基于表面肌电信号的人体动作识别算法研究进展

表面肌电信号(Surface Electromyography,sEMG)是通过相应肌群表面的传感器记录下来的一维时间序列非平稳生物电信号,不但反映了神经肌肉系统活动,对于反映相应动作肢体活动信息同样重要。而模式识别是肌电应用领域的基础和关键。为了在应用基于表面肌电信号模式识别中选取合适算法,本文拟对基于表面肌电信号的人体动作识别算法进行回顾分析,主要包括模糊模式识别算法、线性判别分析算法、人工神经网络算法和支持向量机算法。模糊模式识别能自适应提取模糊规则,对初始化规则不敏感,适合处理s EMG这样具有严格不重复的生物电信号;线性判别分析对数据进行降维,计算简单,但不适合大数据;人工神经网络可以同时描述训练样本输入输出的线性关系和非线性映射关系,可以解决复杂的分类问题,学习能力强;支持向量机处理小样本、非线性的高维数据优势明显,计算速度快。比较各方法的优缺点,为今后处理此类问题模式识别算法选取提供了参考和依据。     

Techniques of EMG signal analysis: detection, processing, classification and applications

Electromyography (EMG) signals can be used for clinical/biomedical applications, Evolvable Hardware Chip (EHW) development, and modern human computer interaction. EMG signals acquired from muscles require advanced methods for detection, decomposition, processing, and classification. The purpose of this paper is to illustrate the various methodologies and algorithms for EMG signal analysis to provide efficient and effective ways of understanding the signal and its nature. We further point up some of the hardware implementations using EMG focusing on applications related to prosthetic hand control, grasp recognition, and human computer interaction. A comparison study is also given to show performance of various EMG signal analysis methods. This paper provides researchers a good understanding of EMG signal and its analysis procedures. This knowledge will help them develop more powerful, flexible, and efficient applications.     

Flexible Analytic Wavelet Transform Based Features for Physical Action Identification Using sEMG Signals

ObjectivesElectromyography (EMG) is recording of the electrical activity produced by skeletal muscles. The classification of the EMG signals for different physical actions can be useful in restoring some or all of the lost motor functionalities in these individuals. Accuracy in classifying the EMG signal indicates efficient control of prosthesis.Material and methodsThe flexible analytic wavelet transform (FAWT) is used for classification of surface electromyography (sEMG) signals for identification of physical actions. FAWT is an efficient method for decomposition of sEMG signal into eight sub-bands, features namely neg-entropy, mean absolute value (MAV), variance (VAR), modified mean absolute value type 1 (MAV1), waveform length (WL), simple square integral (SSI), Tsallis entropy, integrated EMG (IEMG) are extracted from the sub-bands. Extracted features are fed into an extreme learning machine (ELM) classifier with sigmoid activation function.ResultsComprehensive experiments are conducted on the input sEMG signals and the accuracy, sensitivity and specificity scores are used for performance measurement. Experiments showed that among all sub-bands, the seventh sub-band provided the best performance where the recorded accuracy, sensitivity and specificity values were 99.36%, 99.36% and 99.93%, respectively. The comparison results showed best efficiency of proposed method as compared to other methods on the same dataset.ConclusionThis paper investigates the usage of the FAWT and ELM on sEMG signal classification. The results show that the proposed method is quite efficient in classification of the sEMG signals. It is also observed that the seventh sub-band of the FAWT provides the best discrimination property. In the future works, recent wavelet transform methods will be used for improving the classification performance.     

Modeling nonlinear errors in surface electromyography due to baseline noise: A new methodology

The surface electromyographic (EMG) signal is often contaminated by some degree of baseline noise. It is customary for scientists to subtract baseline noise from the measured EMG signal prior to further analyses based on the assumption that baseline noise adds linearly to the observed EMG signal. The stochastic nature of both the baseline and EMG signal, however, may invalidate this assumption. Alternately, “true” EMG signals may be either minimally or nonlinearly affected by baseline noise. This information is particularly relevant at low contraction intensities when signal-to-noise ratios (SNR) may be lowest. Thus, the purpose of this simulation study was to investigate the influence of varying levels of baseline noise (approximately 2–40% maximum EMG amplitude) on mean EMG burst amplitude and to assess the best means to account for signal noise. The simulations indicated baseline noise had minimal effects on mean EMG activity for maximum contractions, but increased nonlinearly with increasing noise levels and decreasing signal amplitudes. Thus, the simple baseline noise subtraction resulted in substantial error when estimating mean activity during low intensity EMG bursts. Conversely, correcting EMG signal as a nonlinear function of both baseline and measured signal amplitude provided highly accurate estimates of EMG amplitude. This novel nonlinear error modeling approach has potential implications for EMG signal processing, particularly when assessing co-activation of antagonist muscles or small amplitude contractions where the SNR can be low.     

Analysis of phasic and tonic electromyographic signal characteristics: Electromyographic synthesis and comparison of novel morphological and linear-envelope approaches

The pattern of tonic and phasic components in an EMG signal reflects the underlying behaviour of the central nervous system (CNS) in controlling the musculature. One avenue for gaining a better understanding of this behaviour is to seek a quantitative characterisation of these phasic and tonic components. We propose that these signal characteristics can range between unvarying, tonic and intermittent, phasic activation through a continuum of EMG amplitude modulation. In this paper, we present two new algorithms for quantifying amplitude modulation: a linear-envelope approach, and a mathematical morphology approach. In addition we present an algorithm for synthesising EMG signals with known amplitude modulation. The efficacy of the synthesis algorithm is demonstrated using real EMG data. We present an evaluation and comparison of the two algorithms for quantifying amplitude modulation based on synthetic data generated by the proposed synthesis algorithm. The results demonstrate that the EMG synthesis parameters represent 91.9% and 96.2% of the variance of linear-envelopes extracted from lumbo-pelvic muscle EMG signals collected from subjects performing a repetitive-movement task. This depended, however, on the muscle and movement-speed considered (F = 4.02, p < 0.001). Coefficients of determination between input and output amplitude modulation variables were used to quantify the accuracy of the linear-envelope and morphological signal processing algorithms. The linear-envelope algorithm exhibited higher coefficients of determination than the most accurate morphological approach (and hence greater accuracy, T = 8.16, p < 0.001). Similarly, the standard deviation of the coefficients of determination was 1.691 times smaller (p < 0.001). This signal processing algorithm represents a novel tool for the quantification of amplitude modulation in continuous EMG signals and can be used in the study of CNS motor control of the musculature in repetitive-movement tasks.     

Decomposition of indwelling EMG signals.

Decomposition of indwelling electromyographic (EMG) signals is challenging in view of the complex and often unpredictable behaviors and interactions of the action potential trains of different motor units that constitute the indwelling EMG signal. These phenomena create a myriad of problem situations that a decomposition technique needs to address to attain completeness and accuracy levels required for various scientific and clinical applications. Starting with the maximum a posteriori probability classifier adapted from the original precision decomposition system (PD I) of LeFever and De Luca (25, 26), an artificial intelligence approach has been used to develop a multiclassifier system (PD II) for addressing some of the experimentally identified problem situations. On a database of indwelling EMG signals reflecting such conditions, the fully automatic PD II system is found to achieve a decomposition accuracy of 86.0% despite the fact that its results include low-amplitude action potential trains that are not decomposable at all via systems such as PD I. Accuracy was established by comparing the decompositions of indwelling EMG signals obtained from two sensors. At the end of the automatic PD II decomposition procedure, the accuracy may be enhanced to nearly 100% via an interactive editor, a particularly significant fact for the previously indecomposable trains.     

An Improved SSA-Based Technique for EMG Removal from ECG

Background

The electrocardiogram (ECG) signals provide important information about the heart electrical activities in medical and diagnostic applications. This signal may be contaminated by different types of noises. One of the noise types which has a considerable overlap with the ECG signals in frequency domain is electromyogram (EMG). Among the exciting approaches for de-noising the ECG signals, those based on singular spectrum analysis (SSA) are popular.

Effects of minimum sampling rate and signal reconstruction on surface electromyographic signals.

Signal oversampling is frequently used to prevent distortion in time-series representations. When sampling at rates just above the Nyquist critical frequency (f(c)), Shannon's reconstruction theorem provides an alternative means of circumventing this problem. The purpose of this study was to determine whether surface electromyographic (EMG) data is compromised when sampled just above f(c) and whether Shannon's reconstruction theorem can correct these deficiencies, if present. Brief isometric elbow flexion contractions were performed at 100%, 80%, 50%, 25%, 10%, 5% and 2.5% maximum voluntary contraction (MVC) followed by one trial of random cyclic isometric contractions and relaxations. The 2kHz signal was resampled at 1kHz and the 2 and 1kHz signals were reconstructed (2RS and 1RS, respectively) to a sampling rate of 20kHz. Data were full-wave rectified (FWR) and low-pass filtered (LE). Peak amplitudes of the FWR and LE signals, average EMG (AEMG) amplitudes of the FWR and LE signals, mean power frequency of the raw data, number of gaps, and mean gap time of the LE signals were calculated. Significant differences were present in peak EMG and AEMG measurements between the 1 and 2kHz, 2RS and 20kHz signals and occasionally between the 1RS and the 2kHz, 2RS and 20kHz signals. These differences, although statistically significant were quite small amounting to less than 0.5% MVC. No significant differences were found for the gaps parameters. The small differences seen, coupled with the processing time required for signal reconstruction, make oversampling as well as signal reconstruction in surface EMG measurements unnecessary.     

High density electromyography data of normally limbed and transradial amputee subjects for multifunction prosthetic control

Pattern recognition based control of powered upper limb myoelectric prostheses offers a means of extracting more information from the available muscles than conventional methods. By identifying repeatable patterns of muscle activity across multiple muscle sites rather than relying on independent EMG signals it is possible to provide more natural, reliable control of myoelectric prostheses. The purposes of this study were to (1) determine if participants can perform distinctive muscle activation patterns associated with multiple wrist and hand movements reliably and (2) to show that high density EMG can be applied individually to determine the electrode location of a clinically acceptable number of electrodes (maximally eight) to classify multiple wrist and hand movements reliably in transradial amputees. Eight normally limbed subjects (five female, three male) and four transradial amputee subjects (two traumatic and congenital) subjects participated in this study, which examined the classification accuracies of a pattern recognition control system. It was found that tasks could be classified with high accuracy (85-98%) with normally limbed subjects (10-13 tasks) and with amputees (4-6) tasks. In healthy subjects, reducing the number of electrodes to eight did not affect accuracy significantly when those electrodes were optimally placed, but did reduce accuracy significantly when those electrodes were distributed evenly. In the amputee subjects, reducing the number of electrodes up to 4 did not affect classification accuracy or the number of tasks with high accuracy, independent of whether those remaining electrodes were evenly distributed or optimally placed. The findings in healthy subjects suggest that high density EMG testing is a useful tool to identify optimal electrode sites for pattern recognition control, but its use in amputees still has to be proven. Instead of just identifying the electrode sites where EMG activity is strong, clinicians will be able to choose the electrode sites that provide the most important information for classification.     

Effects of EMG processing on biomechanical models of muscle joint systems: sensitivity of trunk muscle moments, spinal forces, and stability

Biomechanical models are in use to estimate parameters such as contact forces and stability at various joints. In one class of these models, surface electromyography (EMG) is used to address the problem of mechanical indeterminacy such that individual muscle activation patterns are accounted for. Unfortunately, because of the stochastical properties of EMG signals, EMG based estimates of muscle force suffer from substantial estimation errors. Recent studies have shown that improvements in muscle force estimation can be achieved through adequate EMG processing, specifically whitening and high-pass (HP) filtering of the signals. The aim of this paper is to determine the effect of such processing on outcomes of a biomechanical model of the lumbosacral joint and surrounding musculature. Goodness of fit of estimated muscle moments to net moments and also estimated joint stability significantly increased with increasing cut-off frequencies in HP filtering, whereas no effect on joint contact forces was found. Whitening resulted in moment estimations comparable to those obtained from optimal HP filtering with cut-off frequencies over 250 Hz. Moreover, compared to HP filtering, whitening led to a further increase in estimated joint-stability. Based on theoretical models and on our experimental results, we hypothesize that the processing leads to an increase in pick-up area. This then would explain the improvements from a better balance between deep and superficial motor unit contributions to the signal.     

Influence of amplitude cancellation on the accuracy of determining the onset of muscle activity from the surface electromyogram

The purpose of the study was to quantify the influence of amplitude cancellation on the accuracy of detecting the onset of muscle activity based on an analysis of simulated surface electromyographic (EMG) signals. EMG activity of a generic lower limb muscle was simulated during the stance phase of human gait. Surface EMG signals were generated with and without amplitude cancellation by summing simulated motor unit potentials either before (cancellation EMG) or after (no-cancellation EMG) the potentials had been rectified. The two sets of EMG signals were compared at forces of 30% and 80% of maximum voluntary contraction (MVC) and with various low-pass filter cut-off frequencies. Onset time was determined both visually and by an algorithm that identified when the mean amplitude of the signal within a sliding window exceeded a specified standard deviation (SD) above the baseline mean. Onset error was greater for the no-cancellation conditions when determined automatically and by visual inspection. However, the differences in onset error between the two cancellation conditions appear to be clinically insignificant. Therefore, amplitude cancellation does not appear to limit the ability to detect the onset of muscle activity from the surface EMG.     

EMG-force modeling using parallel cascade identification

Measuring force production in muscles is important for many applications such as gait analysis, medical rehabilitation, and human-machine interaction. Substantial research has focused on finding signal processing and modeling techniques which give accurate estimates of muscle force from the surface-recorded electromyogram (EMG). The proposed methods often do not capture both the nonlinearities and dynamic components of the EMG-force relation. In this study, parallel cascade identification (PCI) is used as a dynamic estimation tool to map surface EMG recordings from upper-arm muscles to the induced force at the wrist. PCI mapping involves generating a parallel connection of a series of linear dynamic and nonlinear static blocks. The PCI model parameters were initialized to obtain the best force prediction. A comparison between PCI and a previously published Hill-based orthogonalization scheme, that captures physiological behaviour of the muscles, has shown 44% improvement in force prediction by PCI (averaged over all subjects in relative-mean-square sense). The improved performance is attributed to the structural capability of PCI to capture nonlinear dynamic effects in the generated force.     

Frequency characteristics of signals and instrumentation: implication for EMG biofeedback studies

Signals can be analyzed in either the time or frequency domain. In the time domain, the analysis consists of manipulating and measuring one or more characteristics of the signal that may vary with time. One can, for instance, rectify a signal, filter it, calculate its mean value, display the histogram of its amplitude, and so forth. Frequency analysis is less well understood because it requires a lengthy mathematical treatment most easily done by computer. However, it gives exclusive information on a signal. For instance, when the frequency content of a signal is known, it is easy to specify which characteristics an amplifier must have in order to amplify the signal without distortion, or to set the cutoff frequencies of filters to eliminate noise. Also, in many circumstances, frequency spectra are more easily interpreted than the original raw data. Such is the case with the EMG where the random aspect of the signal makes some form of processing (i.e., rectification, filtering, etc.) necessary, but not always as meaningful as we would like. Thus we present here the principal characteristics of frequency analysis, and discuss its usefulness in analyzing EMG signals and its application to biofeedback, clinical practice, and research.     

A Real-Time Capable Linear Time Classifier Scheme for Anticipated Hand Movements Recognition from Amputee Subjects Using Surface EMG Signals

Anticipated hand movements of amputee subjects are considered difficult to classify using only Electromyogram (EMG) signals and machine learning techniques. For a long time, classifying such s-EMG signals have been considered as a non-linear problem, and the problem of signal sparsity has not been given detailed attention in a large set of action classes. For addressing these problems, this paper is proposing a linear-time classifier termed as Random Fourier Mapped Collaborative Representation with distance weighted Tikhonov regularization matrix (RFMCRT). RFMCRT attempts to tackle the non-linear problem via Random Fourier Features and sparsity issue with collaborative representation. The projection error of Random Fourier Features is reduced by projecting to the same dimension as the original feature space and later finding the collaborative representation, with an optional non-negative constraint (RFMNNCRT). The proposed two classifiers were tested with time-domain features computed from the EMG signals obtained from NINAPRO databases using a non-overlapping sliding window size of 256 ms. Due to the random nature of our proposed classifiers, this paper has computed the average and worst-case performance for 50 trials and compared them with other reported classifiers. The results show that RFMNNCRT (average case) outperformed state-of-the-art classifiers with the accuracy of 93.44% for intact subjects and 55.67% for amputee subjects. In the worst-case situation, RFMCRT achieves considerable performance for the same, with the reported accuracy of 91.55% and 50.27% respectively. Our proposed classifier guarantees acceptable levels of accuracy for large classes of hand movements and also maintains good computational efficiency in comparison to LDA and SVM.     

Correlations between short-time Fourier- and continuous wavelet transforms in the analysis of localized back and hip muscle fatigue during isometric contractions.

The aims of the current study were to examine the stationarities of surface electromyographic (EMG) signals obtained from eight bilateral back and hip muscles during a modified Biering-Sørensen test, and to investigate whether short-time Fourier (STFT) and continuous wavelet transforms (CWT) provided similar information with regard to EMG spectral parameters in the analysis of localized muscle fatigue. Twenty healthy subjects participated in the study after giving their informed consent. Reverse arrangement tests showed that 91.6% of the EMG signal epochs demonstrated no significant trends (all p > 0.05), meaning 91.6% of the EMG signal epochs could be considered as stationary signals. Pearson correlation coefficients showed that STFT and CWT in general provide similar information with respect to the EMG spectral variables during isometric back extensions, and as a consequence STFT can still be used.     

Estimation of monopolar signals from sphincter muscles and removal of common mode interference

The detection of surface electromyogram (EMG) by multi-electrode systems is applied in many research studies. The signal is usually recorded by means of spatial filters (linear combination of the potential under at least two electrodes) with vanishing sum of weights. Nevertheless, more information could be extracted from monopolar signals measured with respect to a reference electrode away from the muscle. Under certain conditions, surface EMG signal along a curve parallel to the fibre path has zero mean (property approximately satisfied when EMG is sampled by an array of electrodes that covers the entire support of the signal in space). This property allows estimating monopolar from single differential (SD) signals by pseudoinversion of the matrix relating monopolar to SD signals. The method applies to EMG signals from the external anal sphincter muscle, recorded using a specific cylindrical probe with an array of electrodes located along the circular path of the fibres. The performance of the algorithm for the estimation of monopolar from SD signals is tested on simulated signals. The estimation error of monopolar signals decreases by increasing the number of channels. Using at least 12 electrodes, the estimation error is negligible. The method applies to single fibre action potentials, single motor unit action potentials, and interference signals.The same method can also be applied to reduce common mode interference from SD signals from muscles with rectilinear fibres. In this case, the last SD channel defined as the difference between the potentials of the last and the first electrodes must be recorded, so that the sum of all the SD signals vanishes. The SD signals estimated from the double differential signals by pseudoinvertion are free of common mode.     

Discrimination of Parkinsonian tremor from essential tremor by implementation of a wavelet-based soft-decision technique on EMG and accelerometer signals

A wavelet-decomposition with soft-decision algorithm is used to estimate an approximate power-spectral density (PSD) of both accelerometer and surface EMG signals for the purpose of discrimination of Parkinson tremor from essential tremor. A soft-decision wavelet-based PSD estimation is used with 256 bands for a signal sampled at 800 Hz. The sum of the entropy of the PSD in band 6 (7.8125–9.375 Hz) and band 11 (15.625–17.1875 Hz) is used as a classification factor. The data has been recorded for diagnostic purposes in the Department of Neurology of the University of Kiel, Germany. Two sets of data are used. The training set, which consists of 21 essential-tremor (ET) subjects and 19 Parkinson-disease (PD) subjects, is used to obtain the threshold value of the classification factor differentiating between the two subjects. The test data set, which consists of 20 ET and 20 PD subjects, is used to test the technique and evaluate its performance. A “voting” between three results obtained from accelerometer signal and two EMG signals is applied to obtain the final discrimination. A total accuracy of discrimination of 85% is obtained.     

Automatic decomposition of multichannel intramuscular EMG signals

We describe an automatic algorithm for decomposing multichannel EMG signals into their component motor unit action potential (MUAP) trains, including signals from widely separated recording sites in which MUAPs exhibit appreciable interchannel offset and jitter. The algorithm has two phases. In the clustering phase, the distinct, recurring MUAPs in each channel are identified, the ones that correspond to the same motor units are determined by their temporal relationships, and multichannel templates are computed. In the identification stage, the MUAP discharges in the signal are identified using matched filtering and superimposition resolution techniques. The algorithm looks for the MUAPs with the largest single channel components first, using matches in one channel to guide the search in other channels, and using information from the other channels to confirm or refute each identification. For validation, the algorithm was used to decompose 10 real 6-to-8-channel EMG signals containing activity from up to 25 motor units. Comparison with expert manual decomposition showed that the algorithm identified more than 75% of the total 176 MUAP trains with an accuracy greater than 95%. The algorithm is fast, robust, and shows promise to be accurate enough to be a useful tool for decomposing multichannel signals. It is freely available at http://emglab.stanford.edu.     

Feasibility of using EMG driven neuromusculoskeletal model for prediction of dynamic movement of the elbow.

Neuromusculoskeletal (NMS) modeling is a valuable tool in orthopaedic biomechanics and motor control research. To evaluate the feasibility of using electromyographic (EMG) signals with NMS modeling to estimate individual muscle force during dynamic movement, an EMG driven NMS model of the elbow was developed. The model incorporates dynamical equation of motion of the forearm, musculoskeletal geometry and musculotendon modeling of four prime elbow flexors and three prime elbow extensors. It was first calibrated to two normal subjects by determining the subject-specific musculotendon parameters using computational optimization to minimize the root mean square difference between the predicted and measured maximum isometric flexion and extension torque at nine elbow positions (0-120 degrees of flexion with an increment of 15 degrees ). Once calibrated, the model was used to predict the elbow joint trajectories for three flexion/extension tasks by processing the EMG signals picked up by both surface and fine electrodes using two different EMG-to-activation processing schemes reported in the literature without involving any trajectory fitting procedures. It appeared that both schemes interpreted the EMG somewhat consistently but their prediction accuracy varied among testing protocols. In general, the model succeeded in predicting the elbow flexion trajectory in the moderate loading condition but over-drove the flexion trajectory under unloaded condition. The predicted trajectories of the elbow extension were noted to be continuous but the general shape did not fit very well with the measured one. Estimation of muscle activation based on EMG was believed to be the major source of uncertainty within the EMG driven model. It was especially so apparently when fine wire EMG signal is involved primarily. In spite of such limitation, we demonstrated the potential of using EMG driven neuromusculoskeletal modeling for non-invasive prediction of individual muscle forces during dynamic movement under certain conditions.