vector. The loss is calculated based on the teacher's fused feature vector and the training fused feature vector, and the soft label is used to train the student extraction network.

[0068] In this embodiment, the cross-entropy loss function is used to calculate the loss.

[0069] The output of the trained student extraction network is used as the input to the deconvolutional neural network in the teacher network to construct the student network; The output of the student network is compared with the labeled image input to the discriminator to obtain the loss, and the student network is trained.

[0070] Optionally, based on the user's vital sign data and motion data, the step of obtaining a set of vital sign levels and a set of motion levels through a hierarchical structure includes: The system acquires vital sign scores and exercise scores; the vital sign scores represent the scores for the user's vital sign data; and the exercise scores represent the scores for the exercise data.

[0071] In this process, experts assign scores. In this embodiment, the expert scores range from 0 to 10, with higher scores indicating a greater impact on sports injuries. For example, the longer the exercise time exceeds the standard, the greater the probability of sports injury, and the higher the score.

[0072] User vital sign data corresponding to the same range of vital sign scores are treated as the same level. Through the hierarchical structure, multiple sets of vital sign levels are obtained. The multiple sets of vital sign levels are sorted according to the serial number of the corresponding vital sign scores from largest to smallest.

[0073] In this embodiment, the vital signs scores are classified into three levels according to [0,3), [3,6), and [6,10], resulting in a set of three vital signs levels.

[0074] The motion data corresponding to the same range of motion scores are regarded as the same level. Through the hierarchical structure, multiple sets of motion levels are obtained. The multiple sets of motion levels are sorted according to the serial number of the corresponding vital signs scores from largest to smallest.

[0075] In this embodiment, the motion scores are classified into three levels: [0,3), [3,6), and [6,10], resulting in three sets of motion levels.

[0076] Optionally, the method for constructing the motion network is the same as that for the vital signs network.

[0077] The method for constructing the motion network includes: The motion data corresponding to the first index in multiple motion level sets is input into the first neural network to detect the features of multiple motion data and their correlation, and the first motion feature is obtained. The motion data corresponding to the second index in multiple motion level sets is input into the first preparation neural network, and the size is adjusted to obtain the first motion preparation feature; The first motion feature and the first motion preparation feature are input into a multi-head attention mechanism and fused to obtain the first motion fusion feature; The motion data corresponding to the second index in multiple motion level sets is input into the second preparation neural network to extract features and obtain the second motion preparation features. The first motion fusion feature and the second motion preparation feature are input into a multi-head attention mechanism and fused to obtain the second motion fusion feature. By traversing the entire set of motion levels, we obtain the motion feature vectors.

[0078] Example 2: Based on the above-mentioned method for simulating and intervening sports injury risks using generative adversarial networks, this invention also provides a system for simulating and intervening sports injury risks using generative adversarial networks. The system includes an acquisition module, a teacher network module, a student network module, a generation module, and an intervention strategy module.

[0079] The acquisition module is used to acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time.

[0080] The teacher network module is used to construct a teacher network through a hierarchical structure based on the user's vital signs and motion data.

[0081] The student network module is used to build student networks using the teacher network; it uses the student network as a generator.

[0082] The generation module is used to obtain a predicted injury image based on the user's vital sign data and motion data through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss.

[0083] The generator is trained using a discriminator.

[0084] Based on the predicted damage image, a risk intervention strategy is generated.

[0085] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0086] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0087] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A method for simulating and intervening in sports injury risk based on generative adversarial networks, characterized in that, include: Acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time. Based on the user's vital signs and motion data, a teacher network is constructed through a hierarchical structure; Build a student network through a teacher network; Use the student network as a generator; Based on the user's vital signs and motion data, a predicted injury image is obtained through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss. The generator is trained using a discriminator; Based on the predicted damage image, a risk intervention strategy is generated.

2. The method for simulating and intervening in sports injury risk based on generative adversarial networks according to claim 1, characterized in that, The construction of a teacher network based on the user's vital signs and motion data, through a hierarchical structure, includes: Based on the user's vital signs data and motion data, a set of vital signs levels and a set of motion levels are obtained through a hierarchical structure. The teacher network includes a vital signs network, a motor network, a fusion neural network, and a deconvolutional neural network; By using a vital sign network, based on multiple sets of vital sign levels, we obtain vital sign feature vectors. Motion feature vectors are obtained through a motion network based on a set of multiple motion levels; The vital sign feature vector and motion feature vector are input into a fusion neural network to obtain a fusion feature vector; By using a deconvolutional neural network, based on the fused feature vector, deconvolution is performed to obtain the image of the predicted damage location.

3. The method for simulating and intervening in sports injury risk based on generative adversarial networks according to claim 2, characterized in that, The process of obtaining a vital sign feature vector through a vital sign network, based on multiple sets of vital sign levels, includes: The user's vital signs data corresponding to the first index in multiple sets of vital signs are input into the first neural network to detect the features of multiple user vital signs data and their correlation, and the first feature is obtained. The user's vital signs data corresponding to the second index in multiple sets of vital signs are input into the first preparation neural network, and the size is adjusted to obtain the first preparation feature. The first feature and the first prepared feature are input into a multi-head attention mechanism and fused to obtain the first fused feature; The user's vital signs data corresponding to the third index in multiple sets of vital signs are input into the second preparation neural network to extract features and obtain the second preparation features. The first fused feature and the second prepared feature are input into a multi-head attention mechanism and fused to obtain the second fused feature. By traversing the entire set of physical characteristic levels, we obtain the physical characteristic feature vector.

4. The method for simulating and intervening sports injury risk based on generative adversarial networks according to claim 3, characterized in that, The vital signs network includes a first neural network, multiple i-th preparation neural networks, and multiple multi-head attention mechanisms; The number of the i-th prepared neural network is equal to the number of multi-head attention mechanisms; The number of layers in the i-th preparatory neural network is less than the number of layers in the i-th neural network; the i-th preparatory neural network is used for feature extraction and size adjustment; Where i is a positive integer, and i corresponds to the set of physical characteristic levels.

5. The method for simulating and intervening sports injury risk based on generative adversarial networks according to claim 2, characterized in that, The training method for the teacher network includes: The system acquires historical user vital sign data, historical motion data, labeled data, and labeled images; the labeled data includes labeled injury locations and labeled injury degrees; the labeled images represent images marked with labeled injury locations and labeled injury degrees. Based on the historical user vital sign data and historical motion data, a training fusion feature vector is obtained through a vital sign network, a motion network, and a fusion neural network. The fused feature vectors are classified to obtain the training impairment locations and corresponding training impairment severity values; Loss is calculated using labeled data, training injury locations, and corresponding training injury severity values. The training includes a vital signs network, a motor network, and a fusion neural network. The training fusion feature vector is deconvolved using a deconvolutional neural network to obtain a training damage location image. The discriminator obtains the loss value based on the training damage location image and the labeled image; The teacher network is trained based on the stated loss value.

6. The method for simulating and intervening sports injury risk based on generative adversarial networks according to claim 5, characterized in that, The step of obtaining a loss value through a discriminator based on the trained damage location image and the labeled image includes: Based on the training damage location image, extract the second training damage location and the corresponding second training damage degree value; The first loss value is obtained by calculating the loss between the second training damage location and the corresponding second training damage degree value and the labeled data; The training damage location image and the labeled image are input into the discriminator for discrimination to obtain a discrimination value; The first loss value and the discrimination value are added together to obtain the loss value.

7. The method for simulating and intervening in sports injury risk based on generative adversarial networks according to claim 1, characterized in that, The construction of a student network through a teacher network includes: The output of the student network is a predicted image of the injury location; the input of the student network consists of multiple sets of physical signs and multiple sets of motor signs; one input neuron corresponds to one set of physical signs or one set of motor signs. The training method for the student network includes: Historical user vital sign data and historical motion data are input into the trained teacher network to obtain the teacher fusion feature vector; Historical user vital sign data and historical motion data are input into the student extraction network to obtain a training fusion feature vector. The loss is calculated based on the teacher's fused feature vector and the training fused feature vector, and the soft label is used to train the student extraction network. The output of the trained student extraction network is used as the input to the deconvolutional neural network in the teacher network to construct the student network; The output of the student network is compared with the labeled image input to the discriminator to obtain the loss, and the student network is trained.

8. The method for simulating and intervening sports injury risk based on generative adversarial networks according to claim 1, characterized in that, Based on the user's vital sign data and motion data, a hierarchical structure is used to obtain a set of vital sign hierarchies and a set of motion hierarchies, including: Acquire vital sign data scores and exercise scores; the vital sign data represents the score of the user's vital sign data; the exercise score represents the score of the exercise data; User vital sign data with the same range of vital sign scores are treated as the same level. Through the hierarchical structure, multiple sets of vital sign levels are obtained. The multiple sets of vital sign levels are sorted according to the sequence number of the corresponding vital sign scores from largest to smallest. The motion data corresponding to the same range of motion scores are regarded as the same level. Through the hierarchical structure, multiple sets of motion levels are obtained. The multiple sets of motion levels are sorted according to the serial number of the corresponding vital signs scores from largest to smallest.

9. The method for simulating and intervening in sports injury risk based on generative adversarial networks according to claim 2, characterized in that, The method for constructing the motor network is the same as that for constructing the vital signs network.

10. A sports injury risk simulation and intervention system based on generative adversarial networks, characterized in that, include: The acquisition module is used to acquire user vital sign data and exercise data; the exercise data includes exercise type and exercise time. The teacher network module is used to construct a teacher network based on the user's vital sign data and motion data, through a hierarchical structure. The student network module is used to build student networks using the teacher network; it uses the student network as a generator. The generation module is used to generate a predicted injury image based on the user's vital sign data and motion data, through the generator; the predicted injury image represents the location of multiple human bodies with predicted motion loss; The generator is trained using a discriminator; An intervention strategy module is used to generate a risk intervention strategy based on the predicted damage image.

Images