Removing peaks less than threshold

Hi,

I have a time series which has fairly large cyclical peaks (~1sec) that I used the findpeaks function to identify. However, there are sometimes lesser spurious peaks which I am trying to eliminate. I've tried to set a threshold for minimum peak distance of 0.7sec using the diff function to get rid of the smaller peaks, but this has sometimes resulted in also removing the larger peaks that I would like to keep. I know I need to use the y-values from the peaks to keep the larger values but am unsure how to incorporate that into the algorithm. For example, given this data of peaks:

[LMaxValues2,LMaxInd2] = findpeaks(data.Y,'MinPeakProminence',0.1,'Annotate','extents'); %find data peaks
LMaxTimes2 = data.time(LMaxInd2); %turn peak indices into timestamps

badpts = diff(LMaxTimes2)<0.7; %threshold 0.7s to eliminate minor spurious peaks
LMaxTimes2(badpts)=[]; %Eliminate times of fake peaks from array
LMaxValues2(badpts)=[]; %Eliminate values of fake peaks from array

LMaxTimes2 = [292.498 293.517 294.543 295.117 295.556 296.548];

LMaxValues2 = [0.0097 -0.027 0.0412 -0.4127 0.0111 -0.011];

I want it to keep points 1,2,3,4,6, outputting:

LMaxTimes2 = [292.498 293.517 294.543 295.556 296.548]

LMaxValues2 = [0.0097 -0.027 0.0412 0.0111 -0.011];

But I'm getting:

LMaxTimes2 = [292.498 293.517 294.543 296.548]

Since the diff(LMaxTimes2) for points 3-4 and 4-5 both cross the <0.7 threshold.

Hope this was clear, thank you!

[Image Analyst]

Do you just want to find the shorter peaks on the left shoulder of the big peak and delete everything in the x and y arrays between the valleys on each side of the shoulder peak? Or did you want to change the values, like interpolate a smooth function between the valleys?

[Trevor Evans]

I only wanted to extract the x-values of the big peaks, but the idea of smoothing the data between peaks is interesting! I ended up doing a variation of what Voss suggested in his answer. Thanks anyway!

Answer 1

[Voss]

Try using MinPeakHeight and/or MinPeakDistance in findpeaks, instead of MinPeakProminence. You'll likely be able to entirely avoid removing elements afterward based on diff.

[Trevor Evans]

I tweaked the MinPeakDistance value and ended up with the desired results without using diff like you said. Thanks!

[Voss]

You're welcome!