User-Independent EMG Gesture Recognition Method Based on Adaptive Learning

Abstract

In a gesture recognition system based on surface electromyogram (sEMG) signals, the recognition model established by existing users cannot directly generalize to the across-user scenarios due to the individual variability of sEMG signals. In this article, we propose an adaptive learning method to handle the problem. The muscle synergy is chosen as the feature vector because it can well-characterize the neural origin of movement. The initial train set is composed of representative samples extracted from the synergy matrix of the existing user. When the new users use the system, the label is obtained by the adaptive K nearest neighbor algorithm (KNN). The recognition process does not require the pre-experiment for new users due to the introduction of adaptive learning strategy, namely, the qualified data and the label of new user data evaluated by a risk evaluator are used to update the train set and KNN weights, so as to adapt to the new users. We have tested the algorithm in DB1 and DB5 of Ninapro databases. The average recognition accuracy is 68.04, 73.35, and 83.05% for different types of gestures, respectively, achieving the effects of the user-dependent method. Our study can avoid the re-training steps and the recognition performance will improve with the increased frequency of uses, which will further facilitate the widespread implementation of sEMG control systems using pattern recognition techniques.

Keywords: adaptive learning; muscle synergy; pattern recognition; surface electromyogram; user-independent.

Figure 1

Acquisition setups and the placement: (A) DB1 OttoBock MyoBock 13E200, (B) DB5 Double Myo armband (Pizzolato et al., 2017).

Figure 2

Twenty-two daily gesture: (A) four basic movements of the wrist and hand (Group one), (B) eight isometric, isotonic hand configurations (Group two), and (C) 12 basic movements of the fingers (Group three) (Atzori et al., 2014b).

Figure 3

The framework for user-independent gesture recognition.

Figure 4

The process of adapting K-value.

Figure 5

Schematic diagrams of updating the weight.

Figure 6

Envelope and activity segment.

Figure 7

Relationship between average recognition accuracy and test times.

Figure 8

Changes in channel weights at different stages of (A) the existing samples, (B) the initial train set, (C) the updated train set, and (D) test samples.

Figure 9

Comparison of significance among different schemes (*p < 0.05).