12.07.2015 Views

Wavelet basis functions in biomedical signal processing - SPAN LAB

Wavelet basis functions in biomedical signal processing - SPAN LAB

Wavelet basis functions in biomedical signal processing - SPAN LAB

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

J. Rafiee et al. / Expert Systems with Applications 38 (2011) 6190–6201 619320 Normal (Neutral)151000.025 (Neutral)1055005050.08 (Neutral)60Normal (Erotic)20040100200805050.025 (Erotic)0.08 (Erotic)604040202005(Hz)05(Hz)Fig. 5. Power spectrum density (V 2 /Hz) of VPA recorded <strong>in</strong> six classes (one specific subject).The CWT results <strong>in</strong> cont<strong>in</strong>uous wavelet coefficients (CWC),which illustrate how well a wavelet function correlates with a specific<strong>signal</strong>. If the <strong>signal</strong> has a major frequency component correspond<strong>in</strong>gto a particular scale, then the wavelet at this scale issimilar to the <strong>signal</strong> at the location where this frequency componentoccurs, regardless of its amplitudes and phases. Correlationis one of the most common statistical measures, and describesthe degree of l<strong>in</strong>ear dependence between two variables <strong>in</strong> termsof a coefficient between 1 and +1. The closer the coefficient isto either 1 or +1, the stronger the correlation between the twovariables. The negative or positive sign of the coefficient <strong>in</strong>dicatesthe direction of the l<strong>in</strong>ear dependence. The coefficient 0 impliesthat the two variables are completely l<strong>in</strong>early <strong>in</strong>dependent of eachother. q x,y , the correlation between two random variables x and ywith expected values l x and l y and standard deviations r x andr y is def<strong>in</strong>ed as:covðx; yÞq x;y¼ ¼ Eððx l x Þðy l y ÞÞ; ð3Þr x r y r x r ywhere E stands for expected value of the variable and cov stands forcovariance. As E(x)=l x , E(y)=l y and r 2 (x)=E(x 2 ) E 2 (x),r 2 (y)=E(y 2 ) E 2 (y), q x,y can be def<strong>in</strong>ed as:EðxyÞ EðxÞEðyÞq x;y¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiEðx 2 Þ E 2 ðxÞ Eðy 2 Þ E 2 ðyÞIn this research, the outcome of the correlation between <strong>signal</strong>and wavelet <strong>basis</strong> function (CWC) has been established as the <strong>basis</strong>for compar<strong>in</strong>g selected mother wavelet <strong>functions</strong>. Consequently,ð4Þto f<strong>in</strong>d the most similar mother wavelet to bio<strong>signal</strong>s, motherwavelets from different families <strong>in</strong>clud<strong>in</strong>g Haar, Daubechies (db),Symlet, Coiflet, Gaussian, Morlet, complex Morlet, Mexican hat,bio-orthogonal, reverse bio-orthogonal, Meyer, discrete approximationof Meyer, complex Gaussian, Shannon, and frequency B-spl<strong>in</strong>efamilies have been analyzed based on CWC and us<strong>in</strong>g the follow<strong>in</strong>galgorithm:1. Signal segmentation: EMG, EEG, VPA <strong>signal</strong>s were segmented tothe 256-po<strong>in</strong>ts, 500-po<strong>in</strong>ts, 960-po<strong>in</strong>ts (12 s) w<strong>in</strong>dows, respectivelyas shown <strong>in</strong> Figs. 6 and 7 for EMG and VPA.2. One class of recorded data is selected from one channel for onesubject for each type of bio<strong>signal</strong>. The number of subjects is 20,6, and 1 for VPA, EMG, and EEG, respectively. There are 1, 16, 6,and 3 channels for record<strong>in</strong>g the <strong>signal</strong>s for VPA, surface EMG,<strong>in</strong>tramuscular EMG, and EEG, respectively.3. One mother wavelet function is selected from 324 candidates.In the fourth decomposition level, CWC of the segmented <strong>signal</strong>swere calculated (2 4 numbers for each segmented <strong>signal</strong>).4. Absolute value of CWC <strong>in</strong> each scale (2 4 scales) was determ<strong>in</strong>edus<strong>in</strong>g 324 mother wavelets <strong>in</strong> a segmented <strong>signal</strong> for each of the10 hand motions for EMG, as well as each of three specific channelsfor EEG. Next, the sum of this value <strong>in</strong> all 2 4 scales wasdeterm<strong>in</strong>ed for 20 EMG segmented <strong>signal</strong>s (<strong>signal</strong>s with thelength of 256 po<strong>in</strong>ts), 50 EEG segmented <strong>signal</strong>s (<strong>signal</strong>s withthe length of 500 po<strong>in</strong>ts), and 15 VPA segmented <strong>signal</strong>s (<strong>signal</strong>swith the length of 960 po<strong>in</strong>ts). Their average was calculatedfor each type of <strong>signal</strong> and called the evaluationcriterion (EC) for simplicity and to f<strong>in</strong>d the most similar mother

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!