Feature Extraction and Selection for Myoelectric Control Based on Wearable EMG Sensors

Abstract

Specialized myoelectric sensors have been used in prosthetics for decades, but, with recent advancements in wearable sensors, wireless communication and embedded technologies, wearable electromyographic (EMG) armbands are now commercially available for the general public. Due to physical, processing, and cost constraints, however, these armbands typically sample EMG signals at a lower frequency (e.g., 200 Hz for the Myo armband) than their clinical counterparts. It remains unclear whether existing EMG feature extraction methods, which largely evolved based on EMG signals sampled at 1000 Hz or above, are still effective for use with these emerging lower-bandwidth systems. In this study, the effects of sampling rate (low: 200 Hz vs. high: 1000 Hz) on the classification of hand and finger movements were evaluated for twenty-six different individual features and eight sets of multiple features using a variety of datasets comprised of both able-bodied and amputee subjects. The results show that, on average, classification accuracies drop significantly ( p.

Keywords: EMG; L-moments; electromyography; feature extraction; pattern recognition; prosthesis; sampling rate; wearable sensor.

Figure 1

Examples of EMG data sampled at two different sampling rates (1000 Hz vs. 200 Hz) in: (a) time domain; and (b) frequency domain. Samples are acquired from the first EMG channel of Subject 1 during the hook grip motion from Database 3.

Figure 2

Differences in EMG patterns between using: (left) a 1000 Hz sampling rate; and (right) a 200 Hz sampling rate. ZC features are extracted from two different EMG channels (6 and 7) during thumb flexion (green circle markers and solid lines) and index flexion (blue square markers and dashed lines). Samples are from Subject 1 of Database 3.

Figure 3

The classification performance of the multi-feature sets sequentially selected by the SFS method using the two different sampling rates (1000 Hz vs. 200 Hz) for Dataset 3.

Figure 4

The classification performance of the MS1 feature set using an SVM classifier for multiple EMG datasets (Datasets 1–3) when: (a) the training set is acquired from multiple conditions and the testing set is acquired from a different condition, unseen during training; and (b) the training set is acquired from a single condition and the testing set is acquired from multiple conditions. Mean values across subjects and classes of motion represented with bars and their standard deviations with error bars. Asterisks indicate significant difference ($p<0.05$).

Figure 5

The classification performance of 26 individual features of 12 motion classes in Exercise A, 17 motion classes in Exercise B, and 23 motion classes in Exercise C using SVM classification of myoelectric signals recorded from wearable EMG sensors from Dataset 4 (with a sampling rate of 200 Hz).

Figure 6

The classification performance of eight multi-feature sets (MS1–MS8) and two newly proposed multi-feature sets (TD4 and TD9) in the classification of 12 motion classes in Exercise A, 17 motion classes in Exercise B, and 23 motion classes in Exercise C using SVM classification of myoelectric signals recorded from wearable EMG sensors from Dataset 4 (with a sampling rate of 200 Hz).