Muscle fatigue detection method based on transfer learning and mean rounding

Abstract

The invention relates to the field of muscle fatigue detection, in particular to a muscle fatigue detection method based on transfer learning and mean rounding. The method comprises the following steps: acquiring an sEMG signal, and preprocessing the signal; designing a deep learning model, performing rounding processing on the preprocessed data by using a mean rounding label method, and training the model by using the data after rounding processing to learn general features of the sEMG signal; under the condition that new user data is limited, fine adjustment is conducted on the trained model through a transfer learning method, and a final deep learning model is obtained and used for transferring knowledge learned by the model to a new task or a new user; and performing intervention on a model result obtained by transfer learning training by adopting a mean rounding label method to obtain a final muscle fatigue detection result. The method provides an efficient and practical solution for accurate detection and personalized health monitoring of muscle fatigue, and is suitable for scenes of rehabilitation training, exercise monitoring, occupational health management and the like.

Description

Technical Field

[0001] The present invention relates to the field of muscle fatigue detection, and in particular to a muscle fatigue detection method based on transfer learning and mean rounding. Background Art

[0002] With the development of medical rehabilitation technology and sports science, muscle fatigue detection has received widespread attention in the fields of health monitoring, sports training and rehabilitation medicine. Fatigue is a common physiological state. If it is not properly monitored and managed for a long time, it may lead to sports injuries, muscle loss and even more serious health problems. Fatigue detection is a method of evaluating fatigue status by monitoring physiological signals of the human muscle or nervous system (such as sEMG). Fatigue detection can be used to identify muscle fatigue caused by exercise, work or rehabilitation training, so that appropriate adjustments can be made to prevent overuse of muscles. In the process of rehabilitation training, accurate detection of fatigue is particularly important because it is related to the safety of patients and the effectiveness of rehabilitation training.

[0003] Surface electromyography (sEMG) is a non-invasive bioelectric signal that records the electrical activity generated by muscles during activity. It can reflect the intensity and state of muscle activity and is widely used in muscle function assessment and fatigue detection. However, sEMG signals are easily affected by multiple factors such as environmental noise, electrode placement, skin conditions, etc., especially the significant differences between different individuals. Traditional fatigue detection methods usually rely on fixed statistical feature extraction and classification algorithms, which are difficult to overcome the influence of individual differences, resulting in insufficient detection accuracy and robustness. At the same time, machine learning methods rely heavily on large-scale, high-quality labeled data. In actual scenarios, collecting and labeling a sufficient amount of sEMG data is usually time-consuming and labor-intensive, especially for rehabilitation training patients. Due to physical conditions and experimental environments, data collection is even more difficult. In addition, training deep learning models usually takes a long time, which is impractical in real-time monitoring scenarios. Therefore, how to reduce data collection requirements, shorten model training time, and improve cross-individual detection accuracy has become a key technical problem that needs to be solved in the field of fatigue detection.

[0004] In the prior art, the "multi-dimensional feature fusion network" requires large-scale labeled data for training, while the "near infrared spectroscopy device" introduces additional multimodal data collection requirements, which increases data costs. Therefore, the prior art has the following significant shortcomings in surface electromyography (sEMG) fatigue detection: First, the traditional method is usually based on the prediction results of a single frame signal, ignoring the timing characteristics of the signal, resulting in the detection results being easily disturbed by short-term noise and sudden abnormalities, resulting in instability and misjudgment; second, due to the significant differences in physiological characteristics and signal acquisition conditions between individuals, the traditional model shows insufficient accuracy and poor applicability in cross-individual applications; finally, the existing deep learning method has a high dependence on large-scale labeled data, a long training time, and is difficult to meet the needs of real-time detection, especially in scenes such as rehabilitation training and motion monitoring. It is impractical. Summary of the invention

[0005] In order to solve the problems of unstable detection results, significant individual differences and insufficient real-time performance in the prior art, the present invention provides a muscle fatigue detection method based on transfer learning and mean rounding. By introducing the mean rounding labeling method, it is possible to smooth the multi-frame prediction results, capture the dynamic trend of the signal, and significantly improve the stability and robustness of the detection results; through transfer learning technology, it is possible to effectively reduce data collection requirements and model training time, and significantly improve cross-individual detection accuracy and applicability; at the same time, combined with improved signal data preprocessing processes (such as extended Kalman filtering and grayscale image conversion), the detection efficiency of the system is further improved. The method mainly includes the following steps:

[0006] S1: Acquire sEMG signal and preprocess the signal;

[0007] S2: Design a deep learning model, slice the preprocessed data using the mean rounding labeling method, and train the model using the sliced ​​data to learn the general features of sEMG signals;

[0008] S3: When new user data is limited, the trained model is fine-tuned through transfer learning methods to obtain the final deep learning model, which is used to transfer the knowledge learned by the model to new users;

[0009] S4: The mean integer labeling method is used to intervene in the model results obtained by transfer learning training to obtain the final muscle fatigue detection results.

[0010] A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0011] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above method.

[0012] A computer program product comprises a computer program or instructions, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0013] The technical solution provided by the present invention has the following beneficial effects: the present invention introduces the mean rounding labeling method, performs weighted averaging and rounding of multi-frame prediction results, smoothes the classification results, improves the ability to capture time series trends, and effectively enhances the stability and robustness of detection. The present invention uses transfer learning technology to achieve efficient adaptation of the model between multiple users and different individuals, reduces the impact of individual differences on the detection results, and reduces the cost of data acquisition and model retraining. The present invention directly utilizes the deep features of sEMG signals through signal preprocessing and grayscale image conversion, combines deep learning models to achieve end-to-end feature extraction and classification, and reduces the computational complexity of feature engineering. Moreover, the present invention can achieve high-precision detection by relying only on a single sEMG signal, simplifies hardware design and data processing, and significantly improves real-time performance. The present invention provides an efficient and practical solution for accurate detection of muscle fatigue and personalized health monitoring, and is suitable for scenes such as rehabilitation training, sports monitoring, and occupational health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0015] Figure 1 is a flow chart of a muscle fatigue detection method based on transfer learning and mean rounding in an embodiment of the present invention;

[0016] Figure 2 is a schematic diagram of a mean rounding labeling method in an embodiment of the present invention;

[0017] Figure 3 Schematic diagram of the transfer learning method in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0019] Individual differences refer to differences between individuals of the same type due to differences in genetics, anatomy, physiology or environment. For example, individual differences in muscle physiological structure, skin conductivity, signal noise interference, etc. will cause sEMG to show significant characteristic differences between different individuals. This difference will affect the versatility and accuracy of the algorithm in sEMG signal processing. Therefore, individual differences are an important issue that needs to be addressed in fatigue detection.

[0020] In view of the problem of incomplete utilization of timing information in detection: sEMG signals have significant timing characteristics, but traditional fatigue detection methods usually only classify based on single-frame signals, ignoring the correlation between consecutive frames. This approach is easily affected by short-term noise or sudden signal anomalies, resulting in large fluctuations in detection results and failure to accurately capture the dynamic trend of fatigue status. The present invention designs a mean rounding labeling method, which averages and rounds the prediction results of multiple frames to generate the final classification label, greatly improving the accuracy.

[0021] Regarding the problem of data quantity and long model training time: Traditional deep learning models usually require a large amount of labeled data for training, and the training time is long, and they cannot quickly adapt to individual monitoring needs. The present invention significantly reduces data requirements and training time through transfer learning of pre-trained models, and realizes rapid adaptation and efficient personalized fatigue detection.

[0022] According to the influence of individual differences on fatigue detection accuracy: Due to the differences in muscle anatomical structure, skin resistance, and electromyographic signal characteristics between individuals, traditional fatigue detection methods show significant accuracy fluctuations in different individuals. The present invention designs a fatigue detection method that can adapt to individual differences and achieve high-precision fatigue state classification across individuals.

[0023] Example 1

[0024] Please refer to Figure 1 , Figure 1 This is a flow chart of a muscle fatigue detection method based on transfer learning and mean rounding in an embodiment of the present invention, which realizes muscle fatigue detection by preprocessing sEMG signals, designing a deep learning model, using mean rounding labeling method, transferring learning, and using mean rounding labeling method, and specifically includes:

[0025] S1: Acquire sEMG signal and preprocess the signal; the preprocessing includes using an extended Kalman filter method (EKF) to remove noise and interference in the sEMG signal, and then mapping the filtered data to a grayscale image.

[0026] S2: Design a deep learning model, slice the preprocessed data using the mean rounding labeling method, and train the model using the sliced ​​data to learn the general features of sEMG signals;

[0027] S3: When new user data is limited, the trained model is fine-tuned through transfer learning methods to obtain the final deep learning model, which is used to transfer the knowledge learned by the model to new users;

[0028] S4: The mean integer labeling method is used to intervene in the model results obtained by transfer learning training to obtain the final muscle fatigue detection results.

[0029] The present invention adopts the following method in the fatigue detection process: Figure 2 The mean rounding labeling method shown in the figure performs comprehensive processing on the multi-frame prediction results, solving the problem that the single-frame prediction is easily disturbed by short-term noise. Specifically, the continuous sEMG signal is first divided into time slices of fixed length by the sliding window method, and each time slice is sent to the final deep learning model as an independent input frame for prediction. The model output results of each frame signal are summarized, and then the mean of the predicted labels of multiple consecutive frames (10 frames are used in this embodiment) is calculated, and the final label is generated by rounding, that is, the final muscle fatigue detection result is obtained. This method not only reduces the random error of single-frame prediction by smoothing the multi-frame prediction results, but also effectively captures the temporal change trend of the signal, thereby improving the stability and robustness of the detection.

[0030] In order to solve the problem of individual differences affecting detection accuracy and the long time required for data collection and model training, the present invention adopts Figure 3 The transfer learning technology shown in . Transfer learning can transfer the knowledge learned in the trained deep learning model to new tasks or new users, thereby reducing the dependence on large-scale labeled data and accelerating the adaptation of the model in new scenarios. The specific implementation steps are as follows: First, the deep learning model (such as ResNet, DenseNet, etc.) is pre-trained using the sEMG signal data of multiple users to learn the common features of sEMG signals. Subsequently, in the case of limited new user data, the model is fine-tuned through transfer learning so that it can quickly adapt to the personalized signal characteristics of new users. This method significantly reduces the dependence on large-scale data while maintaining high detection accuracy. In addition, the introduction of transfer learning overcomes the impact of differences in physiological characteristics between individuals on the detection results, providing technical support for wide applications across individuals.

[0031] The technical solution implemented by the present invention can achieve the following effects:

[0032] 1. Make full use of timing characteristics

[0033] Most existing technologies only classify or evaluate data from a single frame or a specific time window, ignoring the temporal variation characteristics of sEMG signals. The present invention introduces a mean rounding labeling method, which smoothes the classification results by weighted averaging and rounding the prediction results of multiple frames, improves the ability to capture temporal trends, and effectively enhances the stability and robustness of detection.

[0034] 2. Cross-individual adaptability

[0035] Existing technologies mostly design detection models for fixed data sets and individuals. For example, multidimensional feature fusion networks rely on the calculation of specific muscle positions and muscle contributions, which has limited scope of application. However, the present invention uses transfer learning technology to achieve rapid fine-tuning of the model, and only requires a small amount of new user data to complete the efficient adaptation of the model between multiple users and different individuals, significantly reducing training time and data requirements, as well as the impact of individual differences on detection results, reducing the cost of data collection and model retraining, and improving feasibility in practical applications. It is suitable for a wider range of application scenarios.

[0036] 3. Efficient data utilization and real-time performance

[0037] Compared with the "multi-dimensional feature fusion network" that relies on complex time domain and frequency domain feature extraction, the present invention uses the deep features learned by the deep learning model after signal preprocessing and grayscale conversion for subsequent transfer learning, and combines the deep learning model to achieve end-to-end feature extraction and classification, reducing the computational complexity of feature engineering. In addition, unlike the "near-infrared spectroscopy-based device" that requires a complex process of multimodal data acquisition, the present invention only relies on a single sEMG signal to achieve high-precision detection, simplifies hardware design and data processing, and significantly improves real-time performance.

[0038] Example 2

[0039] A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0040] Example 3

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above method.

[0042] Example 4

[0043] A computer program product comprises a computer program or instructions, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0044] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A muscle fatigue detection method based on transfer learning and mean rounding, characterized in that: include: S1: Acquire sEMG signal and preprocess the signal; S2: Design a deep learning model, slice the preprocessed data using the mean rounding labeling method, and train the model using the sliced ​​data to learn the general features of sEMG signals; S3: When new user data is limited, the trained model is fine-tuned through transfer learning methods to obtain the final deep learning model, which is used to transfer the knowledge learned by the model to new users; S4: The mean integer labeling method is used to intervene in the model results obtained by transfer learning training to obtain the final muscle fatigue detection results.

2. A muscle fatigue detection method based on transfer learning and mean rounding as claimed in claim 1, characterized in that: In S1, the preprocessing includes using an extended Kalman filter method to remove noise and interference in the sEMG signal, and then mapping the filtered data to a grayscale image.

3. A muscle fatigue detection method based on transfer learning and mean rounding as claimed in claim 1, characterized in that: In S2, the deep learning model is ResNet, or DenseNet, or CNN, which is used to directly output muscle fatigue detection results.

4. A muscle fatigue detection method based on transfer learning and mean rounding as claimed in claim 1, characterized in that: In S2, the slicing process is to divide the preprocessed continuous sEMG signal into time slices of fixed length by a sliding window method, and each time slice is used as an independent input frame.

5. A muscle fatigue detection method based on transfer learning and mean rounding as claimed in claim 1, characterized in that: In S4, the process of intervening in the prediction results is to summarize the model output results of each frame signal, then calculate the mean of the predicted labels of multiple consecutive frames, and generate the final label by rounding, that is, to obtain the final muscle fatigue detection result.

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the muscle fatigue detection method based on transfer learning and mean rounding as described in claims 1-5.

7. A computer-readable storage medium, characterized in that: A computer program is stored, and when the program is executed by a processor, the steps of the muscle fatigue detection method based on transfer learning and mean rounding as described in claims 1-5 are implemented.

8. A computer program product, characterized in that It includes a computer program or instruction, which, when executed by a processor, implements the steps of the muscle fatigue detection method based on transfer learning and mean rounding as described in claims 1-5.

Images