Table 1 - uploaded by Angkoon Phinyomark
Content may be subject to copyright.
A list of EMG feature extraction techniques. 

A list of EMG feature extraction techniques. 

Source publication
Article
Full-text available
The success of biological signal pattern recognition depends crucially on the selection of relevant features. Across signal and imaging modalities, a large number of features have been proposed, leading to feature redundancy and the need for optimal feature set identification. A further complication is that, due to the inherent biological variabili...

Citations

... Important factors include electrodes shift, varying contraction force levels, forearm orientation and variation in limb positions, intersubject variability, and fatigue [2]- [6]. The fundamental discriminatory power of EMG pattern classifiers, however, has been found to largely depend on the quality of the extracted features [7], [8]. In this regard, when classifying, EMG data acquired over numerous days, Phinyomark et al. [9] examined 50 feature extraction approaches for EMG pattern identification and concluded that the combination of sample Entropy, fourth-order cepstrum coefficients, root mean square, and waveform length is a promising feature subset. ...
... The literature has so far presented us with a significant number of time-domain feature extraction (TDFE) algorithms that have been well investigated and utilized in myoelectric control [8], [9], [33]. In this article, we decided to utilize the TD-PSD that was proposed in the work of Al-Timemy et al. [6] as the base features, since the TD-PSD has now been well tested and adopted in previous studies [13], [34]. ...
Article
The role of feature extraction in electromyogram (EMG) based pattern recognition has recently been emphasized with several publications promoting deep learning (DL) solutions that outperform traditional methods. It has been shown that the ability of DL models to extract temporal, spatial, and spatio–temporal information provides significant enhancements to the performance and generalizability of myoelectric control. Despite these advancements, it can be argued that DL models are computationally very expensive, requiring long training times, increased training data, and high computational resources, yielding solutions that may not yet be feasible for clinical translation given the available technology. The aim of this paper is, therefore, to leverage the benefits of spatio–temporal DL concepts into a computationally feasible and accurate traditional feature extraction method. Specifically, the proposed novel method extracts a set of well-known time-domain features into a matrix representation, convolves them with predetermined fixed filters, and temporally evolves the resulting features over a short and long-term basis to extract the EMG temporal dynamics. The proposed method, based on Fixed Spatio–Temporal Convolutions, offers significant reductions in the computational costs, while demonstrating a solution that can compete with, and even outperform, recent DL models. Experimental tests were performed on sparse-and high-density EMG (HD-EMG) signals databases, across a total of 44 subjects performing a maximum of 53 movements. Despite the simplification compared to deep approaches, our results show that the proposed solution significantly reduces the classification error rates by 3% to 10% in comparison to recent DL models, while being efficient for real-time implementations.
... Then, five distinct feature sets used in the space of EMG signal characterization were individually extracted, resulting in the formation of a feature vector that is applied in building a machine learning model to decode inherent motor tasks. Briefly put, the extracted features are Novel time-domain features (NTDF) [4]; Time-dependent power spectral density (TD-PSD) [26]; Hudgin's time-domain features (TD4) [27]; Fifth-order autoregressive coefficient (AR5); and root mean square (RMS) [28]. Furthermore, the TD-PSD extracts motor information that compensates for the effect of force variation, while NTDF constructs EMG feature vectors that are robust to the combined impact of muscle contraction force variation and mobility of the subject when performing upper limb movements. ...
Article
Surface electromyogram (sEMG) is arguably the most sought-after physiological signal with a broad spectrum of biomedical applications, especially in miniaturized rehabilitation robots such as multifunctional prostheses. The widespread use of sEMG to drive pattern recognition (PR)-based control schemes is primarily due to its rich motor information content and non-invasiveness. Moreover, sEMG recordings exhibit non-linear and non-uniformity properties with inevitable interferences that distort intrinsic characteristics of the signal, precluding existing signal processing methods from yielding requisite motor control information. Therefore, we propose a multiresolution decomposition driven by dual-polynomial interpolation (MRDPI) technique for adequate denoising and reconstruction of multi-class EMG signals to guarantee the dual-advantage of enhanced signal quality and motor information preservation. Parameters for optimal MRDPI configuration were constructed across combinations of thresholding estimation schemes and signal resolution levels using EMG datasets of amputees who performed up to 22 predefined upper-limb motions acquired in-house and from the public NinaPro database. Experimental results showed that the proposed method yielded signals that led to consistent and significantly better decoding performance for all metrics compared to existing methods across features, classifiers, and datasets, offering a potential solution for practical deployment of intuitive EMG-PR-based control schemes for multifunctional prostheses and other miniaturized rehabilitation robotic systems that utilize myoelectric signals as control inputs.
... Then, five distinct feature sets used in the space of EMG signal characterization were individually extracted, resulting in the formation of a feature vector that is applied in building a machine learning model to decode inherent motor tasks. Briefly put, the extracted features are Novel time-domain features (NTDF) [4]; Time-dependent power spectral density (TD-PSD) [26]; Hudgin's time-domain features (TD4) [27]; Fifth-order autoregressive coefficient (AR5); and root mean square (RMS) [28]. Furthermore, the TD-PSD extracts motor information that compensates for the effect of force variation, while NTDF constructs EMG feature vectors that are robust to the combined impact of muscle contraction force variation and mobility of the subject when performing upper limb movements. ...
Preprint
Surface electromyogram (sEMG) is arguably the most sought-after physiological signal with a broad spectrum of biomedical applications, especially in miniaturized rehabilitation robots such as multifunctional prostheses. The widespread use of sEMG to drive pattern recognition (PR)-based control schemes is primarily due to its rich motor information content and non-invasiveness. Moreover, sEMG recordings exhibit non-linear and non-uniformity properties with inevitable interferences that distort intrinsic characteristics of the signal, precluding existing signal processing methods from yielding requisite motor control information. Therefore, we propose a multiresolution decomposition driven by dual-polynomial interpolation (MRDPI) technique for adequate denoising and reconstruction of multi-class EMG signals to guarantee the dual-advantage of enhanced signal quality and motor information preservation. Parameters for optimal MRDPI configuration were constructed across combinations of thresholding estimation schemes and signal resolution levels using EMG datasets of amputees who performed up to 22 predefined upper-limb motions acquired in-house and from the public NinaPro database. Experimental results showed that the proposed method yielded signals that led to consistent and significantly better decoding performance for all metrics compared to existing methods across features, classifiers, and datasets, offering a potential solution for practical deployment of intuitive EMG-PR-based control schemes for multifunctional prostheses and other miniaturized rehabilitation robotic systems that utilize myoelectric signals as control inputs.
... s (Belter et al. 2013;Tam et al. 2021). On the other hand, human hand can perform a grasping task within a period 300 ms following the neuromuscular time constraint (Phinyomark et al. 2017). Further, Gigli et al. (2020) reported that machine learning based pattern recognition methods for prosthetic hand control has limitation for clinical practice due to requirement of higher order computational resources. ...
Article
Full-text available
Development of prosthetic hands with human-like functionality and controllability is one of the major goals in the area of rehabilitation robotics. Current developments on prosthetic hands have earned higher functionality with multiple fingers and degrees of freedom. However, the issue of time required to perform a grasp type opens avenues for improvement in its controllability. This paper reports a real-time electromyogram (EMG) based embedded controller for prosthetic hands. The focus was on development of an efficient controller in terms of grasping accuracy and time required for grasping vis-á-vis human hand neuromuscular time constraint. The controller has been tested for a prosthetic hand to grasp four objects: cricket ball, coffee mug, screw-driver box and plastic container. EMG from biceps brachii muscles during maximum voluntary contraction versus resting state was classified. With an aim for low computational complexity in the controller such that the reported work can be translated into a low cost commercial product, a finite state algorithm was used to understand user’s grasping intention. Experiments have been accomplished in four sessions, each with 20 trials, by five subjects in both sitting and standing positions. It has been found that the prosthetic hand can perform grasping with an average accuracy of 96.2 ± 2.6%. The controller enables the prosthetic hand to perform grasping operation in 250.80 ± 1.1 ms, which is comparable to the time required by human hands i.e. 300 ms and thereby satisfied the neuromuscular constraint.
... The calculation or the approach we used to determine the k function or decision rules for the K-NN algorithm is the majority voting scheme [40,41] which deepens on the category) or class that has one vote for each instance on the class set of the sample K-neighborhood samples. In that case, the new data sample is classified according to the highest number of votes in the class. ...
Article
Full-text available
The real-time recognition of pain level is required to perform an accurate pain assessment of patients in the intensive care unit, infants, and other subjects who may not be able to communicate verbally or even express the sensation of pain. Facial expression is a key pain-related behavior that may unlock the answer to an objective pain measurement tool. In this work, a machine learning-based pain level classification system using data collected from facial electromyograms (EMG) is presented. The dataset was acquired from part of the BioVid Heat Pain database to evaluate facial expression from an EMG corrugator and EMG zygomaticus and an EMG signal processing and data analysis flow is adapted for continuous pain estimation. The extracted pain-associated facial electromyography (fEMG) features classification is performed by K-nearest neighbor (KNN) by choosing the value of k which depends on the nonlinear models. The presentation of the accuracy estimation is performed, and considerable growth in classification accuracy is noticed when the subject matter from the features is omitted from the analysis. The ML algorithm for the classification of the amount of pain experienced by patients could deliver valuable evidence for health care providers and aid treatment assessment. The proposed classification algorithm has achieved a 99.4% accuracy for classifying the pain tolerance level from the baseline (P0 versus P4) without the influence of a subject bias. Moreover, the result on the classification accuracy clearly shows the relevance of the proposed approach.
... Well known state-of-the-art commercial prostheses such as the Michelangelo and Bebionic hand by Otto Bock [7], the i-limb by Össur [8] and the Hero Arm by Open Bionics [9] incorporate some form of this technology. Within the field, research has focused on feature engineering, e.g., signal amplitude and power, frequency information, or time series modeling and discriminative description [6,10]. Additionally, deep learning-driven approaches that incorporate convolutional networks have gained traction in recent years. ...
Article
Full-text available
Background Existing assistive technologies attempt to mimic biological functions through advanced mechatronic designs. In some occasions, the information processing demands for such systems require substantial information bandwidth and convoluted control strategies, which make it difficult for the end-user to operate. Instead, a practical and intuitive semi-automated system focused on accomplishing daily tasks may be more suitable for end-user adoption. Methods We developed an intelligent prosthesis for the Cybathlon Global Edition 2020. The device was designed in collaboration with the prosthesis user (pilot), addressing her needs for the competition and aiming for functionality. Our design consists of a soft robotic-based two finger gripper controlled by a force-sensing resistor (FSR) headband interface, automatic arm angle dependent wrist flexion and extension, and manual forearm supination and pronation for a shared control system. The gripper is incorporated with FSR sensors to relay haptic information to the pilot based on the output of a neural network model that estimates geometries and objects material. Results As a student team of the Munich Institute of Robotics and Machine Intelligence, we achieved 12th place overall in the Cybathlon competition in which we competed against state-of-the-art prosthetic devices. Our pilot successfully accomplished two challenging tasks in the competition. During training sessions, the pilot was able to accomplish the remaining competition tasks except for one. Based on observation and feedback from training sessions, we adapted our developments to fit the user’s preferences. Usability ratings indicated that the pilot perceived the prosthesis to not be fully ergonomic due to the size and weight of the system, but argued that the prosthesis was intuitive to control to perform the tasks from the Cybathlon competition. Conclusions The system provides an intuitive interface to conduct common daily tasks from the arm discipline of the Cybathlon competition. Based on the feedback from our pilot, future improvements include the prosthesis’ reduction in size and weight in order to enhance its mobility. Close collaboration with our pilot has allowed us to continue with the prosthesis development. Ultimately, we developed a simple-to-use solution, exemplifying a new paradigm for prosthesis design, to help assist arm amputees with daily activities.
... A first motivation relies on the high number of features known to be involved in many biological studies. This is the case of the work in [90] where TDA is integrated into feature selection to detect subgroups of features based on similarity measures. ...
Chapter
The aim of this chapter is to give a handy but thorough introduction to persistent homology and its applications. The chapter’s path is made by the following steps. First, we deal with the constructions from data to simplicial complexes according to the kind of data: filtrations of data, point clouds, networks, and topological spaces. For each construction, we underline the possible dependence on a fixed scale parameter. Secondly, we introduce the necessary algebraic structures capturing topological informations out of a simplicial complex at a fixed scale, namely the simplicial homology groups and the Hodge Laplacian operator. The so-obtained linear structures are then integrated into the multiscale framework of persistent homology where the entire persistence information is encoded in algebraic terms and the most advantageous persistence summaries available in the literature are discussed. Finally, we introduce the necessary metrics in order to state properties of stability of the introduced multiscale summaries under perturbations of input data. At the end, we give an overview of applications of persistent homology as well as a review of the existing tools in the broader area of Topological Data Analysis (TDA).
... The PANDA and SVM methods both require computing multiple features from clean and newly acquired signals, differentiating these from approaches like the statistical tests of normality. Many of the handcrafted features used in sEMG signal quality assessment have been inspired by the myoelectric control community [10], including time domain, frequency domain, and time-frequency domain features. However, feature sets that perform well for pattern recognition in myoelectric control are not necessarily optimal for contamination detection. ...
Article
This work is an exposition of research supporting efforts to automate quality assessment in surface electromyography (sEMG). Electromyography measures electrical activity from skeletal muscles, which are the muscles associated with voluntary movements. Skeletal muscles can consist of tens to hundreds of thousands of contractile fibers which are grouped into functional units called Motor Units (MUs). Each MU is driven to contract its fibers by electrochemical impulses sent from a motor neuron, and a muscle may contain anywhere from tens to hundreds of MUs. When an MU is activated, all fibers associated with it are simultaneously activated. When fibers are activated, an electrochemical impulse propagates along each of them to cause contraction. This impulse is known as a Single Fiber Action Potential (SFAP). A Motor Unit Action Potential (MUAP) is the summation of all SFAPs corresponding to an activated MU. To generate a continuous contraction, MUs are repeatedly activated, producing a train of MUAPs, and this pulse train is called the Motor Unit Action Potential Train (MUAPT). The combined electrical activity of all of the MUAPTs associated with a contraction, as recorded on the surface of the skin using non-invasive electrodes, is a surface electromyography (sEMG) signal. In general, it is not possible to directly observe the SFAPs, MUAPs, and MUAPTs non-invasively (though there have been algorithmic efforts to decompose sEMG signals into its constituent MUAPTs). SFAPs and MUAPTs can be observed by inserting needle electrodes into specific muscle regions for high spatial selectivity; however, such techniques are invasive, making surface measurements a desirable alternative. sEMG signals have been used in a wide variety of applications, including fatigue assessment, myoelectric control, diagnosis of neuromuscular disorders, and tracking performance in sports.
... Based on empirical testing, the HMM was trained using three states and 64 Gaussian mixture components. [18]. ...
Conference Paper
Surface electromyography (sEMG) signals are now commonly used in continuous myoelectric control of prostheses. More recently, researchers have considered EMG-based gesture recognition systems for human computer interaction research. These systems instead focus on recognizing discrete gestures (like a finger snap). The majority of works, however, have focused on improving multi-class performance, with little consideration for false activations from "other" classes. Consequently, they lack the robustness needed for real-world applications which generally require a single motion class such as a mouse click or a wake word. Furthermore, many works have borrowed the windowed classification schemes from continuous control, and thus fail to leverage the temporal structure of the gesture. In this paper, we propose a verification-based approach to creating a robust EMG wake word using one-class classifiers (Support Vector Data Description, One Class-Support Vector Machine, Dynamic Time Warping (DTW) & Hidden Markov Models). The area under the ROC curve (AUC) is used as a feature optimization objective as it provides a better representation of the verification performance. Equal error rate (EER) and AUC are then used as evaluation metrics. The results are computed using both window-based and temporal classifiers on a dataset consisting of five different gestures, with a best EER of 0.04 and AUC of 0.98, recorded using a DTW scheme. These results demonstrate a design framework that may benefit the development of more robust solutions for EMG-based wake words or input commands for a variety of interactive applications.
... The feature extraction process was done on both frequency domain and time domain [19]. For each channel, 4 frequency features were extracted: the integral values of amplitude, root mean square frequency (RMSF) the frequency centroid (FC), and root var frequency (RVF) [20]. ...
Preprint
Human taste sensation can be qualitatively described with surface electromyography. However, the pattern recognition models trained on one subject (the source domain) do not generalize well on other subjects (the target domain). To improve the generalizability and transferability of taste sensation models developed with sEMG data, two methods were innovatively applied in this study: domain regularized component analysis (DRCA) and conformal prediction with shrunken centroids (CPSC). The effectiveness of these two methods was investigated independently in an unlabeled data augmentation process with the unlabeled data from the target domain, and the same cross-user adaptation pipeline were conducted on six subjects. The results show that DRCA improved the classification accuracy on six subjects (p < 0.05), compared with the baseline models trained only with the source domain data;, while CPSC did not guarantee the accuracy improvement. Furthermore, the combination of DRCA and CPSC presented statistically significant improvement (p < 0.05) in classification accuracy on six subjects. The proposed strategy combining DRCA and CPSC showed its effectiveness in addressing the cross-user data distribution drift in sEMG-based taste sensation recognition application. It also shows the potential in more cross-user adaptation applications.