FindPeaks() of a 1024 x 116 Matrix

I have a matrix that I read into MatLab that is 116 spectra (columns) each being 1024 elements (rows). Each spectrum has a peak and I'm looking to determine the following information from each spectrum:

- Peak value

- Peak position

- Peak width at a set height, like 10% or 30% of the peak height. What I'd really like is to know the first and last x element of each peak

I've completed the Signal Processing Toolbox course and it didn't go over the FindPeaks() function. Please, any help would be greatly appreciate. Even if it's direction to a good source on the function. Thank you!

Answer 1

[dpb]

findpeaks only works on vector input; just use a loop and pass each column of your array in turn.

There's a fair amount of background information and examples of using findpeaks in the documentation; I'm unaware of any other tutorial but the web if full of stuff.

I'd suggest "just try it" and see what you get. If the spectra are only one-peaked, it seems it could hardly go wrong.

ADDENDUM

Since findpeaks can return anything from an empty set to some large number of peaks, you'll probably want to use a cell array to store the results initially or else wrap the call into a routine that processes the results of each column to ensure expected results rather than just presume every column will give one and only one peak. Of course, if the data are very clean, that may actually happen....I was rarely that lucky. :)

[Christopher Gordon]

Thank you so much. And please bear with me as I'm a novice and I'm working off existing code. This is the code that I've tried to make a loop to find the peak value, peak location, and peak width at a given threashold. Then, I want to normalize each curve to the peak intensity.

% do calc ourselves starting from the raw data, the raw calibration curves
AllCurves = readmatrix('770UVVIS_Allcurves.csv');
[npix, ncurve]=size(AllCurves); %number of pixels and calibraton curves
%find peak pixel positions for calibration curves as well as left and right
%threshold position (for figuring width of peak and IB)
curvepeaks=zeros( ncurve,3); %reserve memory to hold the returned info
for curve=1:ncurve
    %for each curve, get the info on peak, IB range, peak int, etc.
    [peak,locs,width] = findpeaks(AllCurves,x,"Threashold",0.3);
    curveinfo(curve,:)=[pks,locs,width];
    AllCurves(:,curve)=AllCurves(:,curve)/peak; %normalize so that peak integrated intensity is 1; no further normalization needed
end

I'm getting an error that it's not recognizing 'x', but I thought that would perform the peakfinds() values relative to the x variable. Could you help me with what I'm doing wrong?

[dpb]

You don't have any 'x' variable.

What is the content of the file? Does it have a frequency or time vector associated with the spectral data? You've got to tell findpeaks what 'x' is if it's supposed to know anything about it.

The parameter 'Threshold' is misspelled.

Also, you have to pass the column as a vector, not the full array; findpeaks works only on vectors; since it can return a variable size of output arrays they did not vectorize it.

[peak,locs,width] = findpeaks(AllCurves(:,curve),"Threshold",0.3);

would run and return the locations in terms of ordinal position; is you really have an x, then define it and put it back.

Then

curveinfo(curve,:)=[pks,locs,width];

uses pks instead of peak so that won't work and does assume there will be one and only one peak.

AllCurves(:,curve)=AllCurves(:,curve)/peak; 

will just redefine your input array; you really want to do that? It's OK if that is what you really intend, but you could also wait until done and then

AllCurves=AllCurves./curveinfo(:,1); 

which would be "the MATLAB way" of using vectorized operations. NB: the "dot" division operator.

This assumes, of course, that all the other assumptions are met about one and only one peak being found.

[Christopher Gordon]

Thank you so much for your awesome feedback. I honestly appreciate it so much!

[Christopher Gordon]

Sorry, just a quick follow up. For the threshold, is that a fraction of the total signal, or do I need to specify an exact value? Each curve has one main peak and may have smaller features, but the peak will be an order or more larger in magnitude.

[dpb]

'Threshold' is absolute but it's probably not what you want here; it's more for isolating a peak from multiple sidelobe-type peaks that are relatively close in amplitude.

Since you think your spectra are just one dominant peak and want only it, I'd think

[peak,locs,width] = findpeaks(AllCurves(:,curve),'NPeaks',1,'SortStr','descend');

would be the way to go. The documentation for 'NPeaks' is a little misleading, perhaps, in that one might infer that only the first peak positionally would be returned which would be true if didn't also use the illnamed 'Sortstr' parameter along with it.

The doc should make a note that when used together all peaks will be found, sorted and then only the requested number returned, sorted as requested. 'none' as the choice there is the same as if didn't use the 'Sortstr' parameter at all, it will return in the order in the input vector.