The Application of Wavelet Transform In The Feature Extraction of Plant Electrical Signals

Size: px

Start display at page:

Download "The Application of Wavelet Transform In The Feature Extraction of Plant Electrical Signals"

Christopher Malone
5 years ago
Views:

1 , pp The Applicatio of Wavelet Trasform I The Feature Extractio of Plat Electrical Sigals Zhag Xiaohui 1 *, Zag Haihe 1, Su baopig 1, Zhag zhixia 2 (1 College of Electrical Egieerig, Hea Uiversity of Techology, Zhegzhou , Chia 2 School of Iformatio Egieer, Hea Istitute of Sciece ad Techology, Xixiag , Chia) Abstract: A method of feature extractio i o-statioary sigals is proposed to improve the correct classificatio rates of plat electrical sigals. The samples are composed of plat electrical sigals datum which icludes the period of four circumstaces. For wavelets have the trait of arbitrary distictio ad decompositio, eigevector which reflect differet state of plat electrical sigals are extracted from differet frequecy segmets with the techology of wavelet decompositio. The we take them iput eural etwork as samples to establish the model of BP eural etwork. Experimetal results demostrate that the classificatio accuracy of the proposed feature extractio method for experimet plat electrical sigals is high. Keywords: plat electrical sigals, wavelet trasform, feature extractio, BP eural etwork 1 Itroductio Plat electrical sigal, the trasmitted iformatio related to the processes of plat physiology is a importat physiological sigal. It is the plat i respose to evirometal chage stimulatio. Scietists from verify the presece of radio sigals of the plat to ascertai the characteristics ad electrical trasmissio mechaisms of the plat, ad developmet to the curret model for establishig electrical plats ad related physiological effects study[1-5]. With the developmet of weak sigal detectio processig techology ad computer iformatio collectio ad data processig system realizatio, the extractio ad aalysis of weak plat electrical sigal will be more scietific; feasibility ad quatitative aalysis of the wave sigals of the plat is also possible. There are commoly used methods [6,7]: plats wave sigals i time domai, frequecy domai aalysis, wavelet aalysis ad time series aalysis, electrical plats AR model ad eural etwork predictio, ad so o. Wavelet trasform sigal projected oto a set of mutually orthogoal wavelet fuctio sheets ito space, which ca be decomposed ito two parts: low frequecy part ad high frequecy part. I the ext layer decompositio, the low frequecy part was coducted re-decompositio, while the high frequecy part was ot coducted re-decompositio, which is very suitable for the weak low-frequecy electrical sigal ISSN: ASTL Copyright 2014 SERSC

2 feature extractio of the plat. Based o wavelet trasform, this paper presets a method, which is suitable for o-statioary sigal feature extractio, ad wavelet coefficiets eergy eigevalues as BP eural etwork classifier for automatic idetificatio, achievig good recogitio effect. 2 Basic Priciples 2.1. Wavelet decompositio priciple It is a effective tool for sigal processig because wavelet trasform ca localize i both time ad frequecy field. I practical, we ofte use Mallat fast algorithm. Here will be described with three-layer decompositio, ad wavelet decompositio tree show i Figure 1.Where, S is the sigal to be decomposed, represetig the decomposed approachig sigal of ( 1,2,3, ) the details of the decomposed sigal[8]. A L. D ( 1,2,3, L ) represets Fig.1 Three-level structure of the wavelet decompositio tree As ca be see i Figure 1, the low frequecy part was further decomposed, but the high frequecy part will ot be decomposed. Decompositio has relatio S=A3+D3+D2+Dl. I additio, there is just explaiig three-layer decompositio. If we wat to further decompose, the low-frequecy part of the A3 ca be decomposed ito low frequecy A4 ad high frequecy D4.The followig further decompositio, ad so o Electric sigal of the plat The electric sigal of the plat is weak low-frequecy. Studies [9,10] showed that the amplitude of electric sigal of the plat is oly a few pv to a few mv rage. Experimetal data is the electric sigal of aloe Vera collected uder the darkess, broke braches, smoke, ad burig coditios or stimuli coditio. I this study, We used the biological fuctio experimet system to measure electrical sigals. The measurig system is show i [11-12]. The samplig frequecy of the origial data is Fs 5000 Hz. After wavelet decompositio, the -lever detail coefficiets correspodig to frequecy rage is Copyright 2014 SERSC 77

3 1 ( Fs / 2, Fs / 2 ) ; the -lever approximatio coefficiets correspodig to 1 frequecy rage is (0, F s / 2 ). Thus, we ca calculate the details of the wavelet decompositio of the sigal ad the approximatio sigals the frequecy rage, which are show i table 1. Table 1. Frequecy rage after wavelet decompositio Sigals D5 d4 d3 d2 d1 a5 Frequecy rage ~ 312.5~ 625~ 1250~ 2500~ 0~ (HZ) BP eural etwork establishmet ad traiig BP eural etwork has bee successful i may applicatios of recogitio ad classificatio problems. BP etwork is a former propagatio multilayered etwork, i additio to the iput layer ad the output layer, there are oe or more hidde layers. The etwork traiig method is called error back propagatio method (Back Propagatio, BP),which uses the miimum error betwee the actual etwork output ad the desired output(target vector) to correct the multilayer etwork coectio weights layer by layer from back to frot. This successive costatly modify the weight vector method called successive correctio method [13]. 3 Eergy characterizatio extractio algorithm The origial sigal is deoted as A s, accordig to the priciple of wavelet decompositio, A ca be decomposed as: A s s A1 D1 A2 D1 D2 The decompositio of the recostructed sigal, which icludes the time-domai of the origial sigal i differet bads, Research uder differet frequecy bads sigal characteristics, ad the extract the character frequecy rages, which ca most express the fault features. The expressio for the total eergy of the sigal is: E EA j ED (2) j The approximatio sigal of the j-th layer ad the layers detail sigal eergy ca be selected as are prevetatives to aalyze the characteristics of the sigal. The (1) 78 Copyright 2014 SERSC

4 T [ ED, ED, L, ED j ]. Feature feature vector is costructed as followig: 1 2 extractio algorithm of the plat electric sigal based o wavelet aalysis is described as follows: 1) Coduct five-layer wavelet decompositio after de-oisig a log period of time about 20s plat's electric sigal, extractig the detail coefficiets of wavelet bad divisio, usig S to represet wavelet coefficiet of the i-th layer j-th ode. 2) Calculate the variace ad the orm of wavelet coefficiets, accordig to the calculated results to class the experimetal data. 3) Calculate the eergy [14] of the wavelet coefficiets, set the sigal S correspodig to the eergy is E, the, E S 2 k ( t) dt d (3) k 1 2 k Where, d represets sequece value of the i-th layer j-th ode s wavelet coefficiet. 4) Structure feature vector. Usig five-layers of the wavelet decompositio eergy value of detail wavelet coefficiets as elemets to structure feature vector T, which ca be described as follows: T [ E5,1, E5,2, E5,3, E5,4, E5, 5] Iput the obtaied feature vector to BP eural etwork to testig. (4) 4 Experimetal results ad aalysis Four groups of plat electrical data refereced i this study are the aloe plat electrical sigals from Xi'a Uiversity of Techology Laboratory collected. We'll represet the plat electrical sigals collected uder the darkess, smoke, broke braches ad burig coditios usig symbols: A, B, C, ad D. Adoptig db5 wavelet to coduct five layers wavelet decompositio for each plat electrical sigal, extractig the wavelet coefficiet of each ode, calculatig the orm ad variace of wavelet coefficiets as a kid of feature to aalyze the sigal. Because the sample data are more, so oly 2 set of values of each characteristic group are listed i this paper, which are show i Tables2-3. By cotrast, it ca be foud that the characteristics of the data set A ad B are i a order of magitude, the characteristics of the data set C ad D are i the other a order of magitude, which usually higher tha the previous. the stimulatio of aloe uder the darkess ad smoke coditios is weaker tha it uder the coditios of broke braches ad burig. I order to make the classificatio accuracy of the plat electrical higher ad the simulatio experimets easier, we divided the experimetal Copyright 2014 SERSC 79

5 data ito two categories: the sigal uder the coditios of darkess ad smoke is a kid of class, uder the coditios of broke braches ad burig is a kid of the other class. Through a lot of experimetal results, it ca easily distiguish these two categories. Table 2. The ormal of wavelet coefficiets Data Set Number T1 T2 T3 T4 T5 T6 A(Darkess) B(Smoke) C(broke braches) D(Burig) Table 3. The variace of wavelet coefficiets Data Set Number T1 T2 T3 T4 T5 T6 A(Darkess) B(Smoke) C(broke braches) D(Burig) BP eural etwork achieved successfully applied i may recogitio ad classificatio problems. I this paper, we adopt three-layer BP eural etwork to classify [15], which is widely used i the curret. Ad iput T to the BP eural etwork for testig. Each category of data cotais two the plat electrical, amely category vector icludes two categories. Because the eergy extracted feature vector dimesio is 5, so the BP eural etwork iput layer odes is 5. Based o empirical formulas ad specific experimetal results set the odes of hidde layer is 12 ad the odes of output layer is Copyright 2014 SERSC

6 Neural etwork is a itelliget algorithm, each time the results are ot ecessarily able to achieve the set parameters. Settig the maximum umber of traiig steps is 20000, traiig target is 0.06, ad the program successfully reaches to set parameters deotig process efficiecy. Traiig results are show i Figure 2. I the experimet, the plat electrical were coducted o 6-layer, 5-layer ad 4-layer wavelet decompositio ad obtaiig each feature vector. Iput the feature vector, which was obtaied by 6-layer wavelet decompositio, ito BP eural etwork. Although the sigal recogitio rate is relatively higher, yet the program efficiecy is relatively lower. Iput the feature vector, which was obtaied by 4-layer wavelet decompositio, ito BP eural etwork. Although the program efficiecy is relatively higher, yet the sigal recogitio rate is relatively lower. The results are show i table 4. Fig.2 Traiig results Table 4. The experimetal results of BP Neural Network The layers of Wavelet decompositio Sigal recogitio rate Program Efficiecy 6 96% 50% 5 91% 83% 4 77% 100% Therefore, cosiderig the experimetal results, we selected the plat sigal recogitio method, which is based o the eergy feature of 5-layer wavelet decompositio. The experimetal results show that the plat electrical sigal recogitio rate of the proposed method reaches 91%. 5 Coclusio I this paper, the wavelet trasform was carried out o the plat electrical sigal decompositio. Choose the extracted orm ad variace which iclude i differet Copyright 2014 SERSC 81

7 frequecy bads of the sigal, as the characteristics of prelimiary idetificatio of the plat electrical sigal. Fially, a feature vector was iput ito BP eural etwork. The feature vector which was costructed by the eergy values of the wavelet packet coefficiet. The proposed method is oly exploratory research i plat electrical aspects. I our further research work, other feature extractio methods ad the improved BP eural etwork will be tried. Therefore, the plat weak electrical sigals study will play a icreasigly importat role i facility agriculture, the agricultural aspects of iformatio techology. Refereces 1. STANKOVIB, DAVIESE. Both actio potetials ad variatio potetials iduce proteiase ihibitor gee expressio i tomato, J. FEBS Letters.390,3 (1996) 2. Lou Cheghou. I higher plats electrochemical wave messeger, J. Acta Biophysica Siica. 4(1996) 3. Gurovich L, Hermosilla P. Electric sigalig i fruit trees i respose to water applicatios ad light-darkess coditios,j. J Plat Physiol.166 (2009) 4. Eric Davies. New fuctios for electrical sigals i Plats,J. New Phytologist. 161(2004) 5. Hastei S M, Felle H H. Naoifusio: a itegratig tool to study elicitor perceptio ad sigal trasductio i itact leaves,j. New Phytologist. 161(2004) 6. Wag Lazhou, Li Dogsheg, Li Jiao. Advace o the testig research of weak electrical sigals i plats,j. Joural of Chia Jiliag Uiversity.19,1(2008) 7. Wag Xu, Xue Lifag, Yag Da, Ha Zhiya. Speech visualizatio based o wavelet trasform,j. Applicatio Research of Computers.26,1(2009) 8. Xiaofei Ya, Zhogyi Wag, La Huag, Cheg Wag, Ruifeg Hou, Zhilog Xu, Xiaoju Qiao. Research progress o electrical sigals i higher plats,j. Progress i Natural Sciece. 19,5(2009) 9. Che Wei, Weg Shaojie, Wag Liglig, Chu Chuhua, Liu Hog. Desig of the measuremet system for week electrical sigal based of virtual istrumet ad wavelet de-oisig,j.chiese Agricultural Mechaizatio.231, 5(2010) 10. Dig Jili, Dig Guiyig, Li Haixial. Study o Electrical Sigals of Seedligs of Cucumis sativus L. by a ew Isecticide "ZJ0967",J.Joural of Zhejiag Uiversity of Sciece ad Techology.18,3 1(2006) 11. Zhag Xiaohui, Yu Nigmei, Xi Gag, Meg Xiaoli. Chages i the power spectrum of electrical sigals i maize leaf iduced by osmotic stress, J.Chiese Sciece Bulleti. 57,4 (2012) 12. Xiaohui Zhag,Nigmei Yu,Xiaoli Meg,Gag Xi.Power spectrum aalysis of Maize based o wavelet de-oisig. The 2d iteratioal coferece o iformatio sciece ad egieerig,(2010),october 12-14; Hagzhou, Chia 13. Li Xiaobig,Shu Xiaoli,Xia Liazhe. Wavelet eural etwork i automatic target recogitio based o wavelet momet, J.Joural of Southeast Uiversity(Natural Sciece Editio). S1(2006) 14. Zhu Yaqig, He Yigag. Fuzzy Neural Network Based Aalog Circuit Fault Diagosis Usig Geetic Algorithms, J. Computer Sciece. 37, 12 (2010) 15. Ge Zhexue, Su Zhiqiag,Editor,Neural etwork theory ad the realizatio of MATLAB2007.,Publishig House of electroics idustry, Beig(2012) 82 Copyright 2014 SERSC

Radar emitter recognition method based on AdaBoost and decision tree Tang Xiaojing1, a, Chen Weigao1 and Zhu Weigang1 1

Radar emitter recognition method based on AdaBoost and decision tree Tang Xiaojing1, a, Chen Weigao1 and Zhu Weigang1 1 Advaces i Egieerig Research, volume 8 d Iteratioal Coferece o Automatio, Mechaical Cotrol ad Computatioal Egieerig (AMCCE 7) Radar emitter recogitio method based o AdaBoost ad decisio tree Tag Xiaojig,