Multi-dimensional motion fatigue recognition method based on wearable sensing multi-source information fusion
By integrating multi-source wearable sensor information through a multi-task learning network, the problem of multi-dimensional identification in sports fatigue detection in existing technologies has been solved, and comprehensive, accurate quantitative perception and real-time identification of sports fatigue state has been achieved.
Patent Information
- Application Number
- CN202510869694.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-26
- Publication Date
- 2025-11-21
AI Technical Summary
Existing methods for detecting exercise fatigue mainly rely on subjective questionnaires, biochemical indicators, or single types of wearable sensors. These methods are characterized by strong subjectivity, poor real-time performance, difficulty in capturing dynamic differences between individuals, and limitations in identifying multi-dimensional fatigue states. They cannot achieve accurate quantitative perception of acute/chronic, local/global, or physical/psychological fatigue.
A multi-source information fusion method based on multi-task learning networks is adopted. Through a shared layer, a task relationship modeling module and a specific task layer, a motion fatigue recognition model is constructed. By utilizing multi-source heterogeneous wearable sensor information and combining residual and gating mechanisms, a precise quantitative perception of multi-dimensional fatigue states can be achieved.
It enables a comprehensive assessment of exercise fatigue, improves the model's efficiency, performance, and generalization ability, and can simultaneously identify multi-dimensional fatigue states, thereby enhancing the accuracy and real-time performance of fatigue state perception.
Smart Images
Figure CN120995039A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computer application, more particularly, to a multi-dimensional exercise fatigue recognition method based on wearable sensor multi-source information fusion. BACKGROUND
[0002] Physical exercise is beneficial to maintaining the health status of the body. However, if scientific guidance is lacking during exercise, it is easy to cause exercise fatigue, and further cause muscle damage, cardiovascular accidents and even long-term health problems. Especially in high-intensity exercise scenarios such as competitive sports and rehabilitation medicine, timely identification and intervention of exercise fatigue state is particularly critical.
[0003] Exercise fatigue is a self-protection mechanism of the body, and different fatigue states have different recovery needs. Accurate identification of exercise fatigue state helps to develop targeted recovery strategies, thereby improving exercise performance and avoiding injury. However, exercise fatigue has the characteristics of multi-dimensionality and complexity. From the time dimension, exercise fatigue includes acute fatigue and chronic fatigue; from the spatial dimension, exercise fatigue is divided into local muscle fatigue and systemic fatigue; according to the mechanism causing fatigue, it can be divided into central fatigue and peripheral fatigue. Therefore, how to realize accurate quantitative perception of multi-dimensional and complex fatigue state is a key scientific problem to be solved.
[0004] Exercise fatigue detection aims to monitor physiological and exercise indicators to identify the fatigue state of individuals in a timely manner to prevent overtraining or exercise injury. Traditional exercise fatigue detection methods mainly rely on subjective questionnaires, biochemical indicators and standardized exercise tests. These methods can effectively reflect the fatigue condition after exercise in a laboratory environment, but there is a certain subjective bias, or the detection cost is high, the real-time performance is poor and it is difficult to capture the dynamic differences between individuals.
[0005] With the development of wearable technology, researchers began to use physiological signals as indicators of exercise fatigue detection, and used a single type of wearable sensor to monitor exercise fatigue in real time. For example, Beijing Sports University proposed an intelligent fatigue detection method based on electrocardiogram (ECG), which provides a high-precision, low-complexity fatigue detection scheme for wearable devices through hierarchical multi-scale modeling and adaptive feature fusion. The research team of West China Hospital of Sichuan University used the probability density function (PDF) shape feature of surface electromyography (sEMG) signal to detect muscle fatigue induced by exercise in community elderly people, and proposed a fatigue detection index with low computational complexity and high sensitivity-Temporal-Mean-Kurtosis (TMK), which provides a feasible method for real-time muscle fatigue monitoring. Since a single type of sensor can only reflect information in one dimension, it has limitations in dealing with multi-dimensional fatigue status in complex exercise scenarios.
[0006] Through analysis, the prior art mainly has the following defects:
[0007] 1) Traditional fatigue evaluation methods such as subjective feeling scale (such as Borg scale), blood lactic acid concentration detection, heart rate variability analysis, etc. are effective to some extent, but have strong subjectivity, complex operation, poor real-time performance and difficulty in capturing dynamic differences between individuals.
[0008] 2) The existing wearable-based fatigue monitoring method only uses a single type of sensor, which can only reflect information in one dimension, and has limitations in dealing with multi-dimensional fatigue status in complex exercise scenarios.
[0009] 3) The existing method generally only detects instantaneous and overall fatigue, and cannot realize the analysis and identification of multi-dimensional fatigue states such as acute / chronic fatigue, local / overall fatigue, physical / psychological fatigue, etc. SUMMARY
[0010] The purpose of the present application is to overcome the above-mentioned defects of the prior art, and to provide a multi-dimensional exercise fatigue recognition method based on wearable sensor multi-source information fusion. The method comprises the following steps:
[0011] Collecting multi-source information using different types of sensors;
[0012] Inputting the multi-source information into a trained exercise fatigue recognition model to obtain exercise fatigue recognition results;
[0013] The motion fatigue recognition model comprises a shared layer, a task relationship modeling module and a specific task layer, the shared layer is used for fusion representation of the multi-source information, the task relationship modeling module comprises a task expert network, a shared expert network and an auxiliary task network, the task expert network is used for capturing task-specific details and patterns, the shared expert network is used for capturing common knowledge between different tasks, and the auxiliary task network is used for identifying motion intensity and motion type, and the specific task layer is used for jointing output features of the task expert network and the shared expert network.
[0014] Compared with the prior art, the wearable sensor multi-source information fusion-based multi-dimensional motion fatigue recognition method has the advantages that on the basis of multi-source sensor information fusion representation learning, the multi-task learning network with knowledge sharing and mechanism constraint is established, the fatigue state is accurately quantified from multiple dimensions, and the all-around evaluation of the motion fatigue state is realized.
[0015] Other features and advantages of the present application will become apparent from the following detailed description of illustrative embodiments thereof, which proceeds with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS
[0016] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments of the application and, together with the description, serve to explain the principles of the application.
[0017] Figure 1 is a flowchart of a wearable sensor multi-source information fusion-based multi-dimensional motion fatigue recognition method according to an embodiment of the present application;
[0018] Figure 2 is a process schematic diagram of a wearable sensor multi-source information fusion-based multi-dimensional motion fatigue recognition method according to an embodiment of the present application. DETAILED DESCRIPTION
[0019] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. Note that the relative arrangement, numerical expressions and numerical values of components and steps set forth in these embodiments are not limiting to the scope of the present application unless otherwise specifically stated.
[0020] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting to the scope of the application or its applications or uses.
[0021] Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the specification, where appropriate.
[0022] In all of the examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Thus, other examples of the exemplary embodiments can have different values.
[0023] It should be noted that like reference numerals and letters refer to like items throughout the attached drawings, and once an item is defined in one drawing, it is not necessary to discuss it further in subsequent drawings.
[0024] The exercise fatigue state has multidimensionality and complexity, and the existing exercise fatigue evaluation technology generally only evaluates a single dimension, and cannot take into account multiple dimensions of exercise fatigue. The present application is based on the similarity of multiple dimensions of the fatigue state task, and establishes a multi-task network-based multi-dimensional quantification perception method of the exercise fatigue state. On the basis of multi-source heterogeneous wearable sensing information fusion representation, an adaptive task-to-task fusion network is established for the prediction task of the multi-dimensional exercise fatigue state correlation index, and efficient perception of the multi-dimensional exercise fatigue state in the human body movement process is realized. In view of the problem that the correlation between each sub-task is difficult to effectively model to realize knowledge sharing, the present application introduces a specific task expert network to learn the exclusive knowledge of a specific task, and further learns the shared knowledge between tasks from the exclusive knowledge, effectively establishing the correlation between tasks; in view of the challenge that specific task exclusive knowledge learning and inter-task shared knowledge learning are difficult to balance, the present application introduces a shared expert network and a gating network, and realizes task-to-task interactive fusion by using a residual mechanism and a gating mechanism. In addition, by introducing an auxiliary task network for identifying exercise dose, an exercise dose-based fatigue state quantification constraint mechanism is established at the specific task level, realizing exercise dose perception and improving the accuracy of fatigue state perception.
[0025] Specifically, as shown in Figure 1 and Figure 2 The provided multi-dimensional exercise fatigue recognition method based on wearable sensing multi-source information fusion includes the following steps:
[0026] Step S110, an exercise fatigue recognition model for perceiving a multi-dimensional fatigue state is constructed, and the model includes a shared layer, a task relationship modeling module, and a specific task layer.
[0027] In traditional single-task learning methods, each task usually needs to build an independent network model, and the tasks are independent of each other, and the correlation information between tasks cannot be effectively utilized. Such design not only leads to model parameter redundancy and waste of computing resources, but also cannot fully utilize the correlation between tasks, resulting in low efficiency of the model in processing multi-task scenarios, and it is also difficult to capture the upstream and downstream dependencies between tasks, which limits the generalization ability and performance of the model. In contrast, the multi-task learning method adopted by the present application can simultaneously process multiple related tasks by sharing feature extractors and model parameters, thereby fully utilizing the correlation between tasks. This method can obtain the results of multiple tasks at one time, not only improving the efficiency and performance of the model, but also enhancing the flexibility and generalization ability of the model.
[0028] On the basis of multi-source heterogeneous wearable sensing information fusion learning, the present application constructs a motion fatigue recognition model for multi-dimensional complex fatigue state prediction task. The model is an adaptive task-to-task fusion network (Adaptive Task-to-Task Fusion Network, AdaTT), which can realize efficient perception of multi-dimensional fatigue state in human motion process. The prediction task includes time dimension fatigue state prediction, space dimension fatigue state prediction, physiological system dimension fatigue state prediction, and calculation of motion dose, etc. Time dimension fatigue state includes acute fatigue and chronic fatigue. The space dimension fatigue state is divided into local muscle fatigue (such as waist fatigue, thigh fatigue, calf fatigue, upper arm fatigue, etc.) and overall fatigue. The physiological system dimension fatigue state includes mental fatigue, physical fatigue, etc. The calculation of motion dose takes the energy consumption level in the motion process as the reference, including real-time energy metabolism level and total energy consumption.
[0029] In view of the problem that the relationship between each sub-task is difficult to effectively model and knowledge sharing cannot be realized, in an embodiment of the present application, a task-specific expert network (or simply referred to as a task expert network) is introduced to learn specific knowledge of a specific task and shared knowledge between tasks. In view of the challenge of balancing specific task-specific knowledge learning and shared knowledge learning between tasks, by introducing a task-specific expert network, a shared expert network and a gating network, the interaction and fusion between tasks are explicitly modeled by using a residual mechanism and a gating mechanism. These network units use different experts and fusion strategies in each fusion module, and then adaptively learn shared knowledge and task-specific knowledge.
[0030] In combination with Figure 2As shown, the exercise fatigue recognition model as a whole includes a shared layer of early multi-source information fusion representation, a task relationship modeling module in the middle stage, and a specific task layer in the later stage. The task relationship modeling module in the middle stage includes specific task expert networks unique to each task, shared expert networks shared by all tasks, and auxiliary task networks. The specific task layer in the later stage combines the output features of each specific task expert network and shared expert network and performs forward reasoning.
[0031] The shared expert network (Shared Experts Network) is used to extract common features between multi-dimensional fatigue and exercise dose perception tasks, and the goal is to capture shared information between multiple tasks, thereby reducing model parameter redundancy and improving generalization ability. The shared expert network is shared by all tasks and can capture common knowledge between different tasks, thereby providing a basic feature representation for each task. The input of the shared expert network is the common features of all tasks, and the output is a feature representation shared by all experts. These shared features are dynamically weighted by the gating network of all tasks to adapt to the needs of different tasks. Increasing the number of shared experts can enhance the model's ability to model shared knowledge of tasks. For example, the output of the sth shared expert in the th layer can be represented as:
[0032]
[0033] where X l-1 is the output feature of the (l-1)th layer, and the initial feature X 0 may all come from the initial shared layer. Here, l refers to the th layer in the shared expert network.
[0034] The task-specific expert network (Task-specific Experts Network) is a set of expert networks designed independently for each task, used to extract task-specific features. The task-specific expert network focuses on handling the unique needs of each task and can capture task-specific details and patterns. Through the task-specific expert network, the model can better adapt to the specific needs of each task, thereby improving the performance of the model in multi-task learning. The input of the task-specific expert network combines the features output by the previous layer and the features obtained from the following gating module, and the output is a task-specific feature representation. The task-specific expert network determines the model's ability to model task specificity. Increasing the number of task-specific experts can improve the model's ability to express complex task features. For example, each group of expert networks contains e experts, and the internal experts will extract or learn features from external experts as needed (gating weights) through the gating module. The external experts include s shared experts in the shared expert network and experts in other task-specific expert networks. Since the interaction lines between expert networks are not convenient to show, they can only be simplified as connected to each other between adjacent ones. The output of the e th task-specific expert in the th layer of task e can be represented as:
[0035]
[0036] where X l-1 is the output feature of the l-1th layer, the initial feature X 0 are all from the initial shared layer. Here, l refers to the lth layer of the task expert network.
[0037] The gating network is a dynamic weight distribution mechanism in the model, responsible for distributing the weights of shared experts and task experts. Its core idea is to dynamically adjust the contribution of shared experts and task experts to the final output through task-related gating weights. This dynamic routing mechanism enables the model to select the most suitable experts for processing according to the different characteristics of the input, thereby improving the adaptability and generalization ability of the model. The input of the gating network is the original feature of the task, and the output is the weight of each expert. The design of the gating network enables the model to adaptively balance the commonality and characteristics of the task, thereby achieving efficient feature sharing among different dimensions of fatigue perception tasks. The gating weight of task e to all experts can be represented as:
[0038]
[0039] where, represents the multi-layer perceptron module that processes the previous layer feature X l-1 at the lth layer and tth channel, B represents the number of task samples, and E taSk represents the output of the task expert network, and E shared represents the output of the shared expert network.
[0040] The fusion module is responsible for dynamically fusing the outputs of shared experts and task experts into the final output of the current level. Its core idea is to combine the outputs of shared experts and task experts through the weights of the gating network, thereby generating task-specific feature representations. The design of the fusion module enables the motion fatigue perception model to adaptively balance the commonality and characteristics of the task, enabling the model to better handle complex relationships in multi-task learning. Through the fusion module, the model can dynamically adjust the way of feature fusion, thereby improving the performance of the model in multi-task learning. For example, the fusion module combines the outputs of task experts and shared experts through gating weights, represented as:
[0041]
[0042] where, represents the fusion output of the task expert network and the shared expert network of the lth layer and the tth task, represents the weight of the e th task expert network of the lth layer and the tth task, output of the e-th task-specific network of the I-th task t, weight of the S-th shared network of the I-th task t, output of the S-th shared network of the I-th task t, S represents the S-th shared network, E Shared represents the number of shared networks, E task represents the number of task-specific networks, t is the task index, and e is the task-specific network index.
[0043] Residual Connections are mainly applied to the output of the task-specific private network, which enhances the modeling ability of the model to the task specificity by weighting the output of the task-specific private expert.
[0044] Finally, the output of the task-specific private expert is added to the fusion output through the residual connection, for example, represented as:
[0045]
[0046] Formula (5) weights and sums E task task-specific private experts, is the residual weight parameter of the e-th private expert of task t.
[0047] Step S120, based on the set multi-task joint loss function, the motion fatigue recognition model is trained.
[0048] For the constructed motion fatigue recognition model, the constraint optimization mechanism based on the auxiliary task network is jointly optimized. For example, the auxiliary task network is introduced to perform real-time perception calculation on the motion type and motion intensity, the prior knowledge of the motion fatigue induction mechanism is fused, and the dynamic constraint term of the core task is constructed. The prediction result of the auxiliary task can be used as implicit prior knowledge to indirectly guide the learning of the core task. This design encodes the dependency relationship of the core task to the auxiliary task into the loss function through mathematical coupling, thereby realizing knowledge transfer and collaborative optimization. Finally, the multi-task joint loss of the model is as formula (6), which is represented by the weighted loss of the core task and the weighted loss of the auxiliary task. The weighted loss of the auxiliary task can be represented as the dynamic constraint term of the core task.
[0049]
[0050] wherein, L total represents the total loss, L i represents the loss of the i-th core task, T Core represents the number of core tasks, L j represents the loss of the j-th auxiliary task, T auxiliary represents the number of auxiliary tasks. ω jLj represents the loss of the jth auxiliary task j ωi represents the weight of the ith core task i Lj represents the loss of the jth auxiliary task i ωi represents the weight of the ith core task. The auxiliary task refers to a task of identifying the intensity or type of movement. The core task refers to a task of identifying the movement fatigue relative to the auxiliary task.
[0051] In step S130, for the real-time collected multi-source information, the trained movement fatigue identification model is used to realize multi-dimensional movement fatigue identification.
[0052] After the training of the movement fatigue identification model is completed, the optimized model parameters such as weights, biases, etc. can be obtained, and then used for actual movement fatigue identification. For example, the multi-source data collected in real time by the wearable device is input into the trained movement fatigue identification model, and multi-dimensional fatigue identification results are obtained, including acute or chronic fatigue in the time dimension, local muscle fatigue or overall fatigue in the space dimension, mental fatigue or physical fatigue in the physiological dimension, etc.
[0053] In summary, in view of the complexity and subjectivity of the fatigue state, the present application innovatively proposes a multi-dimensional movement fatigue state quantitative perception method based on a multi-task learning network and a movement dose constraint mechanism. By utilizing the correlation between the fatigue state identification tasks, an adaptive task-to-task fusion network is established to realize efficient perception of multi-dimensional fatigue states in the human movement process; further, to improve the accuracy of fatigue quantitative evaluation, an auxiliary network of movement dose identification is introduced to identify the intensity and type of movement, and a dynamic constraint mechanism of fatigue state quantitative evaluation based on movement dose is established to improve the dimension and precision of fatigue state perception.
[0054] The present application can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon, the computer readable program instructions being used to cause a processor to implement various aspects of the present application.
[0055] A computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
[0056] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.
[0057] Computer readable program instructions for carrying out operations of the present application can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate array (FPGA), or programmable logic array (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present application.
[0058] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.
[0059] These computer readable program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions can also be stored in a computer readable storage medium that can include random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or nonvolatile memory, or a suitable combination of the different types of computer readable storage media. The computer readable program instructions can also be downloaded to a computer, other programmable data processing apparatus, or other device from a computer readable storage medium or to an external computer or external storage device via a data signal that can be transmitted for example via a wired medium or a wireless medium such as the Internet or wireless media.
[0060] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.
[0061] The flow diagrams and the block diagrams in the drawings are presented to illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flow diagrams and the block diagrams can represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logic functions. In some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow diagrams, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and
[0062] Embodiments of the application have been described above. The descriptions are intended to be illustrative, and not restrictive, of possible embodiments of the application. Many modifications and variations of the described embodiments are possible, given the benefit of the present disclosure, without departing from the scope and spirit of the described embodiments. The scope of the application is defined by the appended claims.
Claims
1. A multi-dimensional motion fatigue recognition method based on wearable sensor multi-source information fusion, comprising the following steps: collecting multi-source information by using different types of sensors; inputting the multi-source information into a trained motion fatigue recognition model to obtain a motion fatigue recognition result; wherein the motion fatigue recognition model comprises a shared layer, a task relationship modeling module and a specific task layer, the shared layer is used for fusion representation of the multi-source information; the task relationship modeling module comprises a task-specific expert network shared by each task, a shared expert network shared by all tasks and an auxiliary task network, the task-specific expert network is used to capture task-specific details and patterns, the shared expert network is used to capture common knowledge between different tasks, and the auxiliary task network is used to identify motion intensity and motion type; the specific task layer is used to combine the output features of the task-specific expert network and the shared expert network.
2. The method of claim 1, wherein, For the shared expert network, the output of the s-th shared expert of the l-th layer is represented as: wherein X l-1 is the output feature of the l-1th layer, and SharedExpert represents the shared expert network.
3. The method of claim 1, wherein, For the task specialist network, the output of the e-th task specialist of the l-th layer task t is represented as: where X l-1 is the output feature of the l-1th layer, e denotes the number of internal experts, and Expert denotes a task expert network.
4. The method of claim 1, wherein, distributing the weights of the shared experts of the shared expert network and the task experts of the task expert network using a gating network, the gating weights of task t for all experts is represented as: wherein, represents a multi-layer perceptron that processes the previous layer feature X l-1 on the l-th layer, t-th channel.
5. The method of claim 1, wherein, For the task-specific expert network and the shared expert network, the fusion output is obtained according to the following formula: wherein, denotes the fusion output of the task-specific network and the shared network for the t-th task in the l-th layer, denotes the weight of the e-th task-specific network for the t-th task in the l-th layer, denotes the output of the e-th task-specific network for the t-th task in the l-th layer, denotes the weight of the S-th shared network for the t-th task in the l-th layer, denotes the output of the S-th shared network for the t-th task in the l-th layer, E shared denotes the number of shared networks, E task denotes the number of task-specific networks.
6. The method of claim 5, wherein, the output of the task-specific expert network is added to the fusion output through residual connection, represented as: wherein, is a residual weight parameter of the e-th task specialist network of task t.
7. The method of claim 1, wherein, the overall loss function of training the motion fatigue recognition model is set as: wherein L total represents the overall loss value, T core represents the number of core tasks, L i represents the loss of the i-th core task, T auxiliary represents the number of auxiliary tasks, L j represents the loss of the j-th auxiliary task, ω j represents the weight of the loss L j of the j-th auxiliary task, ω i represents the weight of the loss L i of the i-th core task, the auxiliary tasks being used for recognizing the intensity and the type of movement.
8. The method of claim 1, wherein, the multi-source information is collected by using different types of sensors arranged on a wearable device, including two or more of electrocardiogram signal, pulse wave signal, respiration signal, surface electromyogram signal, acceleration signal, angular velocity signal and plantar pressure signal.
9. A computer readable storage medium having stored thereon a computer program, wherein, The computer program is executed by a processor to implement the steps of the method according to any one of claims 1 to 8.
10. A computer device comprising a memory and a processor, having stored on the memory a computer program capable of running on the processor, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.