Problem finding "valleys" in signal

Hello,

I am using the findpeaks function to find the valleys in a signal, but it is not consistent and was wondering if there was a better way of doing this.

My code is below along with an image of a "missed" valley around the 8 second mark. Any help on this would be greatly appreciated.

  [yfft_, freqvec_, yfft_dB_, freq_res] = calcFFT(temp_snip, 'hamming', Fs*100, Fs);

  [fft_pks fft_locs] = findpeaks(yfft_, 'MinPeakHeight', 0.004);
  
  BreathrateHz = freqvec_(fft_locs(2));
  RespRate = (BreathrateHz * snippp)/2
 
  %Get valleys
  inverted_snip = max(temp_snip) - temp_snip;
  snip_lo = max(inverted_snip);
  [v1, vv] = findpeaks(inverted_snip, "MinPeakDistance", Fs*((RespRate)/2), "MinPeakHeight", snip_lo *.85); %-(Fs*BreathInstance_error))/5);

Answer 1

[Steven Lord]

Have you tried using islocalmin on your original data rather than findpeaks on the "flipped" data?

Answer 2

[Image Analyst]

Is it possible your MinPeakDistance is too large?

You could also try to smooth your data with movmean or sgolayfilt before using findpeaks

help sgolayfilt

 SGOLAYFILT Savitzky-Golay Filtering.
    SGOLAYFILT(X,ORDER,FRAMELEN) smooths the signal X using a
    Savitzky-Golay (polynomial) smoothing filter.  The polynomial order,
    ORDER, must be less than the frame length, FRAMELEN, and FRAMELEN must
    be odd.  The length of the input X must be >= FRAMELEN.  If X is a
    matrix, the filtering is done on the columns of X.
 
    Note that if the polynomial order ORDER equals FRAMELEN-1, no smoothing
    will occur.
 
    SGOLAYFILT(X,ORDER,FRAMELEN,WEIGHTS) specifies a weighting vector
    WEIGHTS with length FRAMELEN containing real, positive valued weights
    employed during the least-squares minimization. If not specified, or if
    specified as empty, WEIGHTS defaults to an identity matrix.
 
    SGOLAYFILT(X,ORDER,FRAMELEN,[],DIM) and
    SGOLAYFILT(X,ORDER,FRAMELEN,WEIGHTS,DIM) operate along the dimension DIM.
 
    % Example:
    %   Smooth the mtlb signal by applying a cubic Savitzky-Golay filter 
    %   to data frames of length 41.
    load mtlb                        % Load data
    smtlb = sgolayfilt(mtlb,3,41);   % Apply 3rd-order filter
    plot([mtlb smtlb]);
    legend('Original Data','Filtered Data');
 
    See also SGOLAY, MEDFILT1, FILTER

    Documentation for sgolayfilt
       doc sgolayfilt

Answer 3

[Star Strider]

I prefer using 'MinPeakProminence' instead of 'MinPeakHeight' since that is usually more robust.

t = linspace(0, 10);
y = 0.03*sin(2*pi*t*0.7) + 0.06 + randn(size(t))/100;

[pks,locs] = findpeaks(-y, 'MinPeakDistance', 10, 'MinPeakProminence',0.025);

figure
plot(t, y)
hold on
plot(t(locs), -pks, 'xr')
hold off

Adjust the 'MinPeakProminence' value to get the desired result with your data.

.