ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
RECOGNITION OF EMG BASED HAND GESTURES FOR PROSTHETIC CONTROL USING ARTIFICIAL NEURAL NETWORKS Sasipriya.S, Prema.P Department of Biomedical Engineering, PSG College of Technology, Coimbatore. email id: sashpsg@gmail.com, ppr@bme.psgtech.ac.in Abstract
–
EMG
(Electromyography)
is
a
I. INTRODUCTION
biological signal derived from the summation of
Congenital defects or accidental loss of
electrical signals produced by muscular actions.
limbs can be corrected by the use of artificial limbs
This EMG can be integrated with external hardware
and
rehabilitation.
control
Pattern
prosthetics
recognition
plays
or prostheses. The most efficient way of controlling
in
the prosthesis is by the use of EMG signals
an
obtained from the active muscles. EMG signals
important role in developing myo-electric control
obtained by different actions can be used to make
based interfaces with prosthetics and Artificial
the prosthesis perform those different functions in
Neural Networks (ANN) are widely used for such
real time. For doing so, the EMG signal obtained is
tasks. The main purpose of this paper is to
filtered, windowed and after digital conversion,
classify different predefined hand gestured EMG
based on the features extracted, it is classified and
signals (wrist up and finger flexion) using ANN
given as control inputs to a prosthetic system for
and to compare the performances of four
activating correct functions. The classifier is an
different Back propagation training algorithms
important element in this system. Artificial neural
used to train the network. The EMG patterns are
networks (ANN) are mathematical modeling of
extracted from the signals for each movement and
biological
then ANN is utilized to classify the EMG signals
neuronal
systems
and
they
are
particularly useful for complex pattern recognition
based on their features. The four different
and classification tasks that are difficult for
training algorithms used were SCG, LM, GD and
conventional computers or human beings. The
GDM with different number of hidden layers
nonlinear nature of neural networks, their ability
and neurons. The ANNs were trained with those
to learn from their environments in supervised as
algorithms using the available experimental data
well as unsupervised ways, as well as their
as the training set. It was found that LM
universal approximation property make them
outperformed the others in giving the best
highly suited for solving difficult signal processing
performance within short time elapse. This
problems. But, it is critical to select the most
classification can further be used to control
appropriate neural network paradigm that can be
devices based on EMG pattern recognition.
applied for specific function. Artificial neural networks based on Multi-Layer Perceptron Model
Keywords - Electromyography, Myo-electric control, Artificial Neural Network, Back propagation algorithms, Pattern recognition.
(MLP) are most commonly used as classifiers for EMG pattern recognition tasks and selecting an appropriate
training
function
and
learning
algorithm for a classifier is a crucial task. The time
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
115
www.iaetsd.in
ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
delay between the command and activation of
EMG signals in which 75 represented finger flexion
prosthetic function should be minimum (not more
and the rest represented wrist up functions
than 100ms) for the comfort of the users [1], [2], [3],
obtained from volunteers and inputted to the
[4]. Hence the classifier should be precise and
neural network for classification.
rapid in producing correct signals for controlling
B. NEURAL NETWORK ARCHITECHTURE
the prosthesis, inspite of inaccuracies that may occur during the process of detection and
A neural network is a general mathematical
acquisition of EMG signals. So it is necessary to
computing paradigm that models the operations of
assess the impact of neural network algorithms on
biological neural systems [4]. Weights (which are
the performance, robustness, and reliability aspects
determined by the training algorithm), bias and
and to identify those that work the best for solving
activation function are important factors for the
our problem of interest. This work has attempted
response of a neural network. A Multi-Layer
to classify EMG based on two predefined hand
Perceptron
gestures and to identify the better performing training
functions
in
Feed
model
based
on
Feed
forward
Backpropagation algorithm is used here for
Forward
classification. Fig 1 shows the basic architecture of
Backpropagation algorithm (standard strategy of
a neural network with hidden layers between the
MLP) used in recognizing the patterns.
input and the output layers.
II. .METHODOLOGY A.
EMG
SIGNAL
ACQUISITION
AND
FEATURE EXTRACTION EMG signals used in this study are acquired from the muscles of the forearm namely Flexor Carpi Ulnaris (FCU), Extensor Carpi Radialis (ECR) and Extensor Digitorum (reference) for two types of hand movements-finger flexion and wrist up. FCU assists in wrist flexion with ulnar deviation and ECR assists in extension and radial
Fig 1: Neural network with hidden layers.
abduction of the wrist. The myoelectric signals are acquired by means of single channel differential
The designed ANN for EMG pattern
electrodes (disposable Ag/AgCl surface electrodes)
recognition consists of 3-layers: input layer, tan-
which are then amplified and filtered before
sigmoid (standard activation function) hidden
further processing.
layer and a linear (purelin) output layer. Each layer except input layer has a weight matrix, a bias
13 different statistical features extracted
vector and an output vector. The learning rule for
from the acquired EMG signals are Integrated
the propagation of neural network defines how the
EMG, Mean Absolute value, Modified Mean
weights between the layers will change [5], [6].
Absolute value 1, Modified Mean Absolute value
Here, the input is a 13 x 150 matrix and the
2, Simple square integral(energy), Variance, Root
corresponding target is a 2 x 150 matrix, (as 13
Mean Square, Waveform length, Zero Crossing,
features extracted from 150 samples and there are 2
Slope sign change, Willison amplitude, Difference
types of gestures to be classified). The classification
Absolute Mean Value and Histogram of EMG.
was divided into 3 stages: training (70% of
These 13 features are extracted from 150 samples of
samples), validation (15% of samples) and test
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
116
www.iaetsd.in
ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
(15% of samples). Four main training algorithms used were Scaled Conjugate Gradient (SCG), Levenberg-Marquardt
(LM),
Gradient descent
(GD), and Gradient descent with momentum (GDM).
The
sample
input
vectors,
their
corresponding target vectors and the output vectors after classification are shown in Table.1. III. RESULTS AND DISCUSSION The classification is done by altering the training functions, learning rate (GD and GDM), and number of neurons in a hidden layer with a Fig 2: Best Performance of LM
standard performance function MSE (Mean Square Error) and the better performing algorithm is identified. The summary of the results of various training algorithms used and the corresponding number of neurons is listed in Table.2. It is found that, on training the network with 10 neurons, SCG gives the least error and best performance. But on considering the fast convergence (which is the actual need for prosthetic control), LM with 20 neurons gives good classification performance earlier than SCG. If number of neurons is increased to 30, performance of LM is better than SCG, with a least time elapse. Also the classification rate of SCG algorithm is saturated at 81.3 % after 30 neurons, no matter how many neurons increased, the rate of correct classification remains constant. But higher
the neuron number, better
the
classification rate in LM algorithm. The other two
Fig 3: Classification Rate of LM
algorithms, GD and GDM require more time and IV. CONCLUSION AND FUTURE WORK
number of iterations needed for classification.
This work has been carried out to find the
Their performances are also undesired. Hence the classification
efficiencies
of
SCG
and
appropriate classifier for
LM
EMG
signals. The
algorithms alone are shown in Table 3 and Table 4
experimental results show that the Levenberg-
respectively..
Marquardt
The
best
performance
and
algorithm
gives
good
and
fast
classification rate of LM based ANN with 50
performance with least computations. The main
neurons are also shown in Fig 2 and Fig 3. Thus the
disadvantages of this algorithm are requirements
LM network outperforms the other algorithms by
of large memory and increased number of neurons
providing
which may affect the hardware design (DSP
the
least
response
time
(fastest less
processors) of the prosthetics. Hence the future
number of iterations required and less error values.
will be to optimize the performance and provide
convergence),
higher
classification
rate,
interface to the hardware.
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
117
www.iaetsd.in
ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
.
Fig 4: LM Training tool
Table.1 : SAMPLE INPUT DATA AND SIMULATED OUTPUT DATA
Input vectors
13 extracted features from EMG
Target vectors
Wrist up Finger flexion
Output vectors
Wrist up Finger flexion
Sample 1 257.92 0.42987 0.32246 0.20606 643.08 1.0647 0.80911 67.429 89.086 0.14872 6.5714 67.571 116.29 1
Sample 2 318.26 0.53044 0.39805 0.28215 695.51 1.1527 0.88938 88 115.55 0.1929 9.1429 68 111.86 1
Sample 3 154.49 0.25749 0.19358 0.1467 425.03 0.70138 0.58973 26.857 46.511 0.077647 0.85714 73.857 108.86 0
0 0.7866 0.2339
0 0.9530 0.0411
1 0.3495 0.6390
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
118
www.iaetsd.in
ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
Table.2 : COMPARISON OF VARIOUS ALGORITHMS
BEST PERFORMANCE
GDM
GD
LM
SCG
0.15275
0.14585
0.18598
0.14172
HIDDEN NEURONS 10
0.27954
0.16829
0.0991680
0.13132
20
0.12168
0.1526
0.14919
0.17166
30
1000
1000
5
21
10
1000
968
9
20
20
1000
1000
7
27
30
EPOCH AT WHICH BEST PERFORMANCE IS GOT
Table.3 : SHOWING THE CLASSIFICATION RATE SATURATION OF SCG HIDDEN NEURONS
10
20
30
40
50
Correctly classified
76.7
80
81.3
81.3
81.3
Misclassified
23.3
20
18.7
18.7
18.7
BEST PERFORMANCE
0.17323
0.10074
0.22587
0.14376
0.13385
EPOCH AT WHICH BEST PERFORMANCE OBTAINED
11
12
19
15
6
CLASSIFICATION RATES (PERCENTAGE)
Table.4 : SHOWING THE CLASSIFICATION RATE EFFICIENCY OF LM
HIDDEN NEURONS
10
20
30
40
50
Correctly classified
78.7
76
84
87.3
88
Misclassified
21.3
24
16
12.7
12
BEST PERFORMANCE
0.20428
0.22699
0.14131
0.30642
0.20215
EPOCH AT WHICH BEST PERFORMANCE OBTAINED
2
2
7
4
10
CLASSIFICATION RATES (PERCENTAGE)
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
119
www.iaetsd.in
ISBN: 378-26-138420-01
INTERNATIONAL CONFERENCE ON CURRENT TRENDS IN ENGINEERING RESEARCH, ICCTER - 2014
REFERENCES
bimodal drug delivery’, International Journal of Pharmaceutics, 2006.
[1] Alcimar Soares et al., ‘Development of virtual myo electric prosthetic controlled by EMG pattern recognition system based on neural networks’ ,Journal of Intelligent Information Systems,2003
[6]. Claudio Castellini, Patrick van der Smagt, ‘Surface EMG in advanced hand prosthetics’, . Biol Cybern , 2009.
[2] Md. R. Ahsan,Muhammad, I. Ibrahimy, Othman O. Khalifa, ‘EMG Signal Classification for Human Computer Interaction: A Review’ , European Journal of Scientific Research, 2009. [3] Md. Rezwanul Ahsan, Muhammad Ibn Ibrahimy, Othman O. Khalifa, ‘Electromyography (EMG) Signal based Hand Gesture Recognition using Artificial Neural Network (ANN)’, 4th International Conference on Mechatronics (ICOM), 17-19 May 2011, Kuala Lumpur, Malaysia, [4] Yu Hen Hu, Jenq-Neng Hwang,’ Handbook of neural network signal processing’, ‘‘Introduction to Neural Networks for Signal processing’’, CRC PRESS, 2001. [5] A. Ghaffari et al., ‘Performance comparison of neural network training algorithms in modeling of
INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY FOR SKILL DEVELOPMENT
120
www.iaetsd.in