Comparison of machine learning methods in sEMG signal processing for shoulder motion recognition

Zhou, Yang; Chen, Chaoyang; Cheng, Mark Ming‐Cheng; Alshahrani, Yousef; Franovic, Sreten; Lau, Emily; Xu, Guanghua; Ni, Guo‐Xin; Cavanaugh, John M.; Muh, Stephanie J.; Lemos, Stephen

doi:10.1016/j.bspc.2021.102577

Cited by 24 publications

(16 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The support vector machine (SVM) and CNN are two popular ML methods used for ML-based EMG signal processing. Only a few studies successfully demonstrated the ML approaches to process EMG signals for real-time robot control using both RMS and frequency variables (Zhou et al, 2021 ). A powerful computer and LabView software with a machine learning toolbox are required to implement the ML tasks with significant efforts devoted to establishing portal communication between EMG sensor systems and ML processing toolbox software.…”

Section: Discussionmentioning

confidence: 99%

“…Artificial intelligence and machine learning (ML) are emerging techniques in EMG signal processing for motion pattern recognition and robot control. However, the ML-based firmware is expensive and significant efforts are needed for algorithm development (Jiang et al, 2020 ; Zhou et al, 2021 ). In recent years, deep learning-based human-robot interaction (HRI) was developed (Qi et al, 2021 , 2022 ; Su et al, 2021a ) and achieved a higher recognition accuracy and faster inference speed with the help of GPU.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Real-Time EMG-Based Fixed-Bandwidth Frequency-Domain Embedded System for Robotic Hand

Chen

et al. 2022

Front. Neurorobot.

Self Cite

View full text Add to dashboard Cite

The signals from electromyography (EMG) have been used for volitional control of robotic assistive devices with the challenges of performance improvement. Currently, the most common method of EMG signal processing for robot control is RMS (root mean square)-based algorithm, but system performance accuracy can be affected by noise or artifacts. This study hypothesized that the frequency bandwidths of noise and artifacts are beyond the main EMG signal frequency bandwidth, hence the fixed-bandwidth frequency-domain signal processing methods can filter off the noise and artifacts only by processing the main frequency bandwidth of EMG signals for robot control. The purpose of this study was to develop a cost-effective embedded system and short-time Fourier transform (STFT) method for an EMG-controlled robotic hand. Healthy volunteers were recruited in this study to identify the optimal myoelectric signal frequency bandwidth of muscle contractions. The STFT embedded system was developed using the STM32 microcontroller unit (MCU). The performance of the STFT embedded system was compared with RMS embedded system. The results showed that the optimal myoelectric signal frequency band responding to muscle contractions was between 60 and 80 Hz. The STFT embedded system was more stable than the RMS embedded system in detecting muscle contraction. Onsite calibration was required for RMS embedded system. The average accuracy of the STFT embedded system is 91.55%. This study presents a novel approach for developing a cost-effective and less complex embedded myoelectric signal processing system for robot control.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Real-Time EMG-Based Fixed-Bandwidth Frequency-Domain Embedded System for Robotic Hand

Chen

et al. 2022

Front. Neurorobot.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Jose et al (2017) extracted the time domain sEMG features of the subjects' forearm movement and classified the features using a multi-layer perception network, with a classification accuracy of 91.6%. In the literature (Zhou et al, 2021), the machine learning method was applied to the recognition of shoulder movements, and the support vector machine (SVM) method with a sliding time window of 270ms was used, and the classification accuracy was more than 90% (Hinton et al, 2012). Narayan et al (2018) extracted the features of the sEMG signal by first-order differential and classified the features by a medium tree classifier, which improved the classification accuracy by 6% compared with other features.…”

Section: Semg-based Hri Related Studymentioning

confidence: 99%

“…Traditional sEMG-based HRI usually extracts hand-crafted features and then uses machine learning methods to build mappings of hand-crafted features and different lower limb movements (Jose et al, 2017;Motoche and Benalcázar, 2018;Narayan et al, 2018;Khiabani and Ahmadi, 2021;Zhou et al, 2021). The interface is computationally cheap and can achieve a relatively good lower-limb movements prediction performance in most cases.…”

Section: Introductionmentioning

confidence: 99%

An sEMG-Based Human-Exoskeleton Interface Fusing Convolutional Neural Networks With Hand-Crafted Features

Yang

et al. 2022

Front. Neurorobot.

View full text Add to dashboard Cite

In recent years, the human-robot interfaces (HRIs) based on surface electromyography (sEMG) have been widely used in lower-limb exoskeleton robots for movement prediction during rehabilitation training for patients with hemiplegia. However, accurate and efficient lower-limb movement prediction for patients with hemiplegia remains a challenge due to complex movement information and individual differences. Traditional movement prediction methods usually use hand-crafted features, which are computationally cheap but can only extract some shallow heuristic information. Deep learning-based methods have a stronger feature expression ability, but it is easy to fall into the dilemma of local features, resulting in poor generalization performance of the method. In this article, a human-exoskeleton interface fusing convolutional neural networks with hand-crafted features is proposed. On the basis of our previous study, a lower-limb movement prediction framework (HCSNet) in patients with hemiplegia is constructed by fusing time and frequency domain hand-crafted features and channel synergy learning-based features. An sEMG data acquisition experiment is designed to compare and analyze the effectiveness of HCSNet. Experimental results show that the method can achieve 95.93 and 90.37% prediction accuracy in both within-subject and cross-subject cases, respectively. Compared with related lower-limb movement prediction methods, the proposed method has better prediction performance.

show abstract

“…sEMG signals have verified that there is good performance for the human intention prediction because sEMG signals can immediately reflect the muscle activity and movement intention [23]. In general, the basic process of sEMG-based human intention prediction includes data preprocessing, data segmentation, feature extraction and classifier design [24]. The action-recognition accuracy, as well as the real-time and model robustness, are very important factors during the process of rehabilitation training.…”

Section: Introductionmentioning

confidence: 99%

sEMG-Based Motion Recognition of Upper Limb Rehabilitation Using the Improved Yolo-v4 Algorithm

Guo

2022

Life

View full text Add to dashboard Cite

The surface electromyography (sEMG) signal is widely used as a control source of the upper limb exoskeleton rehabilitation robot. However, the traditional way of controlling the exoskeleton robot by the sEMG signal requires one to specially extract and calculate for complex sEMG features. Moreover, due to the huge amount of calculation and individualized difference, the real-time control of the exoskeleton robot cannot be realized. Therefore, this paper proposes a novel method using an improved detection algorithm to recognize limb joint motion and detect joint angle based on sEMG images, aiming to obtain a high-security and fast-processing action recognition strategy. In this paper, MobileNetV2 combined the Ghost module as the feature extraction network to obtain the pretraining model. Then, the target detection network Yolo-V4 was used to estimate the six movement categories of the upper limb joints and to predict the joint movement angles. The experimental results showed that the proposed motion recognition methods were available. Every 100 pictures can be accurately identified in approximately 78 pictures, and the processing speed of every single picture on the PC side was 17.97 ms. For the train data, the mAP@0.5 could reach 82.3%, and mAP@0.5–0.95 could reach 0.42; for the verification data, the average recognition accuracy could reach 80.7%.

show abstract

Comparison of machine learning methods in sEMG signal processing for shoulder motion recognition

Cited by 24 publications

References 39 publications

A Real-Time EMG-Based Fixed-Bandwidth Frequency-Domain Embedded System for Robotic Hand

A Real-Time EMG-Based Fixed-Bandwidth Frequency-Domain Embedded System for Robotic Hand

An sEMG-Based Human-Exoskeleton Interface Fusing Convolutional Neural Networks With Hand-Crafted Features

sEMG-Based Motion Recognition of Upper Limb Rehabilitation Using the Improved Yolo-v4 Algorithm

Contact Info

Product

Resources

About