Surface electromyography (sEMG) signals can reflect the body motion information and are widely used in military, medical rehabilitation, industrial production. The lower limb motion classification mainly includes feature extraction and classification model establishment. Firstly, we proposed a feature extraction method based on the wavelet packet transform (WPT) and principal component analysis (PCA). We used the wavelet packet method to decompose the sEMG signals of three muscles in the lower limb and got the 24-dimensional eigenvector. To reduce the calculation and improve the speed of the classification model, we used the PCA method to reduce the dimension of the feature vector and got the 3-dimensional eigenvector. Then, we proposed a method based on the scale unscented Kalman filter (SUKF) and neural network (NN) for lower limb motion classification. Through the scale correction unscented transform (SCUT) could optimize the neural network weight and improve lower limb motion classification accuracy. Finally, the experimental results showed that the average accuracy was 93.7%. Compared with the backpropagation neural network (BPNN) and wavelet neural network (WNN), this method could improve the accuracy and reliability of the lower limb motion classification.

The real-time and accuracy of motion classification plays an essential role for the elderly or frail people in daily activities. This study aims to determine the optimal feature extraction and classification method for the activities of daily living (ADL). In the experiment, we collected surface electromyography (sEMG) signals from thigh semitendinosus, lateral thigh muscle, and calf gastrocnemius of the lower limbs to classify horizontal walking, crossing obstacles, standing up, going down the stairs, and going up the stairs. Firstly, we analyzed 11 feature extraction methods, including time domain, frequency domain, time-frequency domain, and entropy. Additionally, a feature evaluation method was proposed, and the separability of 11 feature extraction algorithms was calculated. Then, combined with 11 feature algorithms, the classification accuracy and time of 55 classification methods were calculated. The results showed that the Gaussian Kernel Linear Discriminant Analysis (GK-LDA) with WAMP had the highest classification accuracy rate (96%), and the calculation time was below 80 ms. In this paper, the quantitative comparative analysis of feature extraction and classification methods was a benefit to the application for the wearable sEMG sensor system in ADL.

Due to the strong dynamic and time-varying characteristics of the surface electromyography (sEMG) signal and the coupling between lower limb muscles, it is challenging to accurately estimate the joint angle of complex lower limb motion. Firstly, we proposed a wavelet packet decomposition correlation dimension (WPD_CD) analysis method. A three-level wavelet packet decomposes the sEMG signal, and the optimal and stable feature vector is extracted by the correlation dimension analysis. A singular spectrum nonlinear autoregressive exogenous model (NARX_SSA) is proposed to map the optimal eigenvector to the lower limb joint angle to improve the angle accuracy and smoothness. The experimental results show that this method only needs to collect three lower limb muscles and accurately estimate four movements of knee and hip angles. The overall RMSE was 0.32, the minimum RMSE was 0.20, and the standard deviation (SD) was 20.23. Compared with other methods, the accuracy of angle estimation was improved by six times, and the smoothness was improved by 13%. Therefore, this method can be effectively applied to the angle estimation of complex lower limb motion patterns.

Exoskeleton robot is a typical application to assist the motion of lower limbs. To make the lower extremity exoskeleton more flexible, it is necessary to identify various motion intentions of the lower limbs of the human body. Although more sEMG sensors can be used to identify more lower limb motion intention, with the increase in the number of sensors, more and more data need to be processed. In the process of human motion, the collected sEMG signal is easy to be interfered with noise. To improve the practicality of the lower extremity exoskeleton robot, this paper proposed a wavelet packet transform- (WPT-) based sliding window difference average filtering feature extract algorithm and the unscented Kalman neural network (UKFNN) recognition algorithm. We established an sEMG energy feature model, using a sliding window difference average filtering method to suppress noise interference and extracted stable feature values and using UKF filtering to optimize the neural network weights to improve the adaptability and accuracy of the recognition model. In this paper, we collected the sEMG signals of three muscles to identify six lower limb motion intentions. The average accuracy of 94.83% is proposed in this paper. Experiments show that the algorithm improves the accuracy and anti-interference of motion intention recognition of lower limb sEMG signals. The algorithm is superior to the backpropagation neural network (BPNN) recognition algorithm in the lower limb motion intention recognition and proves the effectiveness, novelty, and reliability of the method in this paper.

Continuous joint angle estimation is essential for enhancing man-machine collaboration performance. However, it is challenging to estimate the complex multi-joint angle of the lower limb accurately. Firstly, a non-redundant feature extraction algorithm for muscle synergy was proposed. The non-negative matrix factorization (NMF) algorithm was used to extract the muscle activation coefficient matrix, and the muscle activation coefficient matrix was divided into non-redundant and redundant feature vectors. Then, a state-space frame model with non-redundant features as input and redundant features as measurement output to reduce system error was proposed. The square root unscented Kalman filter (SRUKF) algorithm was used to estimate the multi-joint angle of lower limbs. We recruited ten subjects to participate in seven daily activities, including going upstairs (US), downing stairs (DS), going uphill (UH), going downhill (DH), and walking at three speeds of 0.6 m/s, 1.0 m/s, and 1.4 m/s. The results showed that the average RMSE of the proposed approach for estimating hip and knee joint angles was 0.44±0.1 and 0.73±0.5, respectively, which was significantly smaller than the common neural networks (p<0.05). Particularly, the anti-interference performance of the proposed model was tested. Meanwhile, the adaptability test was carried out through the developed lower-limb multi-joint angle estimation verification system, which proved that the proposed approach could provide accurate and stable estimation results by making full use of redundant features. It can improve the safety of online applications for sEMG auxiliary equipment.

Accurate recognition of transitional and steady states in lower limb wearable exoskeletons using surface electromyography (sEMG) signals remains challenging due to the transition sEMG signals exhibiting short duration, strong nonlinearity, and similarity to the preceding and following steady states. This study proposed a novel classification method to classify various modes, including transition and steady modes. Specifically, an improved complete ensemble empirical mode decomposition adaptive noise (ICEEMDAN) method, combined with the crested porcupine optimization (CPO) algorithm, was proposed to enhance the ability to capture sEMG signal features. A dual-branch fusion recognition network was proposed, it extracts the correlation motion features between different channel. The multi-scale CNN was employed to extract motion features for steady and transition modes. The attention mechanism was applied to enhance the representation of fused features from the CNN. Finally, a DenseNet network was employed for precise classification. Eight participants were recruited for the experiment, and sEMG data were collected for five steady modes and eight transition modes. It was validated through extensive experiments that the proposed method can effectively classify transition and steady modes, achieving an average accuracy of 97.39%, which outperforms existing methods. The proposed method achieves a recognition time of 25.5108 ms, satisfying the requirements for real-time deployment. Additionally, it attains accuracies of 0.8616 on two publicly available datasets, demonstrating generalization capability in the field of lower limb motion recognition.