A robust myoelectric pattern recognition using online sequential extreme learning machine for finger movement classification

Using Antonyan Vardan Transform and Extreme Learning Machines for Accurate sEMG Signal Classification

In this paper, we present an evaluation of an adaptation of the Antonyan Vardan Transform (AVT) used in combination with an Extreme Learning Machines (ELM) classifier to process surface electromyography (sEMG) data used to classify six finger movements and a rest state. A total of 12 assays formed by three repetitions performed by four volunteers is analyzed. Additionally, a sample-by-sample output label comparison was performed to make a more comprehensive analysis of the system which was tested on a PC and embedded on a Rasp.berry Pi platform. Compared to literature papers, our system was capable to match or outperform similar solutions even using a simpler model, reaching mean accuracy rates above 94.

Adaptive myoelectric pattern recognition for arm movement in different positions using advanced online sequential extreme learning machine

The performance of the myoelectric pattern recognition system sharply decreases when working in various limb positions. The issue can be solved by cumbersome training procedure that can anticipate all possible future situations. However, this procedure will sacrifice the comfort of the user. In addition, many unpredictable scenarios may be met in the future. This paper proposed a new adaptive myoelectric pattern recognition using advance online sequential extreme learning (AOS-ELM) for classification of the hand movements to five different positions. AOS-ELM is an improvement of OS-ELM that can verify the adaptation validity using entropy. The proposed adaptive MPR was able to classify eight different classes from eleven subjects by accuracy of 95.42 % using data from one position. After learning the data from whole positions, the performance of the proposed system is 86.13 %. This performance was better than the MPR that employed original OS-ELM, but it was worse than the MPR that utilized the batch classifiers. Nevertheless, the adaptation mechanism of AOS-ELM is preferred in the real-time application.

Proof of Concept of an Online EMG-Based Decoding of Hand Postures and Individual Digit Forces for Prosthetic Hand Control

INTRODUCTION

Options currently available to individuals with upper limb loss range from prosthetic hands that can perform many movements, but require more cognitive effort to control, to simpler terminal devices with limited functional abilities. We attempted to address this issue by designing a myoelectric control system to modulate prosthetic hand posture and digit force distribution.

METHODS

We recorded surface electromyographic (EMG) signals from five forearm muscles in eight able-bodied subjects while they modulated hand posture and the flexion force distribution of individual fingers. We used a support vector machine (SVM) and a random forest regression (RFR) to map EMG signal features to hand posture and individual digit forces, respectively. After training, subjects performed grasping tasks and hand gestures while a computer program computed and displayed online feedback of all digit forces, in which digits were flexed, and the magnitude of contact forces. We also used a commercially available prosthetic hand, the i-Limb (Touch Bionics), to provide a practical demonstration of the proposed approach's ability to control hand posture and finger forces.

RESULTS

Subjects could control hand pose and force distribution across the fingers during online testing. Decoding success rates ranged from 60% (index finger pointing) to 83-99% for 2-digit grasp and resting state, respectively. Subjects could also modulate finger force distribution.

DISCUSSION

This work provides a proof of concept for the application of SVM and RFR for online control of hand posture and finger force distribution, respectively. Our approach has potential applications for enabling in-hand manipulation with a prosthetic hand.