TWI886079B - Muscle fatigue analysis device using surface electromyography signals - Google Patents

Abstract

The present invention is a muscle fatigue analysis device using surface electromyography signals, comprising an input unit and a processing unit. The processing unit receives time-domain electromyography signals through the input unit and converts them into frequency-domain integrated electromyography signals. The processing unit has a deep learning model and takes time-domain electromyography signals and frequency-domain electromyography signals as inputs. The deep learning model has a time-domain branch, a frequency-domain branch, and a fully connected layer. The time-domain branch extracts the time-domain features of the time-domain electromyography signals. The frequency-domain branch extracts the frequency-domain features of the frequency-domain electromyography signals. The fully connected layer generates an analysis result based on the time-domain features and the frequency-domain features. The analysis result is a muscle fatigue analysis result or a muscle non-fatigue analysis result. This will further improve the accuracy of the analysis results and help sports and medical professionals effectively prevent sports injuries when formulating training plans or rehabilitation plans.

Description

Muscle fatigue analysis device using surface electromyography signals

The present invention relates to a muscle fatigue analysis device, and more particularly, to a muscle fatigue analysis device using surface electromyography signals.

Muscle fatigue is an inevitable phenomenon during exercise, and if it is severe, it may even cause muscle injury. However, for athletes or manual laborers, muscle fatigue is a common occurrence, especially in order to improve muscle endurance through training or to complete work tasks, muscle fatigue caused by long-term exercise is an inevitable phenomenon.

In recent years, the trend of exercising has become more popular, and the risk of injury caused by muscle fatigue has also increased. In traditional strength training, coaches will guide trainees to perform at an intensity of 10 to 12 maximum repetitions. The maximum repetition (RM) is an intensity setting indicator during muscle training, representing the maximum weight that can be lifted according to the specified number of repetitions before muscle fatigue. Although the maximum repetition is very helpful for strength training, there is currently no instrument that can accurately assist in measurement. It is usually evaluated by the coach's judgment or the trainee's own feedback.

Since the risk of muscle damage is high when the muscles are fatigued due to high-intensity loads, how to identify muscle fatigue becomes an important issue in order to accurately assess the maximum number of repetitions.

In view of this, one of the purposes of the present invention is to provide a "muscle fatigue analysis device using surface electromyography signals", which can measure surface electromyography signals through a non-invasive measurement method and evaluate the muscle fatigue status through surface electromyography signals for users' reference.

To achieve the aforementioned purpose, the present invention "Muscle fatigue analysis device using surface electromyography signal" includes an input unit and a processing unit.

The input unit is connected to a surface electromyography signal detection device to receive a time domain electromyography signal.

The processing unit is connected to the input unit to receive the time-domain electromyographic signal and has a deep learning model. The processing unit converts the time-domain electromyographic signal into a frequency-domain electromyographic signal and inputs the time-domain electromyographic signal and the frequency-domain electromyographic signal into the deep learning model.

The deep learning model has a time domain branch, a frequency domain branch and a fully connected layer. The time domain branch receives the time domain electromyographic signal and extracts at least one time domain feature of the time domain electromyographic signal. The frequency domain branch receives the frequency domain electromyographic signal and extracts at least one frequency domain feature of the frequency domain electromyographic signal. The fully connected layer receives the at least one time domain feature and the at least one frequency domain feature, and generates an analysis result according to the at least one time domain feature and the at least one frequency domain feature. The analysis result is a muscle fatigue analysis result or a muscle non-fatigue analysis result.

Since surface electromyographic signals can be measured at any time during exercise and are measured in a non-invasive manner, they can be easily operated. In addition, the muscle fatigue analysis device using surface electromyographic signals accurately identifies the surface electromyographic signals through deep learning technology, and extracts the time domain features and the frequency domain features through the time domain branch and the frequency domain branch, respectively, thereby further improving the accuracy of the analysis results, which will help sports and medical professionals to effectively prevent sports injuries when formulating training plans or rehabilitation plans.

Referring to FIG. 1 and FIG. 2 , in one embodiment of the present invention, a muscle fatigue analysis device 10 using surface electromyography signals includes an input unit 11, a processing unit 12, and an output unit 13. The input unit 11 is connected to a surface electromyography signal detection device 20 to receive a time-domain electromyography signal S1. In this embodiment, the surface electromyography signal detection device 20 is an electromyograph, and the time-domain electromyography signal is an electromyography (EMG).

The processing unit 12 is connected to the input unit 11 to receive the time-domain electromyographic signal S1, and has a deep learning model 120. The processing unit 12 converts the time-domain electromyographic signal S1 into a frequency-domain electromyographic signal S2, and inputs the time-domain electromyographic signal S1 and the frequency-domain electromyographic signal S2 into the deep learning model 120. The deep learning model 120 has a time-domain branch 121, a frequency-domain branch 122, and a fully connected layer 123. In this embodiment, the processing unit 12 converts the time-domain electromyographic signal S1 into the frequency-domain electromyographic signal S2 by a time-frequency conversion algorithm. For example, the processing unit 12 converts the time domain electromyographic signal S1 into the frequency domain electromyographic signal S2 through Fourier Transform, but the present invention is not limited thereto.

The time domain branch 121 receives the time domain electromyographic signal S1 and extracts at least one time domain feature of the time domain electromyographic signal S1. The frequency domain branch 122 receives the frequency domain electromyographic signal S2 and extracts at least one frequency domain feature of the frequency domain electromyographic signal S2. The fully connected layer 123 receives the at least one time domain feature and the at least one frequency domain feature and generates an analysis result according to the at least one time domain feature and the at least one frequency domain feature. The analysis result is a muscle fatigue analysis result or a muscle non-fatigue analysis result.

Since the surface electromyography signal can be measured at any time during exercise and is measured in a non-invasive manner, it can be easily operated. In addition, the muscle fatigue analysis device 10 using the surface electromyography signal accurately identifies the surface electromyography signal through deep learning technology, and extracts the time domain feature and the frequency domain feature respectively through the time domain branch 121 and the frequency domain branch 122, thereby further improving the accuracy of the analysis result, which will help sports and medical professionals to effectively prevent sports injuries when formulating training plans or rehabilitation plans.

Furthermore, through the muscle fatigue analysis device 10 using surface electromyography signals, sports coaches or medical personnel can more accurately assess the muscle fatigue of trainers or rehabilitation personnel, thereby formulating more effective strategies in sports training and injury prevention, thereby improving sports performance and reducing the risk of sports injuries. Furthermore, it has been verified that the analysis results generated by the muscle fatigue analysis device 10 using surface electromyography signals, such as the muscle fatigue analysis results or the muscle non-fatigue analysis results, have an accuracy rate of up to 90.07%, which shows its potential and reliability in practical applications.

Furthermore, in this embodiment, the time domain branch extracts the at least one time domain feature of the time domain electromyographic signal through a temporal convolutional network (TCN) model. The TCN uses 32 filters and a convolution kernel with a kernel size of 3, and the dilation rate of the TCN is set to 1, 2, 4, or 8.

The temporal convolutional network model can significantly improve the speed of training and inference and has fewer parameters, which can improve the speed of training. In addition, in this embodiment, the temporal convolutional network model introduces a self-attention mechanism in the second layer, and further adjusts the at least one time domain feature extracted by the temporal convolutional network model by calculating the relationship between different positions in the sequence.

In addition, in the present embodiment, the frequency domain electromyographic signal S2 is a complex one-dimensional matrix. And the frequency domain branch 122 extracts the at least one frequency domain feature of the frequency domain electromyographic signal through a complex one-dimensional convolution (1D-Convolution) layer and a maximum pooling (Max Pooling) layer. And the convolution kernel size of the one-dimensional convolution layers increases layer by layer. For example, the one-dimensional convolution layers are 4 one-dimensional convolution layers, and the convolution kernel size of the four one-dimensional convolution layers is 5, 15, 15, 45 layer by layer, and the number of filters of the four one-dimensional convolution layers is 32, 64, 64, 96 layer by layer.

The muscle fatigue analysis device 10 using surface electromyography signals uses a larger convolution kernel in order to effectively capture the relationship across multiple frequencies in the signal spectrum, thereby improving the accuracy of the inference result.

Furthermore, an activation function of the frequency domain branch 122 is a rectified linear unit (ReLU) function. The at least one frequency domain feature of the frequency domain electromyographic signal S2 extracted by the frequency domain branch 122 is flattened, concatenated with the at least one time domain feature of the time domain electromyographic signal S1 extracted by the time domain branch 121, and then output to the fully connected layer 123.

Since the output dimension of the frequency domain branch 122 is different due to the different number of convolution kernels of the last convolution layer, the dimension of the final output will be different. For example, when the convolution kernel of the last layer of the frequency domain branch 122 is not 1, the dimension of the final output of the frequency domain branch 122 will be a two-dimensional signal. However, in order to splice with the time domain electromyographic signal S1 extracted by the time domain branch 121, the two-dimensional signal output by the frequency domain branch 122 is flattened to form a one-dimensional signal. For example, the specific method can be to concatenate the signal of the second dimension in the two-dimensional signal and place it behind the signal of the first dimension to form an extended one-dimensional signal.

In order to input the time domain features and the frequency domain features extracted by the time domain branch 121 and the frequency domain branch 122 into the fully connected layer 123 for final classification training, the frequency domain electromyographic signal S2 extracted by the frequency domain branch 122 and the time domain electromyographic signal S1 extracted by the time domain branch 121 must be spliced. For example, the specific method is similar to the flattening method, but the difference is that the at least one frequency domain feature of the flattened frequency domain electromyographic signal S2 and the at least one time domain feature of the flattened time domain electromyographic signal S1 are concatenated to form an extended one-dimensional signal, which is then used as the input of the fully connected layer 123 for learning. The reason for doing this is that the input requirement of the fully connected layer 123 can only be a one-dimensional signal feature.

In addition, the output unit 13 is connected to the processing unit 12 and is connected to an external device 30. The output unit 13 receives the analysis result and transmits the analysis result to the external device 30, and the external device 30 displays the analysis result.

For example, the external device 30 can be a mobile device, a personal computer or other electronic device with a display function, for displaying the analysis result, such as displaying a fatigue diagram or text corresponding to the muscle fatigue analysis result, or displaying a non-fatigue diagram or text corresponding to the muscle non-fatigue analysis result. In this way, sports coaches and medical professionals can confirm the exercise status of trainers or rehabilitators through the external device 30, avoid excessive exercise, and prevent sports injuries.

The above is only the preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as the preferred embodiment, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to the equivalent embodiments of equivalent changes by using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

10: Muscle fatigue analysis device using surface electromyography 11: Input unit 12: Processing unit 120: Deep learning model 121: Time domain branch 122: Frequency domain branch 123: Fully connected layer 13: Output unit 20: Surface electromyography detection device 30: External device S1: Time domain electromyography S2: Frequency domain electromyography

Figure 1: Schematic diagram of the device unit block of the present invention. Figure 2: Schematic diagram of the deep learning model architecture of the present invention.

10: Muscle fatigue analysis device using surface electromyography signals

11: Input unit

12: Processing unit

13: Output unit

20: Surface electromyography signal detection device

30: External devices

Claims (8)

A muscle fatigue analysis device using surface electromyography signals comprises: an input unit connected to a surface electromyography signal detection device to receive a time-domain electromyography signal; a processing unit connected to the input unit to receive the time-domain electromyography signal and having a deep learning model; wherein the processing unit converts the time-domain electromyography signal into a frequency-domain electromyography signal and inputs the time-domain electromyography signal and the frequency-domain electromyography signal into the deep learning model; wherein the deep learning model has a time-domain branch, a frequency-domain branch and a fully connected layer; wherein the time-domain branch receives the time-domain electromyography signal and extracts at least one time-domain feature of the time-domain electromyography signal; Wherein, the frequency domain branch receives the frequency domain electromyographic signal and extracts at least one frequency domain feature of the frequency domain electromyographic signal; Wherein, the fully connected layer receives the at least one time domain feature and the at least one frequency domain feature, and generates an analysis result according to the at least one time domain feature and the at least one frequency domain feature; Wherein, the analysis result is a muscle fatigue analysis result or a muscle non-fatigue analysis result. A muscle fatigue analysis device using surface electromyography signals as described in claim 1, wherein the time domain branch extracts at least one time domain feature of the time domain electromyography signal through a time convolution network model. A muscle fatigue analysis device using surface electromyography signals as described in claim 2, wherein the temporal convolution network uses 32 filters and a convolution kernel with a kernel size of 3, and the expansion rate of the temporal convolution network is set to 1, 2, 4, 8. A muscle fatigue analysis device using surface electromyography signals as described in claim 1, wherein the frequency domain electromyography signals are multiple one-dimensional matrices; wherein the frequency domain branch extracts the at least one frequency domain feature of the frequency domain electromyography signals through multiple one-dimensional convolution layers and a maximum pooling layer; wherein the convolution kernel size of the one-dimensional convolution layers increases layer by layer. A muscle fatigue analysis device using surface electromyography signals as described in claim 4, wherein the one-dimensional convolution layers are 4 one-dimensional convolution layers, and the convolution kernel sizes of the 4 one-dimensional convolution layers are 5, 15, 15, 45 layer by layer, and the number of filters of the 4 one-dimensional convolution layers are 32, 64, 64, 96 layer by layer. As in claim 1, a muscle fatigue analysis device using surface electromyography signals is provided, wherein an activation function of the frequency domain branch is a rectified linear unit function. As in claim 1, a muscle fatigue analysis device using surface electromyography signals is used, wherein the at least one frequency domain feature of the frequency domain electromyography signal extracted by the frequency domain branch is flattened, spliced with the at least one time domain feature of the time domain electromyography signal extracted by the time domain branch, and then output to the fully connected layer. The muscle fatigue analysis device using surface electromyography signals as claimed in claim 1 further comprises: an output unit connected to the processing unit and provided for connecting to an external device; wherein the output unit receives the analysis result and transmits the analysis result to the external device, and the external device displays the analysis result.

Images