finding accurate number of peaks and eleminate unwanted peaks

HI

I'm trying to find peaks in this data :

the actual data has 10 peaks as shown in the image below (peaks are in rectangle)

. I applied sgf filter got the peaks below

     order = 7;
     framelen =15;
     x=data;
     lx = 34;
     sgf = sgolayfilt(x,order,framelen);

    plot(x);
    hold on;
   plot(sgf);

   sgf=-sgf;
   [pks_smoothed,locs_smoothed,widths_smoothed,proms_smoothed] = findpeaks(sgf,'MinPeakProminence',0.05);
   pks_smoothed = -pks_smoothed;
   plot(locs_smoothed,pks_smoothed,'g*');
   text(locs_smoothed+.02,pks_smoothed,num2str((1:numel(pks_smoothed))'))

it detects many unwanted peaks how can I find the required peaks accurately and get rid of unwanted peaks ?

[Adam]

That seems like a very small MinPeakProminence value

Answer 1

[Greg Dionne]

Try:

findpeaks(sgf,'MinPeakProminence',1.2);

[sarmad m]

I tried , it improves the result . is there is a another way to obtain the correct number of peaks rather than changing MinPeakProminence values ?

[Adam]

Well, you can just set the 'NPeaks' property to 10, but whether it picks the 10 you want or not is another matter. You may still have to tweak around with some of the other options for findpeaks.

[Greg Dionne]

An alternative is to perhaps take the output of your widths_smoothed and prom_smoothed and come up with some heuristic that matches the most of the peaks you care about then filter out the rest based upon that somehow. Some of your peaks look kind of "flat". Maybe use a combination of medfilt1() and diff() to identify the edges.