Method for establishing quantifiable muscle fatigue model and wearable real-time monitoring system

Abstract

This invention relates to a method for establishing a quantifiable muscle fatigue level model and a wearable real-time monitoring system. The method includes: establishing a dataset related to muscle fatigue level; extracting a training dataset from the dataset and preprocessing the training dataset to form preprocessed data; using the preprocessed data to train the muscle fatigue level model, obtaining a training result and a post-training muscle fatigue level model each time the model is trained; determining whether the difference in muscle fatigue level is greater than a critical value; if the difference is greater than the critical value, retraining the post-training muscle fatigue level model until a new difference in muscle fatigue level is less than the critical value, at which point training stops, resulting in a completed training model.

Claims

1. A method for establishing a quantifiable model of muscle fatigue, the method comprising the following steps performed by an electronic device: establishing a dataset related to muscle fatigue, the dataset including: The system uses one or more standard blood samples to obtain components related to muscle fatigue levels. A sensor is used to sense the muscles of at least one user to obtain one or more blood samples containing components related to muscle fatigue levels. The sensor is a near-infrared spectroscopy (NIRS) sensor, which includes a three-band near-infrared light-emitting diode (LED) and a near-infrared light receiving sensor. A training dataset is extracted from the dataset, and the training dataset is preprocessed, including normalization and data augmentation, to form multiple preprocessed data samples. Using these preprocessed data to train a muscle fatigue model, the training process includes: extracting local features from the preprocessed data through complex convolutional layers to generate convolutional data; The convolutional data is pooled through a complex pooling layer to reduce computation and generate pooled data. This pooled data is then linearly combined through a fully connected layer. Each training iteration of the muscle fatigue model yields a training result and a post-training muscle fatigue model, which includes a predicted muscle fatigue level and a difference value between the predicted and actual muscle fatigue levels. The model then checks if the difference value exceeds a threshold. If it does, the post-training model is retrained to obtain a new difference value and a retrained model. Training continues until the new difference value falls below the threshold, at which point a completed training model is obtained. The muscle fatigue-related components in the standard blood sample are calculated using the modified Beer-Lambert's law (MBLL).

2. The method for establishing a quantifiable muscle fatigue level model as described in claim 1, wherein the blood components related to muscle fatigue level include oxyheme (HbO2), deoxyheme (Hb), total heme (tHb), muscle oxygen saturation (SmO2), and oxygenation index (OI).

3. The method for establishing a quantifiable muscle fatigue level model as described in claim 1, wherein the muscle fatigue level model is a fusion deep learning architecture that combines at least one selected from one-dimensional convolutional neural networks (CNN), two-dimensional convolutional neural networks (2D CNN), one-dimensional residual pooling networks, residual convolutional neural networks, self-attention mechanisms, traditional signal processors, and combinations thereof, to extract local features for learning and classifying fatigue levels.

4. The method for establishing a quantifiable muscle fatigue level model as described in claim 3, wherein the muscle fatigue level model further incorporates at least one selected from the following: gated recurrent unit (GRU), time-series transformation device, temporal convolutional neural network (TCN), hidden Markov model (HMM), state-space model (SSMs), long short-term memory network (LSTM), and combinations thereof, to integrate time-series features and local features for learning and classifying fatigue levels.

5. A wearable real-time monitoring system for quantifying muscle fatigue levels, comprising: A sensor performs one or more bands of near-infrared light sensing for at least one user to acquire one or more real-time blood component data. The sensor is a near-infrared spectroscopy (NIRS) sensor comprising a three-band near-infrared light-emitting diode (LED) and a near-infrared light receiving sensor. The real-time blood component data includes oxyheme (HbO2), deoxyheme (Hb), and total heme (tHb). A processor, connected to the sensor, receives the real-time blood component data from the sensor and stores and manages multiple muscle fatigue level tags and the preprocessed data. The processor is an embedded edge computing board, built into a neural network processing unit (NPU), a graphics processing unit (GPU), a field-programmable logic array (FPGA), a digital signal processor (DSP), and combinations thereof. At least one of the following, possessing hardware AI acceleration capabilities; a muscle fatigue level grading model, established according to the method for establishing a muscle fatigue level model as described in any one of claims 1 to 4, the muscle fatigue level grading model being connected to the processor to receive the real-time blood component data transmitted from the processor and perform calculations to generate multiple predicted muscle fatigue level labels, comparing the predicted muscle fatigue level labels with the muscle fatigue level labels stored in the processor to generate a comparison result, and transmitting the comparison result to the processor; and a user feedback module, connected to the processor, connecting to a human-computer interaction interface via wired or wireless Bluetooth communication to provide the comparison result to the user, displaying the real-time muscle fatigue level, and issuing a warning to the user when the muscle fatigue level exceeds a threshold.

6. The wearable real-time monitoring system as described in claim 5, wherein the muscle fatigue level grading model adopts a fusion deep learning architecture, combining at least one selected from one-dimensional convolutional neural networks (CNN), two-dimensional convolutional neural networks (2D CNN), one-dimensional residual pooling networks, residual convolutional neural networks, self-attention mechanisms, conventional signal processors, and combinations thereof, to extract local features for learning and classifying fatigue levels.

7. The wearable real-time monitoring system as described in claim 6, wherein the muscle fatigue level classification model further incorporates at least one selected from the following: gated recurrent unit (GRU), time-series transformation device, temporal convolutional neural network (TCN), hidden Markov model (HMM), state-space model (SSMs), long short-term memory network (LSTM), and combinations thereof, to integrate time-series features and local features for learning and classifying fatigue levels.

8. The wearable real-time monitoring system as described in claim 5, wherein the three-band near-infrared light-emitting diode (LED) emits wavelengths of 750nm, 810nm, and 870nm, respectively.

9. The wearable real-time monitoring system as described in claim 5, wherein the user feedback module includes an OLED display module and a buzzer for displaying fatigue level values ​​and emitting warning sounds; or, the user feedback module is further integrated with the human-machine interface as a functional element or a functional device.

10. The wearable real-time monitoring system as described in claim 5, wherein the muscle fatigue level tags stored and managed by the processor are obtained by means of the Berg scale score or by further grading based on the Berg scale score, and are used to assess and monitor the muscle fatigue level of the upper and lower limb muscle groups.

Images