Feature extraction of the first difference of EMG time series for EMG pattern recognition

This paper demonstrates the utility of a differencing technique to transform surface EMG signals measured during both static and dynamic contractions such that they become more stationary. The technique was evaluated by three stationarity tests consisting of the variation of two statistical properties, i.e., mean and standard deviation, and the reverse arrangements test. As a result of the proposed technique, the first difference of EMG time series became more stationary compared to the original measured signal. Based on this finding, the performance of time-domain features extracted from raw and transformed EMG was investigated via an EMG classification problem (i.e., eight dynamic motions and four EMG channels) on data from 18 subjects. The results show that the classification accuracies of all features extracted from the transformed signals were higher than features extracted from the original signals for six different classifiers including quadratic discriminant analysis. On average, the proposed differencing technique improved classification accuracies by 2–8%.     

A feasibility study on the use of anthropometric variables to make muscle–computer interface more practical

High classification accuracy has been achieved for muscle–computer interfaces (MCIs) based on surface electromyography (EMG) recognition in many recent works with an increasing number of discriminated movements. However, there are many limitations to use these interfaces in the real-world contexts. One of the major problems is compatibility. Designing and training the classification EMG system for a particular individual user is needed in order to reach high accuracy. If the system can calibrate itself automatically/semi-automatically, the development of standard interfaces that are compatible with almost any user could be possible. Twelve anthropometric variables, a measurement of body dimensions, have been proposed and used to calibrate the system in two different ways: a weighting factor for a classifier and a normalizing value for EMG features. The experimental results showed that a number of relationships between anthropometric variables and EMG time-domain features from upper-limb muscles and movements are statistically strong (average r=0.71?0.80) and significant (p<0.05). In this paper, the feasibility to use anthropometric variables to calibrate the EMG classification system is shown obviously and the proposed calibration technique is suggested to further improve the robustness and practical use of MCIs based on EMG pattern recognition.     

Fractal analysis features for weak and single-channel upper-limb EMG signals

Electromyography (EMG) signals are the electrical manifestations of muscle contractions. EMG signals may be weak or at a low level when there is only a small movement in the major corresponding muscle group or when there is a strong movement in the minor corresponding muscle group. Moreover, in a single-channel EMG classification identifying the signals may be difficult. However, weak and single-channel EMG control systems offer a very convenient way of controlling human–computer interfaces (HCIs). Identifying upper-limb movements using a single-channel surface EMG also has a number of rehabilitation and HCI applications. The fractal analysis method, known as detrended fluctuation analysis (DFA), has been suggested for the identification of low-level muscle activations. This study found that DFA performs better in the classification of EMG signals from bifunctional movements of low-level and equal power as compared to other successful and commonly used features based on magnitude and other fractal techniques.     

Feature reduction and selection for EMG signal classification

Feature extraction is a significant method to extract the useful information which is hidden in surface electromyography (EMG) signal and to remove the unwanted part and interferences. To be successful in classification of the EMG signal, selection of a feature vector ought to be carefully considered. However, numerous studies of the EMG signal classification have used a feature set that have contained a number of redundant features. In this study, most complete and up-to-date thirty-seven time domain and frequency domain features have been proposed to be studied their properties. The results, which were verified by scatter plot of features, statistical analysis and classifier, indicated that most time domain features are superfluity and redundancy. They can be grouped according to mathematical property and information into four main types: energy and complexity, frequency, prediction model, and time-dependence. On the other hand, all frequency domain features are calculated based on statistical parameters of EMG power spectral density. Its performance in class separability viewpoint is not suitable for EMG recognition system. Recommendation of features to avoid the usage of redundant features for classifier in EMG signal classification applications is also proposed in this study.     

EMG feature evaluation for improving myoelectric pattern recognition robustness

In pattern recognition-based myoelectric control, high accuracy for multiple discriminated motions is presented in most of related literature. However, there is a gap between the classification accuracy and the usability of practical applications of myoelectric control, especially the effect of long-term usage. This paper proposes and investigates the behavior of fifty time-domain and frequency-domain features to classify ten upper limb motions using electromyographic data recorded during 21 days. The most stable single feature and multiple feature sets are presented with the optimum configuration of myoelectric control, i.e. data segmentation and classifier. The result shows that sample entropy (SampEn) outperforms other features when compared using linear discriminant analysis (LDA), a robust classifier. The averaged test classification accuracy is 93.37%, when trained in only initial first day. It brings only 2.45% decrease compared with retraining schemes. Increasing number of features to four, which consists of SampEn, the fourth order cepstrum coefficients, root mean square and waveform length, increase the classification accuracy to 98.87%. The proposed techniques achieve to maintain the high accuracy without the retraining scheme. Additionally, this continuous classification allows the real-time operation.