function P=findpeaksGSS(x,y,SlopeThreshold,AmpThreshold,smoothwidth,peakgroup,smoothtype) % function % P=findpeaks(x,y,SlopeThreshold,AmpThreshold,smoothwidth,peakgroup,smoothtype) % Function to locate the positive peaks in a noisy x-y time series data % set. Detects peaks by looking for downward zero-crossings in the first % derivative that exceed SlopeThreshold. Returns list (P) containing peak % number and position, height, width, area, and start and stop position of % each peak, assuming a Gaussian peak shape. Arguments "slopeThreshold", % "ampThreshold" and "smoothwidth" control peak sensitivity. Higher values % will neglect smaller features. "Smoothwidth" is the width of the smooth % applied before peak detection; larger values ignore narrow peaks. If % smoothwidth=0, no smoothing is performed. "Peakgroup" is the number % points around the top part of the peak that are taken for measurement. If % Peakgroup=0 the local maximum is takes as the peak height and position. % The argument "smoothtype" determines the smooth algorithm: % If smoothtype=1, rectangular (sliding-average or boxcar) % If smoothtype=2, triangular (2 passes of sliding-average) % If smoothtype=3, pseudo-Gaussian (3 passes of sliding-average) % Skip peaks if peak measurement results in NaN values % See http://terpconnect.umd.edu/~toh/spectrum/Smoothing.html and % http://terpconnect.umd.edu/~toh/spectrum/PeakFindingandMeasurement.htm % T. C. O'Haver, Version 1.1, Last revised December, 2013 % % Example: Three noisy Gaussian peaks at x=20,50,80; % all heights=1.0; all widths=10. % x=1:.2:100; % y=gaussian(x,20,10)+gaussian(x,50,10)+gaussian(x,80,10)+.1.*randn(size(x)); % findpeaksGSS(x,y,0.0004,0.3,17,21,3) % Peak Position Height Width Area Start End % 1 20.496 1.0033 9.2482 9.8784 8.5766 32.414 % 2 50.32 1.0235 7.7165 8.4077 40.375 60.265 % 3 79.691 1.0191 10.716 11.626 65.881 93.501 % Same signal without noise: % 1 20 1 10 10.646 7.1122 32.888 % 2 50 1 10 10.646 37.112 62.888 % 3 80 1 10 10.646 67.112 92.888 % % Related functions: % findvalleys.m, findpeaksL.m, findpeaksb.m, findpeaksplot.m, peakstats.m, % findpeaksnr.m, findpeaks.m, findpeaksLSS.m, findpeaksfit.m. if nargin~=7;smoothtype=1;end % smoothtype=1 if not specified in argument if smoothtype>3;smoothtype=3;end if smoothtype<1;smoothtype=1;end smoothwidth=round(smoothwidth); peakgroup=round(peakgroup); if smoothwidth>1, d=fastsmooth(deriv(y),smoothwidth,smoothtype); else d=y; end n=round(peakgroup/2+1); P=[0 0 0 0 0 0 0]; vectorlength=length(y); peak=1; AmpTest=AmpThreshold; for j=2*round(smoothwidth/2)-1:length(y)-smoothwidth, if sign(d(j)) > sign (d(j+1)), % Detects zero-crossing if d(j)-d(j+1) > SlopeThreshold, % if slope of derivative is larger than SlopeThreshold if y(j) > AmpTest, % if height of peak is larger than AmpThreshold xx=zeros(size(peakgroup));yy=zeros(size(peakgroup)); for k=1:peakgroup, % Create sub-group of points near peak groupindex=j+k-n+1; if groupindex<1, groupindex=1;end if groupindex>vectorlength, groupindex=vectorlength;end xx(k)=x(groupindex);yy(k)=y(groupindex); end if peakgroup>3, [Height, Position, Width]=gaussfit(xx,yy); PeakX=real(Position); % Compute peak position and height of fitted parabola PeakY=real(Height); MeasuredWidth=real(Width); % if the peak is too narrow for least-squares technique to work % well, just use the max value of y in the sub-group of points near peak. else PeakY=max(yy); pindex=val2ind(yy,PeakY); PeakX=xx(pindex(1)); MeasuredWidth=0; end % Construct matrix P. One row for each peak % detected, containing the peak number, peak % position (x-value) and peak height (y-value). % If peak measurements fails and results in NaN, skip this % peak if isnan(PeakX) || isnan(PeakY) || PeakY> v=[1 2 3 4 Inf 6 7 Inf 9]; % >> rmnan(v) % ans = % 1 2 3 4 4 6 7 7 9 la=length(a); if isnan(a(1)) || isinf(a(1)),a(1)=0;end for point=1:la, if isnan(a(point)) || isinf(a(point)), a(point)=a(point-1); end end function [Height, Position, Width]=gaussfit(x,y) % Converts y-axis to a log scale, fits a parabola % (quadratic) to the (x,ln(y)) data, then calculates % the position, width, and height of the % Gaussian from the three coefficients of the % quadratic fit. This is accurate only if the data have % no baseline offset (that is, trends to zero far off the % peak) and if there are no zeros or negative values in y. % % Example 1: Simplest Gaussian data set % [Height, Position, Width]=gaussfit([1 2 3],[1 2 1]) % returns Height = 2, Position = 2, Width = 2 % % Example 2: best fit to synthetic noisy Gaussian % x=50:150;y=100.*gaussian(x,100,100)+10.*randn(size(x)); % [Height,Position,Width]=gaussfit(x,y) % returns [Height,Position,Width] clustered around 100,100,100. % % Example 3: plots data set as points and best-fit Gaussian as line % x=[1 2 3 4 5];y=[1 2 2.5 2 1]; % [Height,Position,Width]=gaussfit(x,y); % plot(x,y,'o',linspace(0,8),Height.*gaussian(linspace(0,8),Position,Width)) % Copyright (c) 2012, Thomas C. O'Haver maxy=max(y); for p=1:length(y), if y(p)<(maxy/100),y(p)=maxy/100;end end % for p=1:length(y), z=log(y); coef=polyfit(x,z,2); a=coef(3); b=coef(2); c=coef(1); Height=exp(a-c*(b/(2*c))^2); Position=-b/(2*c); Width=2.35482/(sqrt(2)*sqrt(-c));