sEMG-Based Gesture Recognition with Convolution Neural Networks

Abstract

Abstract:The traditional classification methods for limb motion recognition based on sEMG have been deeply researched and shown promising results. However, information loss during feature extraction reduces the recognition accuracy. To obtain higher accuracy, the deep learning method was introduced. In this paper, we propose a parallel multiple-scale convolution architecture. Compared with the state-of-art methods, the proposed architecture fully considers the characteristics of the sEMG signal. Larger sizes of kernel filter than commonly used in other CNN-based hand recognition methods are adopted. Meanwhile, the characteristics of the sEMG signal, that is, muscle independence, is considered when designing the architecture. All the classification methods were evaluated on the NinaPro database. The results show that the proposed architecture has the highest recognition accuracy. Furthermore, the results indicate that parallel multiple-scale convolution architecture with larger size of kernel filter and considering muscle independence can significantly increase the classification accuracy.

Surface-EMG based Wrist Kinematics Estimation using Convolutional Neural Network

“…According to previous research, two types of sEMG images have been most widely applied, i.e. the 2D time domain image [9] and 2D spectrum image [10]. The latter is commonly converted from the former via fast Fourier transform (FFT).…”

“…In our experiments hyper-parameters of CNN were firstly identified referring to studies in PR scheme [8][9][10] and then determined via empirical manual tuning. The final settings were fixed to experiments in all subjects.…”

“…This process, also known as feature learning, is considered to be more efficient than classical feature engineering. Therefore, explorations on DL, mainly the convolutional neural networks (CNN), are also becoming a trend in myoelectric control [6,[8][9][10]. However, compared with numerous literatures in PR approaches, studies of DL for regressionbased schemes are quite inadequate.…”

Force Estimation Based on sEMG using Wavelet Analysis and Neural Network

“…Sun et al based on DS evidence theory and effectively improved the recognition efficiency of human upper limb movements through sEMG signal and Kinect multi-sensor data fusion [13]. The method of deep learning has been well implemented based on image gesture recognition, so some research attempts to introduce deep learning methods based on surface EMG signals, and achieved certain results [14]- [16]. Accuracy, diversity and real-time are essential for complex applications such as rehabilitation and human-computer interaction.…”

SEMG based Intention Identification of Complex Hand Motion using Nonlinear Time Series Analysis

“…Xiaochuan Yin and Qijun Chen [8] presented a nonlinear time alignment method with deep autoencoder to extract spatio-temporal features for human action recognition. In a more recent study Ding et al [9] proposed a parallel multiple-scale convolution architecture for gesture recognition by using convolution neural networks. The results show that with in some reasonable range of features, human hand motions could be recognized with a satisfactory recognition rate from SEMG signal, but the approach suffers from some issues such as training of data and classification errors.…”