A fully connected deep learning approach to upper limb gesture recognition in a secure FES rehabilitation environment
Qi Liu,Xueyan Wu,Yinghang Jiang,Xiaodong Liu,Yonghong Zhang,Xiaolong Xu,Lianyong Qi
DOI: https://doi.org/10.1002/int.22383
IF: 8.993
2021-02-16
International Journal of Intelligent Systems
Abstract:<p>Stroke is one of the leading causes of death and disability in the world. The rehabilitation of Patients' limb functions has great medical value, for example, the therapy of functional electrical stimulation (FES) systems, but suffers from effective rehabilitation evaluation. In this paper, six gestures of upper limb rehabilitation were monitored and collected using microelectromechanical systems sensors, where data stability was guaranteed using data preprocessing methods, that is, deweighting, interpolation, and feature extraction. A fully connected neural network has been proposed investigating the effects of different hidden layers, and determining its activation functions and optimizers. Experiments have depicted that a three‐hidden‐layer model with a softmax function and an adaptive gradient descent optimizer can reach an average gesture recognition rate of 97.19%. A stop mechanism has been used via recognition of dangerous gesture to ensure the safety of the system, and the lightweight cryptography has been used via hash to ensure the security of the system. Comparison to the classification models, for example, <i>k</i>‐nearest neighbor, logistic regression, and other random gradient descent algorithms, was conducted to verify the outperformance in recognition of upper limb gesture data. This study also provides an approach to creating health profiles based on large‐scale rehabilitation data and therefore consequent diagnosis of the effects of FES rehabilitation.</p>
computer science, artificial intelligence