Transfer learning in hand movement intention detection based on surface electromyography signals.

Electromyography (EMG) signals are commonly used in prosthetic limb studies. We have proposed a system to detect six basic hand movements using unsupervised and supervised classification algorithms. In this study, two-channel EMG recordings belonging to six different hand movements are analyzed and the performance of the wavelet-based features for hand movement clustering and classification are examined for six subjects (three females and three males). The approximation and detail components are obtained by four-level symmetric wavelet transform. The energy, mean, standard deviation, and entropy values of the wavelet components are calculated and the feature sets are generated. After feature extraction, feature set dimensionality is reduced using principal component analysis, and then the k-nearest neighbor method and k-means clustering are applied for classification and clustering, respectively. The analyses are performed subject-specifically and gender-specifically. Thus, it is possible to evaluate the gender effect on classification performances. Subject-specific hand movements were detected with accuracy in the range of 86.33–100%. Gender-specific hand movements were detected with an accuracy of 96.67% for males and 92.78% for females. The classification and clustering results support each other. It was observed that the samples of hand movements that were classified incorrectly were concentrated in the same clusters. Similarly, it was found that the hand movements that were easily detected were homogeneously clustered.

Surface Electromyography (sEMG) signal has been widely applied in solving many mechanical control problems such as prosthetic control, medical rehabilitation and sports science. Similarly, this is becoming relatively applicable for interventional surgery where robotic catheter systems are being proposed for safe vascular catheterization. In this paper, two basic features are investigated for characterization of users’ hand movements during vascular catheterization. For this purpose, controlled experiments were setup to acquire sEMG signals from six different muscles of nine volunteers who were asked to perform four basic movements namely, move-out, move-in, turn-in, turn-out; used during intravascular catheterization. Two features namely, Average Electromyography and Root Mean Square, were defined over an acquired database of sEMG signals from the nine subjects. The features were utilized to characterize and analyze the different motions in the hand movements dataset. Analysis of the processed signal shows that both features are highly comparable in terms of mean amplitude for all movements across the nine subjects. A significant difference (\(p=0.048\)) was observed in the thumb abductor muscle, and just between move-in and turn-in movements.

Surgical training using 3D virtual reality simulators has become an important routine in medical education. Recent research points to Leap Motion as an exciting interface for the control of virtual surgical instruments due to its simplicity and low-cost characteristics. However, previous studies using Leap Motion only evaluated movements of the whole hand, without considering individual finger movements during the manipulation of surgical instruments. This work investigates the use of Leap Motion as an interface for the capture of basic hand and finger movements during a simulated hysteroscopy using a 3D-printed hysteroscope model. We created a virtual simulated uterine environment containing a hysteroscope controlled by movements of the hand and fingers of a user actuating on a 3D-printed model hysteroscope. The model hysteroscope was positioned in a pivot basis allowing the capture of the following basic movements: leftward/rightward, upward/downward, forward/backward, and extrusion/retraction of the virtual resection loop (which rests on the end of the virtual virtual resectoscope). The findings indicate that the arc-shaped paths of the hysteroscope's alpha plane (rightward/leftward) and beta plane (upward/downward) movements are satisfactorily simulated by the virtual reality system. Using Intraclass Correlation, was noted that the similarity between the calculated (ideal standard) and measured arcs was highly significant on both planes (r = 0.9599 on the alpha plane, and r = 0.9208 on the beta plane). Also, the forward/backward trajectory is a straight line; the pinch gesture decreases its accuracy when increase its distance from the Interaction Box of Leap Motion. The results were satisfactorily since compared with previous works, which used Leap Motion for the capture of hands-free gesturing.

This paper proposes to monitor and recognize the fatigue state during football training by analyzing the surface electromyography (EMG) signals. The surface electromyography (EMG) signal is closely connected with the state during sports and training. First, power frequency interference, motion artifacts, and baseline drift in the surface electromyography (EMG) signal are removed; second, the authors extract 6 features: rectified average value (ARV), integrated electromyography myoelectric value (IEMG), root mean square of electromyography value (RMS), median frequency (MF), average power frequency (MPF), and electromyography power (TP) to represent the surface electromyography (EMG) signal; lastly, the extracted features are input into a one-class support vector machine to determine whether the player has been fatigued and are input into a weighted support vector machine to determine the degree of fatigue if the player has been fatigued. The experimental results show that more than 95% of the fatigue state can be recognized by surface electromyography (EMG) signal.

Various myoelectric prostheses controlled by electromyography (EMG) signals have been developed. However, there have been few studies that provide fast and accurate methods to predict handgrip force from EMG signals. Rapid and precise handgrip force prediction is required, especially for the real-time control system of myoelectric prostheses. In this study, extreme learning machine (ELM) is applied to predict handgrip force from surface EMG signals of forearm muscles. Furthermore, ELM is compared with support vector machine (SVM) and multiple nonlinear regression (MNLR). The below 10 % of the surface EMG and handgrip force signals were cut away, and then the root mean square feature extracted from the modified surface EMG signals was taken as input vector for these three kinds of predicting mechanisms. For the testing dataset, ELM achieved a slightly larger root mean squared error than SVM did and a smaller one than MNLR did. Meanwhile, all three methods showed high correlation coefficients. For the total processing time, ELM and MNLR consumed much less time than SVM did. Experimental results demonstrate that ELM possesses a relatively good accuracy and little consumed time, although SVM is effective for handgrip force estimation in terms of accuracy. Overall, ELM has a promising potential for predicting handgrip force rapidly and precisely.

Electromyography (EMG) signals hold critical significance in biomedical research, capturing muscle electrical activity during rest and contraction in the upper limb. Their versatility in applications, particularly in human-assisting robotic tools, drives ongoing exploration and research. This paper presents an original study focusing on leveraging machine learning techniques to classify EMG datasets and efficiently control a robotic arm based on predicted gestures. Data acquisition involves strategically placing an EMG muscle sensor on the forearm to ensure precise measurement of signals associated with hand gestures and movements. Diverse classifiers including random forest, support vector machine (SVM), K-nearest neighbors (KNN), Gaussian naïve Bayes, gated recurrent unit (GRU), long short-term memory (LSTM), artificial neural network (ANN), recurrent neural network (RNN), convolutional neural network (CNN), and vision transformer (ViT) are employed. Performance results are meticulously analyzed and presented in tabular format, showcasing the ViT classifier as the most successful, achieving an impressive 97.7% accuracy in robotic arm control. Notably, ANN, RNN, and CNN also exhibit high accuracy exceeding 90%. Furthermore, this work is comprehensively compared with existing literature, laying the groundwork for future advancements in human-robot interaction and cutting-edge assistive technologies that markedly enhance the quality of life for individuals with motor impairments or disabilities. The findings carry significant implications for designing and implementing intuitive, responsive robotic systems based on EMG signals.

Identification of the innervation zone is widely used to optimize the accuracy and precision of noninvasive surface electromyography (EMG) signals because the EMG signal is strongly influenced by innervation zones. However, simply structured fusiform muscle, such as biceps brachii muscle, has been employed mainly due to the simplicity with which the propagation from raw EMG signals can be observed. In this study, the optimum electrode location (OEL), free from innervational influence, was investigated by the propagation pattern of action potentials for brachii muscles and more complicated deltoid muscle structures using an automatized signal analysis technique. The technique employed newly developed computer software with additional clinical uses and minimized subjective differences. EMG signals were recorded using surface array electrodes during voluntary isometric contractions obtained from 12 healthy male subjects. Peaks in EMG signals were detected and averaged for each muscle. The propagation patterns and OEL were examined from biceps brachii muscles for all subjects and from deltoid muscles for seven subjects. The estimated locations were partially confirmed by comparing the root mean squares of the EMG signals. These results show that propagation patterns and OEL could be estimated simply and automatically even from the surface EMG signals of deltoid muscles.

With the increasing development of internet, the security of personal information becomes more and more important. Thus, variety of personal recognition methods have been introduced to ensure persons' information security. Traditional recognition methods such as Personal Identification Number (PIN), or Identification tag (ID) are vulnerable to hackers. Then the biometric technology, which uses the unique physiological characteristics of human body to identify user information has been proposed. But the biometrics widely used at present such as human face, fingerprint, iris, and voice can also be forged and falsified. The biometric with living body features such as electromyography (EMG) signal is a good method to achieve aliveness detection and prevent the spoofing attacks. However, there are few studies on personal recognition based on EMG signal. According to the application context, personal recognition system may operate either in identification mode or verification mode. In the personal identification mode, the system recognizes an individual by searching the templates of all the users in the database for a match. While in the personal verification mode, the system validates a person's identity by comparing the captured features with her or his own template(s) stored in the system database. In this paper, both EMG-based personal identification method and EMG-based personal verification method are investigated. First, the Myo armband is placed on the right forearm (specifically, the height of the radiohumeral joint) of 21 subjects to collect the surface EMG signal under hand-open gesture. Then, two different methods are proposed for EMG-based personal identification, i.e., personal identification method based on Discrete Wavelet Transform (DWT) and ExtraTreesClassifier, and personal identification method based on Continuous Wavelet Transform (CWT) and Convolutional Neural Networks (CNN). Experiments with 21 subjects show that the identification accuracy of this two methods can achieve 99.206% and 99.203% respectively. Then based on the identification method using CWT and CNN, transfer learning algorithm is adopted to solve the model update problem when new data is added. Finally, an EMG-based personal verification method using CWT and siamese networks is proposed. Experiments show that the verification accuracy of this method can achieve 99.285%.

Decoding finger and hand movements from sEMG electrodes placed on the forearm of transradial amputees has been commonly studied by many research groups. A few recent studies have shown an interesting phenomenon: simple correlations between distal phantom finger, hand and wrist voluntary movements and muscle activity in the residual upper arm in transhumeral amputees, i.e., of muscle groups that, prior to amputation, had no physical effect on the concerned hand and wrist joints. In this study, we are going further into the exploration of this phenomenon by setting up an evaluation study of phantom finger, hand, wrist and elbow (if present) movement classification based on the analysis of surface electromyographic (sEMG) signals measured by multiple electrodes placed on the residual upper arm of five transhumeral amputees with a controllable phantom limb who did not undergo any reinnervation surgery. We showed that with a state-of-the-art classification architecture, it is possible to correctly classify phantom limb activity (up to 14 movements) with a rather important average success (over 80% if considering basic sets of six hand, wrist and elbow movements) and to use this pattern recognition output to give online control of a device (here a graphical interface) to these transhumeral amputees. Beyond changing the way the phantom limb condition is apprehended by both patients and clinicians, such results could pave the road towards a new control approach for transhumeral amputated patients with a voluntary controllable phantom limb. This could ease and extend their control abilities of functional upper limb prosthetics with multiple active joints without undergoing muscular reinnervation surgery.

Today's advanced muscular sensing and processing technologies have made the acquisition of electromyography (EMG) signal which is valuable. EMG signal is the measurement of electrical potentials generated by muscle cells which is an indicator of muscle activity. Other than rehabilitation engineering and clinical applications, EMG signals can also be employed in the field of human computer interaction (HCI) system. In this work, the detection of different hand movements (left, right, up and down) was obtained using artificial neural network (ANN). A back-propagation (BP) network with Levenberg-Marquardt training algorithm was utilized. The conventional time and time-frequency based feature sets have been chosen to train the neural network. The simulation results show that the designed network is able to recognize hand movements with satisfied classification efficiency in average of 88.4%.

Today, human-computer interfaces are increasingly more often used and become necessary for human daily activities. Among some remarkable applications, we find: Wireless-computer controlling through hand movement, wheelchair directing/guiding with finger motions, and rehabilitation. Such applications are possible from the analysis of electromyographic (EMG) signals. Despite some research works have addressed this issue, the movement classification through EMG signals is still an open challenging issue to the scientific community -especially, because the controller performance depends not only on classifier but other aspects, namely: used features, movements to be classified, the considered feature-selection methods, and collected data. In this work, we propose an exploratory work on the characterization and classification techniques to identifying movements through EMG signals. We compare the performance of three classifiers (KNN, Parzen-density-based classifier and ANN) using spectral (Wavelets) and time-domain-based (statistical and morphological descriptors) features. Also, a methodology for movement selection is proposed. Results are comparable with those reported in literature, reaching classification errors of 5.18% (KNN), 14.7407% (ANN) and 5.17% (Parzen-density-based classifier).

The subject of the study is the methods of controlling a robotic manipulator based on surface electromyography (EMG) signals using individualized normalization and classification of gestures by a convolutional neural network (CNN). The aim of the work is to create an effective control system that combines pre-normalization of EMG signals and deep learning to increase the accuracy of gesture recognition and stability in real time. For this purpose, the following were used: signal collection using the Myo Armband, pre-processing (min-max and zero-mean), signal conversion into images through a sliding window, CNN training (ResNet, Adam optimization) and comparison with SVM and Random Forest models. The results obtained showed a classification accuracy of 97.27% in the test environment and 91.71% in real time. The CNN model outperformed traditional methods by 18–19%, and zero-mean increased the accuracy by 2.34% compared to min-max normalization. The system remained stable even with variations in the bracelet position due to individual normalization. Conclusions: the proposed system demonstrates high accuracy, reliability and adaptability in real time. The scientific novelty lies in the combination of individualized normalization of EMG signals with ResNet, which provides stability and exceeds the accuracy of traditional algorithms. In the future, it is planned to expand the set of gestures, study more complex conditions and optimize neural networks for embedded systems.

Electromyography (EMG) signals have been successfully employed for driving prosthetic limbs of a single or double degree of freedom. This principle works by using the amplitude of the EMG signals to decide between one or two simpler movements. This method underperforms as compared to the contemporary advances done at the mechanical, electronics, and robotics end, and it lacks intuition to perform everyday life tasks. Recently, research on myoelectric control based on pattern recognition (PR) shows promising results with the aid of machine learning classifiers. Using the approach termed as EMG-PR, EMG signals are divided into analysis windows, and features are extracted for each window. These features are then fed to the machine learning classifiers as input. By offering multiple class movements and intuitive control, this method has the potential to power Prosthesis an amputated subject to perform everyday life movements. In this paper, we investigate the effect of the analysis window and feature selection on classification accuracy of different hand and wrist movements using time-domain features. We show that effective data preprocessing and optimum feature selection helps to improve the classification accuracy of hand movements. We use publicly available hand and wrist gesture dataset of 40 intact and 11 trans-radial amputated subjects for experimentation. Results computed using different classification algorithms show that the proposed preprocessing and features selection outperforms the baseline and achieve up to \(98\%\) classification accuracy for both intact and trans-radial subjects.

People with disabilities correspond to about 25% of the Brazilian population. A great part of these people have physical impairments that difficult the use computer peripherals. This article presents the development of a system for detection of hand movements through the acquisition and classification of electromyographic (EMG) signals using machine learning techniques. The purpose of the proposed system is to be used by people with disabilities to control an adapted text editor. The signals are capture by surface EMG electrodes and used to the detect 4 different hand movements. In addition, a database with 3200 EMG signals generated by the hand movements was created, made by one user diagnosed with cerebral palsy and another user without diagnosed motor disabilities. Several tests were carried out, showing the good accuracy of the proposed system, with a success classification rate of 96% to 98%.

Pattern recognition of electromyography (EMG) signals can potentially improve the performance of myoelectric control for upper limb prostheses with respect to current clinical approaches based on direct control. However, the choice of features for classification is challenging and impacts long-term performance. Here, we propose the use of EMG raw signals as direct inputs to deep networks with intrinsic feature extraction capabilities recorded over multiple days. Seven able-bodied subjects performed six active motions (plus rest), and EMG signals were recorded for 15 consecutive days with two sessions per day using the MYO armband (MYB, a wearable EMG sensor). The classification was performed by a convolutional neural network (CNN) with raw bipolar EMG samples as the inputs, and the performance was compared with linear discriminant analysis (LDA) and stacked sparse autoencoders with features (SSAE-f) and raw samples (SSAE-r) as inputs. CNN outperformed (lower classification error) both LDA and SSAE-r in the within-session, between sessions on same day, between the pair of days, and leave-out one-day evaluation (p < 0.001) analyses. However, no significant difference was found between CNN and SSAE-f. These results demonstrated that CNN significantly improved performance and increased robustness over time compared with standard LDA with associated handcrafted features. This data-driven features extraction approach may overcome the problem of the feature calibration and selection in myoelectric control.