A novel feature reduction method for real-time EMG pattern recognition system

Abstract

This paper proposes a novel feature reduction approach for real-time electromyogram (EMG) pattern recognition. This study extracts time and frequency information by wavelet packet transform (WPT) coefficients and uses the node energy as the feature to overcome the translation-invariant property of WPT. Then the non-parametric discriminant analysis (NDA) is used for feature reduction. Because of some inherent properties of the packet node energy, the within-class scatter matrix is usually singular in this approach, which makes feature project unavailable. To solve this problem, a recursive algorithm is proposed to discard some feature components that lead to singularity and contain relatively less discriminant information. Finally, the support vector machine (SVM) is used as the classifier and gives the recognition result. The corresponding pattern of the action could be recognized in a millisecond (ms). The experimental results show that the proposed method has strong robustness and good real-time performance.

Real-Time Embedded EMG Signal Analysis for Wrist-Hand Pose Identification

“…In placed-sensor approaches, the acquired EMG signals undergo time-frequency analysis, enabled by the differing frequency characteristics of the EMGs emanating from the muscles over which they are placed. These techniques include Fourier transform [15], Wavelet Transform (WT), and Wavelet Packet Transform (WPT) [8], [9], [14], [16]. For instance, the work in [16] derives a feature vector, using Discrete Wavelet Transform (DWT), which is then projected via unsupervised Principal Component Analysis (PCA) [17] to detect six different hand poses with 97.50% accuracy.…”

“…The work in [8] combines PCA with a Self-Organizing Feature Map (SOFM) to enable a combined linear/non-linear unsupervised feature projection method. If linear supervised projection is required Linear Discriminant Analysis (LDA) is preferred due to its class separability [9], [19], [20] and its superior performance as compared to PCA/SOFM [20]. The final step in the EMG processing sequence is inference of movements from the reduceddimensionality feature vectors.…”

“…The final step in the EMG processing sequence is inference of movements from the reduceddimensionality feature vectors. Various classifiers have been analysed in this context, including Support Vector Machines (SVMs) [9], [10], [18], [21], [22], bayesian classifiers [14], gaussian mixture models [14], hidden markov models [14], fuzzy classifiers [14] and Multilayer Perceptrons (MLPs) [8], [16], [20], [23].…”

“…Feature projection creates an appropriate subset of new features from an original feature set such that the learning criterion is optimized [19]. For EMG signal analysis, PCA and LDA have been popularly used for this purpose [9], [19], [20]. The most effective choice will be that which, when combined with the proposed classifier, enables the highest classification accuracy.…”

“…The most effective choice will be that which, when combined with the proposed classifier, enables the highest classification accuracy. Similarly, a set of alternative classifiers may also be considered, including SVM [9], [22], k-Nearest Neighbour (k-NN) and MLP [8], [14], [30]. The work in [8], [20] has shown that employing MLP to classify time-frequency feature vectors is able to differentiate the nine classes of wrist-hand movements desired in the proposed approach, and as such a similar approach is considered here.…”