A Comparative Study Of Neural Networks Models For EMG Pattern Classification

Neural Networks 24.6-5

A COMPARATIVE STUDY OF NEURAL NETWORKS MODELS FOR EMG PATTERN CALSSIFICATION

Mahyar Zardoshti Kermani, Kambiz Badie, R.M.

Hashemi, T .

Khoshaba

E l e c t r i c a l Engineering Department,Tarbiat Modares U n i v e r s i t y Djanbazan Biomedical & R e h a b i l i t a t i o n Engineering Research Center Biomedical Engineering Lab., A m i r k a b i r U n i v e r s i t y o f Tech.,Tehran

ABSTRACT A l t h o u g h n e u r a l n e t s have shown t h e i r a b i l i t y f o r t h e EMG p a t t e r n c l a s s i f i c a t i o n i n o r d e r t o c o n t r o l t h e c y b e r n e t i c arm, however, t h e s e l e c t i o n o f topology as w e l l as parameters o f t h e network i s s t i l l a m a t t e r o f question. T h i s paper i s a s t e p towards s o l v i n g t h i s problem through simul a t i o n o f a numer o f networks. P r i l i m i n a r y r e s u l t s show t h a t back-propagation network r e s u l t s i n t o a b e t t e r performance.

purpose o f c l a s s i f y i n g movements i n a s t r u c t u r e d way. Moreover, d i f f e r e n t types o f topology w i t h d i f f e r e n t parameters may be c o n s i d e r e d f o r t h e same t y p e o f t h e network f o r which t h e o p t i m a l s e l e c t i o n needs e m p i r i c a l observations. The purpose o f t h i s paper i s t o approach t h e problem o f s e l e c t i n g a reasonable network through making a comprehensive a n a l y s i s o f t h e r e s u l t s obtained through s i m u l a t i n g t h e above networks.

INTRODUCTION METHOD OF SIMULATION AND RESULTS EMG s i g n a l i n f o r m a t i o n processing has been

w i d e l y used f o r c l a s s i f i c a t i o n o f motion p a t t e r n s i n upper e x t r e m i t y p r o s t h e s e s [ l l . Among t h e so many approaches u t i l i z e d i n t h i s regard, t h e c o n n e c t i o n i s t approach can be meaningful i n t h e sense t h a t i t can f i t the r e a l time nature o f c l a s s i f i c a t i o n as w e l l as t h e f a c t t h a t no w e l l - d e f i n e d mathematical formalism i s r e q u i r e d f o r However, t h e b a s i c classification [2,5]. problem w i t h u t i l i z i n g a c o n n e c t i o n i s t approach i s t o s e l e c t t h e a p p r o p r i a t e topology f o r t h e n e u r a l n e t as w e l l as i t s parameters. Many t y p e s o f n e t w o r k s have been suggested f o r decision-making purpos e s e a c h c a p a b l e i n i t s own s p e c i f i c domain. I n t h i s paper we t r y t o t a c k l e t h i s problem through making a comparative s t u d y on t h e p e r f o r m a n c e s o f d i f f e r n t types o f networks. A

GLANCE

THE PROBLEM OF DIFFERENT NETWORKS

ON

USING

Standard b a c k - p r o p a g a t i o n , f a s t backpropagation, f u n c t i o n a l - l i n k , probabilist i c [3] and counter-propagation models can be mentioned as w e l l - k n o w n n e u r a l n e t models. However, i t seems t h a t each o f these models has u t i l i t y w i t h i n t h e framework o f a d i f f e r e n t s p e c i f i c domain which s t i l l i s n o t well-known. Because o f t h i s , t h e r e i s no systematic approach t o i d e n t i f y t h e mostly-suited network f o r t h e

1460

I n o r d e r t o s i m u l a t e t h e networks, we made use o f a s o f t w a r e package c a l l e d "NeuralWorks P r o f e s s i o n a l I 1 , Which i s a userf r i e n d l y t o o l f o r t h e s i m u l a t i o n purposes p a r t i c u l a r l y r e q u i r i n g graphic visual izat i o n s [4]. Almost 2 5 d i f f e r e n t t y p e s o f networks have been e n c a p s u l a t e d i n t h i s package, o f f e r i n g t h e chance t o t h e user t o make h i s d e s i r e d r e q u e s t t h r o u g h s p e c i f y i n g t h e number o f l a y e r s as w e l l as t h e number o f neurons f o r each l a y e r . 300 EMG p a t t e r n s picked up from t h e biceps and t r i c e p s muscles o f an above e l b o w amputee were i n i t i a l l y u t i l i z e d t o t r a i n each network, i . e . l e a r n i n g t h e weigths o f t h e network c o n n e c t i o n s . These p a t t e r n s are t o be c l a s s i f i e d w i t h i n t h e framework o f a 5-class c l a s s i f i c a t i o n problem which corresponds t o a two-degree-of-freedom p r o s t h e s i s . Having accompolished t h e t r a i n i n g mode f o r each t y p e o f n e t w o r k , t h e t r a i n i n g p a t t e r n s were used f i r s t as t e s t i n g p a t t e r n s i n o r d e r t o s t u d y each network's c l a s s i f i c a t i o n performance. N e x t , 100 r a n d o m l y s e l e c t e d new E M G p a t t e r n s were u t i l i z e d t o a c t u a t e t h e t e s t i n g performances o f t h e networks t r a i n e d through t h e p r e v i o u s l y mentioned t r a i n i n g p a t t e r n s . I n order t o compare t h e perfomances o f t h e n e t w o r k s , error behaviors per c e r t a i n number o f epoches o f t h e i n p u t p a t t e r n s f o r t h e l e a r n i n g mode were used. F i g . 1 i l l u s t r a t e s one o f t h e

Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Vol. 13,No. 3. 1991 CH3068-4/91/0000-1460 $01.00 0 1991 IEEE

simulated networks together corresponding e r r o r behavior.

with

i t s

t h a t o u r c l a s s i f i c a t i o n p r o b l e m can n o t n e c e s s a r i l y be e x p l a i n a b l e u s i n g a s t a t i s t i c a l preassumption. T h i s i n t i a t e s us t o make more knowledgeable i n v e s t i g a t i o n on the nature o f p r o b a b i l i s t i c neural net i n order t o grasp ideas on i t s u t i l i t y domain i n our p r e s e n t c l a s s i f i c a t i o n problem. I t seems t h a t a c o g n i t i v e a p p r o a c h can be q u i t e h e l p f u l i n t h i s regard. E l a b o r a t i n g t h i s p o i n t i s a p a r t o f our research work. REFERENCES

[l]G.N.Saridis, T.P.Gootee, "EMG P a t t e r n A n a l y s i s and C l a s s i f i c a t i o n f o r a Prosthet i c A r m " , I E E E Trans. on BME, Vol. 29, No. 6, June 1982 F i g . l.One o f t h e s i m u l a t e d n e t w o r k s a s w e l l a s i t s l e a r n i n g b e h a v i o r f o r 1000 epoches and a c t i v a t i o n o f t h e o u t p u t neurons t o a samp7e t e s t p a t t e r n .

Results o f simulations followoing facts:

demonstrate

the

1)Although f a s t back-propagation converges f a s t e r than back-propagation, however due t o i t s inexactitudeness i n h a n d l i n g t h e e r r o r f u n c t i o n , i t s p e r f o r m a n c e i s much lower compared t o back-propagation. 2)Altough some redundancies e x i s t i n t h e i n p u t s o f f u n c t i o n a l - l i n k , however, i n t h e s i t u a t i o n where t h e d e c i s i o n - m a k i n g problem r e q u i r e s a h i g h number o f neurons, such as o u r case o f c l a s s i f y i n g EMG p a t t e r n s , t h i s redundancy n o t o n l y does n o t r e s u l t i n t o a h i g h e r performance, b u t makes t h e f u n c t i o n a l - l i n k converge i n t o a h i g h e r e r r o r w i t h i n t h e l i m i t e d t i m e cons i d e r e d f o r t r a i n i n g t h e network.

121M.F. Kel l y , P.A.Parker, R.N. S c o t t , "The A p p l i c a t i o n o f Neural Networks t o Myoelect r i c Signal Analysis: A Preliminary Study", I E E E Trans. on BME, Vol. 3 7 , No. 3, March 1990 C3lD.F.Specht. " P r o b a b i l i s t i c Neural Networks and t h e P o l y n o m i a l Ada1 i n e as Complementory Techniques f o r C l a s s i f i c a t i o n " , I E E E Trans. on N e u r a l N e t w o r k s , Vol. 1 , No. 1 , March 1990 [4]"Neuralworks P r o f e s s i o n a l Guide", Neuralwere I n c . 1990

I1 User's

"A [S]T.Khoshaba, K.Badie, R.M.Hashemi, Neural Network Approach t o C l a s s i f i c a t i o n o f EMG P a t t e r n s i n Cybernetic A r m " , Proc. o f I E E E I n t . Con. On S i g n a l P r o c e s s i n g , B e i j i n g , China, 1990

3 ) A l l back-propagation,functional-link and probabi 1i s t i c networks show a b e t t e r performance compared t o counter propagation, s i n c e our c l a s s i f i c a t i o n problem i s t o be c o n s i d e r e d f o r t h e f e a t u r e space w i t h r e c i p r o c a l over 1aps between t h e c l asses. 4)Back-propagation demostrates a b e t t e r performance compared t o p r o b a b i 1 i s t i c neural n e t Although i t seems t h a t t h i s r e s u l t i s due t o i n s u f f i c i e n c y o f t r a i n i n g p a t t e r n s f o r t h e p r o b a b i l i s t i c neural n e t , we however l e a v e a p o s s i b i l i t y t h a t o u r c l a s s i f i c a t i o n p r o b l e m c a n be w e a k l y e x p l a i n e d on t h e b a s i s o f s t a t i s t i c a l preassumptions.

.

CONCLUDING REMARKS

Out o f t h e r e s u l t s obtained from t h e simul a t i o n s , t h e f a c t t h a t a back-propagation shows a b e t t e r performance t o probabi 1 i s t i c neural n e t , leaves us t h e p o s s i b i l i t y

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Vol. 13. No. 3. 1991 CH3068-4/91/0000-1461 $01.00 0 1991 IEEE

1461