Muscle fatigue detection plays an important role in physical rehabilitation. Previous researches have demonstrated that sEMG offers superior sensitivity in detecting muscle fatigue compared to other biological signals. However, features extracted from sEMG may vary during dynamic contractions and across different subjects, which causes unstability in fatigue detection. To address these challenges, this research proposes a novel neural network comprising an Inception-attention module as a feature extractor, a fatigue classifier and a domain classifier equipped with a gradient reversal layer. The integrated domain classifier encourages the network to learn subject-invariant common fatigue features while minimizing subject-specific features. Furthermore, a supervised contrastive loss function is also employed to enhance the generalization capability of the model. Experimental results demonstrate that the proposed model achieved outstanding performance in three-class classification tasks, reaching 93.54% accuracy, 92.69% recall and 92.69% F1-score, providing a robust solution for cross-subject muscle fatigue detection, offering significant guidance for rehabilitation training and assistance.
Figures & tables
Figure 1: The structure of the Inception module.
Figure 2: The overall structure of the proposed IADAN.
Modules
Layers
Kernel Size
Kernel/Neurons Numbers
Input Shape
Output Shape
Feature Extractor
Dilated Convolution 1
7
32
(n,6,64,64)
(n,32,32,32)
Dilated Convolution 2
5
64
(n,32,32,32)
(n,64,16,16)
Inception Branch 1
1
8
(n,64,16,16)
(n,8,16,16)
Inception Branch 2
3
16
(n,64,16,16)
(n,16,16,16)
Inception Branch 3
5
32
(n,64,16,16)
(n,32,16,16)
Inception Branch 4
1
8
(n,64,16,16)
(n,8,16,16)
Table I: The Model Structure Parameters
Figure 3: The adversarial and contrastive learning process.
Figure 4: The overall experimental setup. (a) Sensors attachment and data acquisition system. (b) Experiment protocol.
Figure 5: The signal preprocessing protocol. (a) The segmented IMU data. (b) The normalized sEMG data (c) The time-frequency image.
Experiment
Accuracy(%)
Recall(%)
F1-score(%)
Fold 1
95.28
94.91
95.09
Fold 2
92.58
91.85
91.47
Fold 3
92.64
91.95
92.20
Fold 4
93.65
92.04
91.99
Average
93.54
92.69
92.69
Table II: The Results of K-fold Cross Validation Experiments
Figure 6: The Model performance analysis result. (a) The loss curve. (b) The accuracy curve.
Figure 7: The t-SNE visualization result. (a) Feature vectors colored according to fatigue states. (b) Feature vectors colored according to subject ID.
Network
Accuracy(%)
Recall(%)
F1-score(%)
IDAN
90.82
88.06
88.10
IAN
87.47
86.45
86.26
ADAN
79.94
73.95
75.94
IADAN
95.28
94.91
95.09
Table III: The Results of Structural Ablation Study
Loss Function
Accuracy(%)
Recall(%)
F1-score(%)
FCE
62.44
57.83
56.01
FCE+DCE
81.87
82.12
81.47
FCE+SC
87.47
86.45
86.26
FCE+DCE+SC
95.28
94.91
95.09
Table IV: The Results of Loss Function Ablation Study
Network
Accuracy(%)
Recall(%)
F1-score(%)
ResNet
86.32
84.03
83.85
CLT-Net
71.60
70.90
71.55
MFFNet
67.46
64.71
62.54
IADAN
95.28
94.91
95.09
Table V: The Results of Model Comparison Experiments
F. Kiki, P. P. Niaz, A. Madani, and C. Basdogan (corresponding author) are with the Robotics and Mechatronics Laboratory (RML) and the KUIS AI Center, Koc University, Sariyer, Istanbul 34450, Turkey.