The robust frequency domain feature and hybrid CNN model for fatigue detection based on sEMG signal

Excessive fatigue may cause secondary damage to the body during muscle rehabilitation. Therefore, muscle fatigue detection is essential for ensuring the safety and effectiveness of the training process. Deep-learning techniques have been widely used in fatigue detection. However, existing fatigue detection algorithms still require further improvement in terms of feature capture, robustness, and generalizability. This study proposed a new feature based on an autoregressive model as a robust frequency-domain feature. Additionally, we proposed a hybrid model for fatigue detection. An Inception block and feature reconstruction module was integrated into the Convolutional Neural Networks to extract multiscale features from the spectrogram. Meanwhile, handcrafted features were combined with deep learning features to increase the diversity of the features. We used four different machine learning algorithms to validate the effectiveness of the new feature. By adding the proposed feature to the feature set, the accuracy improved by an average of 7.4% in dataset1 and 2.6% in dataset2. Using our hybrid model, the accuracy of fatigue detection reached 93.26% on dataset1 and 94.45% on dataset2. Compared with existing methods, the proposed method exhibits better performance in muscle fatigue detection, providing new ideas for the development of surface electromyography signal fatigue detection applications in the future.