Similarity analysis of EEG data based on self organizing map neural network

COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Simila i y Analysis o EEG Da a Based on Sel O ganizing Map Neu al Ne wo k Ib ahim Salem JAHAN1, Michal PRILEPOK2, Vacla SNASEL2, Ma ek PENHAKER3 1Depa men o Compu e Science, Facul y o Elec ical Enginee ing and Compu e Science, VSB–Technical Uni e si y o Os a a, 17. lis opadu 15, 708 33, Os a a, Czech Republic 2Depa men o Compu e Science, Facul y o Elec ical Enginee ing and Compu e Science, IT4 Inno a ions, Eu opean Cen e o Excellence VSB–Technical Uni e si y o Os a a, 17. lis opadu 15, 708 33, Os a a, Czech Republic 3Depa men o Cybe ne ics and Biomedical Enginee ing, Facul y o Elec ical Enginee ing and Compu e Science, VSB–Technical Uni e si y o Os a a, 17. lis opadu 15, 708 33, Os a a, Czech Republic [email protected], mic[email p o ec ed], [email p o ec ed], [email p o ec ed] Abs ac . The Elec oencephalog aphy (EEG) is he eco ding o elec ical ac i i y along he scalp. This eco ded da a a e e y complex. EEG has a big ole in se e al applica ions such as in he diagnosis o human b ain diseases and epilepsy. Also, we can use he EEG signals o con ol an ex e nal de ice ia B ain Com- pu e In e ace (BCI) by ou mind. The e a e many algo i hms o analyse he eco ded EEG da a, bu i s ill emains one o he big challenges in he wo ld. In his a icle, we ex ended ou p e ious p oposed me hod. Ou ex ended me hod uses Sel -o ganizing Map (SOM) as an EEG da a classi ie . The p oposed me hod we can di ide in ollowing s eps: cap u ing EEG aw da a om he senso s, applying il e s on his da a, we will use he equencies in he ange om 0.5 Hz o 60 Hz, smoo h- ing he da a wi h 15- h o de o Polynomial Cu e Fi ing, con e ing il e ed da a in o ex using Tu le G aphic, Lempel-Zi complexi y o measu ing simila - i y be ween wo EEG da a ials and Sel -O ganizing Map Neu al Ne wo k as a inal classi ie s. The expe - imen esul s show ha ou model is able o de ec up o 96 % inge mo emen s co ec ly. Keywo ds EEG da a, elec oencephalog aph, polynomial cu e i ing, SOM, unsupe ised lea ning. 1. In oduc ion To use he EEG signals o communica e be ween he human b ain and an ex e nal de ice becomes one o he cu en big challenges in his esea ch ield. When we a e looking on he EEG da a o di e en men al asks, hey seem o be iden ical, bu in de ails hey a e di - e en . They con ain di e en in o ma ion. So we need o ind an e icien me hod o algo i hm o de ec hese di e ences be ween di e en men al asks and be able o dis inguish be ween hem. When we a e able o dis- inguish be ween wo o mo e a ious men al asks wi h a sa is ying success a e, we can ans o m e e y men al ask o a con ol command o an ex e nal de ice, such as p os hesis and wheelchai . The EEG signals clas- si ica ion was p esen ed by se e al esea che s using a ious echniques, o example Non-nega i e ma ix ac o iza ion (NMF) [1] as a one o e icien me hods o ecognize human men al asks. 2. Rela ed Wo ks In his ield, we can ind many pape s which a e o- cused on EEG da a p ocessing. In his sec ion, we p esen a b ie o e iew o some me hods which a e e- la ed o ou a icle. Zhang e al. applied Polynomial Cu e Fi ing (PCF) o imp o e Image Quali y in Elec- ical Impedance Tomog aphy (EIT). The expe imen s on he 2D model con i med he imp o ing quali y o he econs uc ed image; also PCF can be used o im- p o e econs uc ed image quali y in 3D EIT [2]. Ta- ade and Ka i, compa ed Au o Reg essi e In eg a ed Mo ing A e age (ARIMA), A i icial Neu al Ne wo k (ANN) and polynomial cu e i ing (PCF) o wind speed p edic ion. Thei esul s showed, ha ARIMA is be e han o he me hods [3]. Kang and Lee p e- sen ed algo i hm o compensa ing ne wo k delays in a sma ac ua o based on he Lag ange Polynomial cu e i ing. Thei expe imen al esul s showed, ha his me hod can be used e ec i ely o message de- c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 547 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER lay o sma ac ua o [4]. Zhang e al. hey p oposed a me hod based on he polynomial cu e i ing algo- i hm o p ocess he ligh es ing da a and hei esul showed ha he p oposed me hod can e icien ly au o- ma ically elimina e he ou lie e o s [5]. Jishui e al. hey p oposed mul i-dimensional uzzy easoning algo- i hm o op imize he calcula ion p ocess and imp o e he i ing cu e speed and accu acy o NC Machining G aphics. Thei esul s showed ha his me hod has a sho ime compu a ion and imp o ed he i ing al- go i hm and i ing p ecision and is sui able o cu e i ing o NC machine [6]. Shang e al. hey applied cu e i ing o phase calib a ion algo i hm using e o ol age da a om sa elli e acking. The esul showed ha we could use his me hod in monopulse acking, which does no need o build he sou ce and we can use only he e o ol age [7]. Jiang e al. hey p o- posed me hod o he aul loca ion de ec ion in elec i- cal cables based on la coe icien compu a ion. Cable aul loca ion analysis is combined wi h wa ele ans- o m and i ing cu e echnique. This pape p o ed ha he p oposed me hod educes de ia ion o singu- la i y de ec ion and imp o es he aul loca ion p eci- sion [8]. Yixu Song e al. p oposed new me hod based on cu e i ing echnique combined wi h he clus e ing algo i hm o s o e he da a s eam. The expe imen e- sul s o his me hod depic he bes comp ession a io and i ing accu acy [9]. Zhang and Liu applied cu e i ing echnique me hod o de ec he disloca ion de- ec in polysilicon slices. They compa ed wo me hods o cu e i ing, quad a ic cu e i ing and Gaussian cu e i ing. Thei esul s showed ha he quad a ic cu e i ing is e ec i e and accu a e o de ec he dis- loca ion de ec in polysilicon slices [10]. Nisha e al. hey applied Cu e Fi ing Technique o Pulse Wid h Modula ion (PWM) swi ching angles o h ee phase in e e s. The esul s p o ed ha he quad a ic cu e i ing is su icien o de e mine he op imal swi ching angles in compa ison o he cubic cu e i ing [11]. Dohnalek e al. hey applied Non-nega i e ma ix ac- o iza ion on EEG da a o ind EEG pa e n ma ching and also hey used sho ime Fou ie ans o m o p e- p ocess EEG da a, Cosine Simila i y Measu e o ind simila i y in he EEG da a. The esul s o his pape showed ha his me hod is sui able o implemen a ion on g aphics p ocesso s in eal-wo ld and eal- ime ap- plica ions [12]. Apoy e al. applied he LZ complexi y o analysis biomedical signal, hey conclude ha he LZ complexi y is use ul o es ima e he bandwid h o andom p ocesses and in ha monic a iabili y in liken- pe iodic signal [13]. Hu e al. hey compa ed he LZ complexi y wi h co ela ion en opy o epilep ic seizu e de ec ion in EEG da a. Thei esul concludes ha he LZ complexi y is be e han he co ela ion en opy in he de ec ing epilep ic seizu es p oblem [14]. Guo e al. hey ha e used he LZ complexi y and dy- namic p og amming algo i hm o analyse and measu e he simila i y o DNA sequences. The expe imen e- sul s p o ed he alidi y o hei p oposed me hod [15]. Noshadi e al. hey ha e combined Empi ical mode decomposi ion (EMD) and Lempel Zi (LZ) complex- i y o dis inguish be ween human men al asks. Thei p oposed me hod eached in a e age 92.46 % in men- al ask classi ica ion [16]. Ling e al. hey used he LZ complexi y o analyse he EEG ime se ies du ing di e en sleep s a es o eigh heal hy adul s. The ex- pe imen esul showed ha he p oposed algo i hm is e icien and sui able o disc imina e he sleep s a es o he b ain, bu has some losses o da a in he da a p ep ocessing phase [17]. 3. In oduc ion o EEG The Elec oencephalog aph (EEG) is measu ing and eco ding he di e ences o he ol age om wo si es on he scalp o e ime. The i s eco ding o elec- ical ac i i y o he human b ain was made by B ege (Be ge , 1929), when he explained his me hod o mea- su ing he elec ical ac i i y o he human b ain on he scalp. The EEG signal in common has ampli ude om a ew mic o ol s up o 100 µV wi h equency in he ange om 0.5 o 40 Hz [18]. We can eco d EEG signal be ween wo ac i e elec odes, bipola eco d- ing, o be ween one ac i e elec ode and a e e ence elec ode, monopola eco ding [19]. Elec oencephalo- g aph (EEG) is gene ally used in he diagnosis o b ain diseases and epilepsy, esea ch a eas, due o he alu- able in o ma ion ha con ey by EEG signal [18]. 3.1. 10–20 In e na ional Sys em EEG eco ding is made by pu ing se o senso s on he human b ain acco ding o 10–20 In e na ional Sys- em as Fig. 1. The 10–20 in e na ional EEG Elec ode placemen is he in e na ional sys em o de e mining he EEG elec odes loca ions on he human skull. I con ains 21 EEG elec odes wi hou he ea lobe elec- odes ha called A1 connec ed o he le ea lobe and A2 connec ed o he igh ea lobe. These elec odes a e no mally used as e e ence elec odes [20], The le - e s F, T, C, P, and O s and o F on al, Tempo al, Cen al, Pa ie al and Occipi al [21]. The elec odes ha ha e e en numbe (2, 4, 6) a e placed on he igh side o he skull. The elec odes wi h odd numbe s (1, 3, 5) a e placed on he le side o he skull and Z o ze o on midline o he skull [20]. The 10 and 20 e- e ed o he cu en dis ance be ween an elec ode o o he elec ode, ei he 10 % o 20 % o he whole o dis ance om igh side o he skull o he le side, o om on o back o he skull. In some applica- ions, we need mo e EEG elec odes. In his case we c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 548 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER can pu some elec odes be ween he o iginal elec odes acco ding o 10–20 sys em, as Fig. 1 [20]. 3.2. EEG A i ac s The EEG da a a e e y sensi i e and complica ed. The e o e he EEG da a mus be clea om con ained su ounding in e e ence o ge good and eliable e- sul s. The EEG signal no mally con ains noise and di e - en kinds o in e ace signal (a i ac s). This noises ei he in e nal signals a e p oduced by subjec himsel such as elec ical ac i i y o hea , pulse, body mo e- men , pe spi a ion, eye blinking, eyes mo emen , mus- cles ac i i ies, o ex e nal signals p oduced ou o he subjec , o example 50/60 Hz om elec ical powe supply, EEG de ices, elec onic elemen s [20], [21] and e c. Fig. 1: 10–20 In e na ional Sys em o EEG Elec odes Place- men he Nasion is he place be ween he o ehead and nose, Inion is he ju a back o he skull [20]. Fig. 2: EEG Signal con amina ed by powe line in e e ence [22]. In EEG signal p ocessing ield, he emo ing o hese noises and a i ac s om EEG signal is an impo an opic [18]. Fo example, Fig. 2 show EEG signal con- amina ed by powe line in e e ence, while Fig. 3 show some EEG signal con amina ed by eye blinking a i- ac . EEG mus be il e ed o ob ain clea EEG, wi h- ou in e e ence and a i ac s, so ha da a become eady o u he analysis. The noise signal and un- wan ed signal mus be elimina ed o minimized om EEG da a wi hou losing signi ican in o ma ion and quali y ha embedded in EEG o ensu e an accu a e and pe ec analysis and diagnosis o he EEG. The e a e se e al echniques o il e ing EEG signal such as con en ional il e s and adap i e il e s ha ha e mo e e iciency han con en ional il e s o elim- ina ion o he a i ac s om EEG, because EEG signal and a i ac s ha e o e lapping spec a [23]. Fig. 3: EEG Signal con amina ed by eye blinking a i ac [22]. 4. Tu le G aphics Tu le g aphics (TG) is a e m in compu e g aphics o a me hod o p og amming ec o g aphics using a ela i e cu so posi ion ( he " u le") upon a Ca e- sian plane. In he TG, we ha e a u le wi h a d aw- ing pen on a compu e sc een. This u le mus e- spond on a sequence o commands. The u le can be con olled using hese basic commands: o wa d com- mand, is mo ing he u le in on a ew numbe o uni s, igh commands o a e u le in a clockwise di- ec ion a ew numbe o deg ees. These commands can be ex ended wi h o he mo e complica ed commands. The back and le commands cause same mo emen as o wa d and igh command, bu in he opposi e way. The numbe o commands o de e mine how much o mo e is called inpu commands, depending on he ap- plica ion. When mo ing he u le acco ding o he inpu commands, i lea es a ace, his ace ep esen he desi ed objec [24] as a simple example in Fig. 4. By This way we can ep esen and d aw he objec s, om simple o complex objec s. Using TG we con e ed EEG da a om nume ic al- ues in o ex da a and p ocess hem as ex . This con- e sion helps us o compa e wo EEG da a ial, wo men al asks, such as inge mo emen [25]. E e y EEG ail is ep esen ed by sequence o commands–mo e o wa d and u n le o igh . c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 549 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Fig. 4: A simple example o u le g aphic. 5. Compa ing Da a wi h he LZ Complexi y The Lempel-Zi (LZ) complexi y o sequences o i- ni e leng h was sugges ed by Lempel and Zi [26]. I is a non-pa ame ic, simple- o-calcula e measu e o com- plexi y in a one-dimensional da a. The LZ complexi y is ela ed o he numbe o dis inc subs ings and he a e o hei ecu ence along he gi en sequence [27]. The la ge alues co espond o mo e complexi y in he da a. The compa ing o wo TG commands lis s is he main ask o his a icle. The lis s a e compa ed each o o he . The main p ope y o he compa ison is he numbe o common sequences in bo h lis s. These se- quences a e ob ained a e applying he LZ complexi y o he TG commands lis . This numbe is ep esen ed by he pa ame e in he ollowing o m Eq. (1), which is a me ic o simila i y be ween wo u le commands lis . SM =sc min(c1, c2),(1) whe e sc - Coun o common LZ sequence in bo h com- mand lis s, c1, c2- Coun o LZ sequence in i s o second command lis . This SM gi es a esul in he ange be ween 0 and 1. The 0 esul ells us ha his wo compa ed TG commands lis ha e no hing common. They ha e he highes di e ence. I he esul is equal o 1, he wo compa ed TG commands lis a e same. 6. In e pola ion o he EEG Da a A e eco ding and il e ing o he eco ded EEG da a, we apply polynomial cu e i ing o da a smoo hing. The i ing will emo e noise and in e e ence om he da a and i he da a end. Conside he gene al o m o a polynomial i ing cu e o o de j: (x) = a0+a1x+a2x2+· · · +ajxj= j X k=1 akxk.(2) We minimized he o al e o o polynomial i ing cu e wi h leas squa e app oach. The gene al exp es- sion o any e o using he leas squa es app oach is: e =X(dj)2,(3) e = n X i=1 (yi−(a0+ j X k=1 akxk))2,(4) whe e n- is a coun o da a poin s in one mo e, i- is he cu en da a poin being summed, j- is he polynomial o de . 7. Sel -O ganizing Map (SOM) Sel -O ganizing Map (SOM) is an unsupe ised lea n- ing neu al ne wo k. The SOM in mos common used o he clus e ing and isualiza ion o complex da a. The SOM educes he da a dimension by p oduce map usually in one o wo dimension in he ou pu ha plo s simila i ies o da a oge he as Fig. 5. The SOM is ained a e many o i e a ion in he aining phase un il he map becomes s able a he ou pu . This map is gene a ed in he aining phase and used in he es - ing phase o es ima e in which g oup can belong he es inpu , while in o he ne wo k ypes, Backp opa- ga ion ne wo ks, is he a ge ou pu used o ain he ne wo k [28]. 7.1. SOM Algo i hm The SOM lea ning we can di ide in ollowing s eps: •Ini ializing weigh ec o s wi h small andom al- ues. •Choosing andom ec o om he aining se and p esen o ne wo k. c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 550 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Fig. 5: SOM S uc u e. •Finding winning neu on which has minimum dis- ance om da a inpu based on speci ic c i e ion, o SOM usually using Euclidean dis ance o mea- su e he dis ance be ween da a inpu and neu ons as Eq. (5). D(X, W ) = p(x1−w1)2+· · · + (xn−wn)2.(5) The winning neu on is called Bes Ma ching Uni (BMU). •Calcula e he adius o he neighbo hood o BMU using Eq. (6). ( ) = 0e(− /T ),(6) whe e ( ) - is adius o he neighbo hood, 0- is he adius o he map, T- ime cons an , - Cu en i e a ion. •Any nodes ound wi hin he adius o BMU mus be upda e, his means mo e he BMU and i s neighbo hood nodes owa d da a inpu as Fig. 6 using he Eq. (7). W( + 1) = W( ) + β( )h( )(X( )−W( )),(7) β( ) = β0e(− /T ),(8) h( ) = e(−(dis ance om BMU)2/2 2( )),(9) whe e β( )- Lea ning a e, h( ) - neighbo hood unc ion. •Repea ing he s eps om s ep 2 o s ep 5 o many i e a ions un il he map a ou pu becomes s able [28]. 8. P oposed Me hod The p oposed me hod is using Neu al ne wo k Unsu- pe ised lea ning o classi y EEG da a. Ou model was es ed on EEG da a o de ec index inge mo emen . We made ou p oposed me hod as ollowing: il e and he smoo h EEG da a (T aining da a se ) using 15 h Fig. 6: Upda e o Winne Neu on (BMU) and i s neighbo s, his means mo e hem owa ds da a inpu indica ed wi h X. The solid and do ed line co espond o he s a us be o e and a e upda e espec i ely. o de polynomial cu e i ing, a e ha we con e he smoo hed EEG da a in o ex o m using u le g aphic. The LZ complexi y we used o compa e wo TG commands lis s and assign he ype o mo emen o p ocessed da a ail [29]. This was done o e e y sen- so o p ocessed ial. We made a ec o V wi h dimen- sion 8, 7 channels and one da a ype class. This ec o V is used o ain he Sel -O ganizing Map (SOM) neu- al ne wo k wi h dimension o 5×5nodes o p oduce he map. When he aining is inished, he map o ou pu becomes s able. In he es ing phase we used o he EEG da a (Tes ing da a se ) o es he ne wo k as depic in expe imen scheme in Fig. 7. 8.1. EEG Da a The EEG Da a used in his expe imen was eco ded in ou labo a o y. In ou expe imen we used se en EEG channels, which we e selec ed by ou Biomedical Depa men . These se en channels a e able o cap u e mos inge mo emen da a. The eco ded signals con- ain mo emen s o one index inge . We eco ded EEG Da a om ou di e en subjec s. E e y one o hem pe o med a p ess o a bu on wi h le index inge . We used 320 eco ded inge mo emen s, and 320 eco ded ials wi hou inge mo emen . Fo e e y ask we used 576 ials o he aining se (288 ials wi h mo emen and 288 wi hou mo emen ) and 64 ials o es ing se (32 ials wi h mo emen and 32 wi hou mo emen ). The sampling a e was se o 256 Hz, and he band-pass il e was se o 0.5 Hz o 60 Hz o emo e unwan ed equencies and noises. While ex ac ing he ask da a om cap u ed EEG da a, we added be o e and a e e e y ask a ime in- e al 0.3 second. c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 551 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Fig. 7: The schema ic diag am o p oposed me hod. Black and blue lines ep esen aining phase. Red do ed lines ep esen es ing phase. 8.2. Expe imen Resul s To ain and es ou model we used k- old C oss- alida ion echnique wi h k= 10. The EEG da a se is di ided in o 10 sub-se s, o olds and he expe imen s a e epea ed o 10 imes. The ecogni ion esul s o inge mo emen s a e lis ed in Tab. 1 and he esul s o ials wi hou inge mo emen a e in Tab. 2. Tab. 1: Finge mo emen esul s. k- old Iden i ied Mis ake Iden i ied 1 93.750 % 6.250 % 2 100.00 % 0.000 % 3 93.750 % 6.250 % 4 93.750 % 6.250 % 5 96.875 % 3.125 % 6 100.00 % 0.000 % 7 100.00 % 0.000 % 8 96.875 % 3.125 % 9 90.625 % 9.375 % 10 100.00 % 0.000 % A g 96.563 % 3.438 % The p oposed model is able o de ec index inge mo emen in he ange be ween 90.6 % and 100.00 %. The de ec ion a e o ials wi hou inge mo emen a ied in he ange be ween 90.6 % and 100.00 %. The Table 3 show he pe cen age o o al iden i ied and misiden i ied ials in ou expe imen . The p o- posed model is able o de ec in a e age 96.250 % o inge mo emen ials co ec ly. The o al a e age o misiden i ied ials is abou 3.750 %. The a e age i- nal quan iza ion e o is 0.6556, and he a e age inal opog aphic e o is 0.007. 9. Conclusion This expe imen shows he abili y o ind and ecog- nize di e en men al ask in EEG da a. This helps us Tab. 2: Wi hou inge mo emen esul s. k- old Iden i ied Mis ake Iden i ied 1 93.750 % 6.250 % 2 90.625 % 9.375 % 3 100.00 % 0.000 % 4 96.875 % 3.125 % 5 100.00 % 0.000 % 6 90.625 % 9.375 % 7 93.750 % 6.250 % 8 96.875 % 3.125 % 9 96.875 % 3.125 % 10 100.00 % 0.000 % A g 95.938 % 4.063 % Fig. 8: U ma ix o k- old 1. Fig. 9: SOM o k- old 1, Red Nodes ep esen mo emen i- als and G een Nodes ep esen he wi hou Mo emen ials. o unde s and he aluable in o ma ion which is hid- den in he EEG da a. Ou app oach is able o de- cide be ween wo asks, p essed bu on wi h index in- ge and eleased bu on. We used only se en selec ed elec odes. This coun o elec odes is enough o cap- u e good EEG da a o mo emen . As a i s s ep, we used a band pass il e , o il e ou wan ed equencies which a e use ul o inge mo emen de ec ion. Ou c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 552 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Tab. 3: E alua ion o he esul s. k- old To al Iden i ied To al Misiden i ied Final Quan iza ion E o Final Topog aphic E o 1 93.750 % 6.250 % 0.658 0.007 2 95.313 % 4.688 % 0.664 0.000 3 96.875 % 3.125 % 0.653 0.008 4 95.313 % 4.688 % 0.655 0.008 5 98.438 % 1.563 % 0.658 0.017 6 95.313 % 4.688 % 0.674 0.013 7 96.875 % 3.125 % 0.653 0.003 8 96.875 % 3.125 % 0.654 0.008 9 93.750 % 6.250 % 0.638 0.007 10 100.00 % 00.00 % 0.649 0.000 A g 96.250 % 3.750 % 0.6556 0.007 Fig. 10: U ma ix o k- old 10. Fig. 11: SOM o k- old 10, Red Nodes ep esen mo emen i- als and G een Nodes ep esen he wi hou Mo emen ials. sugges ed app oach is using high o de polynomial i - ing cu e o noise and in e e ence elimina ion, u le g aphic o con e il e ed da a om numbe s in o ex , Lempel-Zi complexi y o compa e wo da a ials a Sel -O ganizing maps as a classi ie . The da a ial we e cu 0.3 second be o e he men- al ask began and 0.3 second a e . In ou expe i- men , we il e ed da a wi h polynomial i ing wi h o - de 15. This o de is enough o i da a end and emo e unwan ed noise and in e e ence su ounding en i onmen . As a classi ie , we chose SOM wi h map dimension 5×5neu ons. The es ing ec o was as- signed o clus e using BMU. Ou model was able o de ec on inge mo emen as a e age abou 96.56 %, he lowes a e we eached was 90.625 %, and he highes was 100.00 %. Fo a ial wi hou inge mo emen he a e age success ul a e is abou 95.93 %, he lowes a e is 90.625 % and he maximal is a 100.00 %. The a e age o bo h o de ec s on inge mo emen and wi hou inge mo e- men abou 96 %. In he u u e, we will con inue wi h o he publica ion ega ded o es ing o he EEG da a o modi ying ou model o imp o e he esul o EEG da a ecogni ion and inc ease he speed o ou model. Acknowledgmen This wo k was suppo ed by he Eu opean Regional De elopmen Fund in he IT4Inno a ions Cen e o Excellence p ojec (CZ.1.05/1.1.00/02.0070) and by P ojec SP2014/110, Pa allel p ocessing o Big da a, o he S uden G an Sys em, VSB–Technical Uni e - si y o Os a a. This pape has been elabo a ed in he amewo k o he p ojec „Suppo esea ch and de elopmen in he Mo a ian-Silesian Region 2013 DT 1 - In e na ional esea ch eams“(RRC/05/2013). Financed om he budge o he Mo a ian-Silesian Region. The pape has been elabo a ed in he amewo k o BIOM ( eg. no. CZ.1.07/2.3.00/20.0073). The wo k and he con- ibu ions we e suppo ed by he p ojec SP2014/194 ’Biomedicinske inzeny ske sys emy X’. c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 553 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER Re e ences [1] JAHAN, I. S. and V. SNASEL. Non-nega i e ma ix ac o iza ion o EEG. In: 2013 The In e na ional Con e ence on Technological Ad- ances in Elec ical, Elec onics and Com- pu e Enginee ing (TAEECE). Konya: IEEE, 2013, pp. 183–187. ISBN 978-1-4673-5612-1. DOI: 10.1109/TAEECE.2013.6557219. [2] ZHANG, J., G. XU, H. CHEN, Q. ZHAO and W. YAN. Using Polynomial Cu e Fi ing Me hod o Imp o e Image Quali y in EIT. In: 2006 12 h Biennial IEEE Con e ence on Elec omagne ic Field Compu a ion. Miami: IEEE, 2006, pp. 439– 439. ISBN 1-4244-0320-0. DOI: 10.1109/CEFC- 06.2006.1633229. [3] TARADE, R. S. and P. K. KATTI. A com- pa a i e analysis o wind speed p edic ion. In: 2011 In e na ional Con e ence on En- e gy, Au oma ion and Signal. Bhubaneswa : IEEE, 2011, pp. 1–6. ISBN 978-1-4673-0137-4. DOI: 10.1109/ICEAS.2011.6147167. [4] KANG, S. and D. LEE. Compensa ing Ne wo k Delays o Sma Ac ua o Using Lag ange Poly- nomial Cu e Fi ing. In: SICE Annual Con- e ence 2010. Taipei: IEEE, 2010, pp. 674–677. ISBN 978-1-4244-7642-8. [5] ZHANG, L., Y. QIN and J. ZHANG. S udy o polynomial cu e i ing algo i hm o ou lie elim- ina ion. In: 2011 In e na ional Con e ence on Compu e Science and Se ice Sys em (CSSS). Nanjing: IEEE, 2011, pp. 760–762. ISBN 978-1- 4244-9762-1. DOI: 10.1109/CSSS.2011.5974936. [6] JISHUI, X., Z. ZIPENG and W. SHUQING. S udy o he cu e i ing algo i hm o NC ma- chining g aphics. In: 2013 8 h In e na ional Con- e ence on Compu e Science. Colombo: IEEE, 2013, pp. 136–139. ISBN 978-1-4673-4464-7. DOI: 10.1109/ICCSE.2013.6553898. [7] SHANG, Y.-H., H. LUO and F.-X. YU. A Phase Calib a ion Me hod o Single Channel Monopulse T acking Sys ems Based on Cu e Fi ing Technique. In: 2012 Second In e na- ional Con e ence on Ins umen a ion, Measu e- men , Compu e , Communica ion and Con ol. Ha bin: IEEE, 2012, pp. 577–580. ISBN 978-1- 4673-5034-1. DOI: 10.1109/IMCCC.2012.142. [8] JIANG, Q., S. ZHAO, J. ZAHO and C. ZAHO. A Powe Cable Faul Loca ion Me hod Combin- ing Wi h Wa ele Analysis and Cu e Fi ing. In: 2010 China In e na ional Con e ence on Elec ic- i y Dis ibu ion (CICED). Nanjing: IEEE, 2010, pp. 1–5. ISBN 978-1-4577-0066-8. [9] SONG, Y., J. HU, X. YANG, J. FU and X. XIE. A me hod o da a s eam p ocessing based on cu e i ing. In: 2010 2nd In e na ional Con e - ence on Signal P ocessing Sys ems. Dalian: IEEE, 2010, pp. V2-542–V2-546. ISBN 978-1-4244-6892- 8. DOI: 10.1109/ICSPS.2010.5555670. [10] ZHANG, W. and T. LIU. Polysilicon Slice Dis- loca ion De ec s Segmen a ion and A ea S a is- ics Based on Cu e Fi ing. In: 2013 Fou h In e na ional Con e ence on Digi al Manu ac- u ing. Qingdao: IEEE, 2013, pp. 925–928. DOI: 10.1109/ICDMA.2013.218. [11] NISHA, G. K., S. USHAKUMARI and Z. V. LAKAPARAMPIL. CFT based op imal PWM s a egy o h ee phase in e e . In: 2012 2nd In e na ional Con e ence on Powe , Con ol and Embedded Sys ems. Allahabad: IEEE, 2012, pp. 1–6. ISBN 978-1-4673-1047-5. DOI: 10.1109/ICPCES.2012.6508066. [12] DOHNALEK, P., P. GAJDOS, T. PETEREK and M. PENHAKER. Pa e n Recogni ion in EEG Cogni i e Signals Accele a ed by GPU. In: In e na ional Join Con e ence CISIS’12, ICEUTE’12, SOCO’12 Special Sessions. Be lin: Sp inge , 2013, pp. 477–485. ISBN 978-364233017- 9. DOI: 10.1007/978-3-642-33018-6_49. [13] ABOY, M., R. HORNERO, D. ABASOLO and D. ALVAREZ. In e p e a ion o he Lempel-Zi Complexi y Measu e in he Con- ex o Biomedical Signal Analysis. IEEE T ansac ions on Biomedical Enginee ing. 2006, ol. 53, iss. 11, pp. 2282–2288. ISSN 0018-9294. DOI: 10.1109/TBME.2006.883696. [14] HU, J., J. GAO and J. C. PRINCIPE. Analysis o Biomedical Signals by he Lempel-Zi Com- plexi y: he E ec o Fini e Da a Size. IEEE T ansac ions on Biomedical Enginee ing. 2006, ol. 53, iss. 12, pp. 2606–2609. ISSN 0018-9294. DOI: 10.1109/TBME.2006.883825. [15] GUO, X., Q. DAI, B. HAN, L. ZHU and L. LI. Simila i y Analysis o DNA Sequences Based on LZ Complexi y and Dynamic P og amming Algo- i hm. In: 5 h In e na ional Con e ence on Bioin- o ma ics and Biomedical Enginee ing. Wuhan: IEEE, 2011, pp. 2151–7614. ISBN 978-1-4244- 5088-6. DOI: 10.1109/icbbe.2011.5780038. [16] NOSHADI, S., V. ABOOTALEBI and T. SADEGHI. A New Me hod Based on EMD and LZ Complexi y Algo i hms o Disc imina ion o Men al Tasks. In: 18 h I anian Con e ence o Biomedical Enginee ing (ICBME). Teh an: IEEE, 2011, pp. 115–118. ISBN 978-1-4673-1004- 8. DOI: 10.1109/ICBME.2011.6168535. c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 554 COMPUTER SCIENCE AND INFORMATION TECHNOLOGY VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER [17] LI, L. and R. WANG. Complexi y Analysis o Sleep EEG Signal. 2010 4 h In e na ional Con e - ence on Bioin o ma ics and Biomedical Enginee - ing. Chengdu: IEEE, 2010, pp. 1–3. ISBN 978-1- 4244-4712-1. DOI: 10.1109/ICBBE.2010.5515699. [18] SORNMO, L. and P. LAGUNA. Bioelec ical sig- nal p ocessing in ca diac and neu ological appli- ca ions. Bos on: Else ie Academic P ess, 2005. ISBN 01-243-7552-9. [19] QUIROGA, R. Q. Quan i a i e analysis o EEG signals ime equency me hods and chaos heo y. Lubeck, 1998. Ph.D. Thesis. Medical Uni e si y o Lubeck. [20] SANEI, S. and J. CHAMBERS. EEG signal p o- cessing. Hoboken: John Wiley, 2007. ISBN 04-700- 2581-6. [21] OCHOA, J. B. EEG Signal Classi ica ion o B ain Compu e In e ace Applica ions. Lau- sanne, 2002. Diploma Thesis. Ecole Poly echnique Fede ale de Lausanne. [22] SCHALK, G. and J. MELLINGER. A p ac ical guide o b ain-compu e in e acing wi h BCI2000: gene al-pu pose so wa e o b ain-compu e in e - ace esea ch, da a acquisi ion, s imulus p esen a- ion, and b ain moni o ing. New Yo k: Sp inge , 2010. ISBN 978-184-9960-915. [23] CORREA, A. G., E. LACIAR, H. D. PATINO and M. E. VALENTINUZZI. A i ac emo al om EEG signals using adap i e il e s in cas- cade. Jou nal o Physics: Con e ence Se ies. 2007, ol. 90, iss. 1, pp. 1–10. ISSN 1742-6588. DOI: 10.1088/1742-6596/90/1/012081. [24] ABELSON, H. and A. DISESSA. Tu le Geom- e y: The Compu e as a Medium o Explo - ing Ma hema ics. London: The MIT P ess, 1986. ISBN 978-0262510370. [25] BIBBO, D., S. CONFORTO, I. BERNABUCCI, M. CARLI, M. SCHMID and T. D’ALESSIO. Analysis o di e en image-based bio eedback models o imp o ing cycling pe o mances. In: P oceedings o SPIE - The In e na ional Socie y o Op ical Enginee ing, Image P ocessing: Al- go i hms and Sys ems X; and Pa allel P ocess- ing o Imaging Applica ions II. (8295). Belling- ham: SPIE, 2012, pp. 1–8. ISBN 978-081948942-5. DOI: 10.1117/12.910605. [26] LEMPEL, A. and J. ZIV. On he Complexi y o Fini e Sequences. IEEE T ansac ions on In o - ma ion Theo y. 1976, ol. 22, iss. 1, pp. 75–81. ISSN 0018-9448. DOI: 10.1109/TIT.1976.1055501. [27] RADHAKRISHNAN, N. and B. N. GANGAD- HAR. Es ima ing egula i y in epilep ic seizu e ime-se ies da a. IEEE Enginee ing in Medicine and Biology Magazine. 1998, ol. 17, iss. 3, pp. 89– 94. ISSN 0739-5175. DOI: 10.1109/51.677174. [28] KOHONEN, T. The sel -o ganizing map. P oceed- ings o he IEEE. 1990, ol. 78, iss. 9, pp. 1464– 1480. ISSN 0018-9219. DOI: 10.1109/5.58325. [29] PRILEPOK, M., J. PLATOS, V. SNASEL and I. S. JAHAN. Comp ession-based simila - i y in EEG signals. In: 2013 13 h In e - na ional Con e ence on In ellien Sys ems De- sign and Applica ions. Kuala Lumpu : IEEE, 2013, pp. 247–252. ISBN 978-147993516-1. DOI: 10.1109/ISDA.2013.6920743. Abou Au ho s Ib ahim Salem JAHAN was bo n in Mis a a Libya in 1974. He ecei ed his M.Sc. om VSB–Technical Uni e si y o Os a a in 2010 a Facul y o Elec ical Enginee ing and Compu e science in Measu emen and con ol Enginee ing Depa men . Now he Ph.D. S uden in VSB–Technical Uni e si y o Os a a, in Compu e Science Depa men . Michal PRILEPOK was bo n in Dolny Kubin, Slo ak Republic in 1985. He ecei ed his M.Sc. om VSB–Technical Uni e si y o Os a a in 2010 a Facul y o Elec ical Enginee ing and Compu e science in Compu e Science and Technology. Since 2010 is a Ph.D. s uden in VSB–Technical Uni e si y o Os a a, in Compu e Science Depa men . Vacla SNASEL was bo n in Olomouc. His esea ch and de elopmen expe ience includes o e 30 yea s in he Indus y and Academia. He wo ks in a mul i-disciplina y en i onmen in ol ing a i- icial in elligence, mul idimensional da a indexing, social ne wo k, o mal concep analysis, in o ma ion e ie al, seman ic web, knowledge managemen , da a comp ession, machine in elligence, neu al ne wo k, web in elligence, na u e and biologically inspi ed compu ing, da a mining, and applied o a ious eal wo ld p oblems. He has gi en mo e han 16 plena y lec u es and con e ence u o ials in hese a eas. He has au ho ed/co-au ho ed se e al e e eed jou nal/con e ence pape s and book chap e s. He has published mo e han 450 pape s. Ma ek PENHAKER was bo n in 1972. He inished M.Sc. in 1996 a Facul y o Elec ical Enginee ing and Compu e science in specializa ion Measu emen and Con ol in Biomedicine a VSB– Technical Uni e si y o Os a a, Czech Republic. He c 2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 555