Surface Emg for Multi-Pattern Recognition With Sensory Feedback Controller of Hand Prosthesis System

Bp–ar-Based Human Joint Angle Estimation Using Multi-Channel Semg

“…The BP neural network has been widely used in the field of sEMG-human motion recognition [16]- [18]. Well arranging the structure and transition functions can make the BP network approximate any non-linear function on a compact set.…”

“…The ANN is a mathematical model used to model complex relationships between inputs and outputs or to find patterns in data [28]. One kind of the most widely used ANN is BP neural network, whose basic function is the approximation of non-linear continuous rational function and has been used for sEMG recognition [16]- [18]. The problem of multi-channel sEMG-knee joint angle estimation can be described as follows:…”

“…This work focuses on providing a reliable real-time human limb motion estimation method using sEMG for active rehabilitation training system. Nowadays, in active rehabilitation training research, the training functions are programmed according to human active motion intention, and sEMG signal is one of the main kinds of biosignals used for human motion pattern recognition, such as limb postures [10]- [13], muscle force [14]- [16], or motion state estimation (such as joint angle, torque) [17]- [20], etc. First, sEMG is used as switching signals to recognize the kind of activities and then program the training system to complete a fixed motor trajectory [12].…”

“…This kind of sEMG application usually has relatively higher recognition accuracy; however, because the robot arms' trajectories are fixed according to motion kinds, it cannot satisfy the variety of human movements. The mathematical models for sEMG-muscle force are always very complicated [14]- [16], some parameters are different according to body physical conditions (such as muscle fatigue, fat content) and environments, and also hard to measure, thus the practical applications are limited.…”

“…For instance, Shrirao et al [17] used neural network models to predict finger joint angles based on sEMG of extensor digitorum superficialis muscle during the flexion-extension cycle. The experiment showed a relatively small RMS (root mean square) error; however, it also caused time delay by 200 ms. Due to the serious disturbance and interference with the recorded sEMG signals, the time delay always seems to be unavoidable [16]- [18] and then brings negative influence to real-time control. Our team tried to use a 20-order BP neural network model to describe the relationship between sEMG time series and human leg joint angles [19], and the average RMS error for treadmill exercise is about average 5-6 • for able-bodied subjects without considering the time delay.…”