[Octave-bug-tracker] [bug #63987] [octave forge] (signal) findpeaks mishandling flat valleys

[John Donoghue]

Follow-up Comment #4, bug #63987 (project octave):

The flat valley gets pulled in with:


## <= in 1st derivative includes the case of oversampled signals.
  idx = find (df1.*[df1(2:end); 0] <= 0 & [df2(2:end); 0] < 0);


But changing the <= to < then causes the oversampled test to fail

If I add a check later on where it loops through each peak, and filter it
out:


diff -r 31c4a47822c1 inst/findpeaks.m
--- a/inst/findpeaks.m  Mon Jun 05 08:33:32 2023 -0400
+++ b/inst/findpeaks.m  Tue Jun 27 09:20:05 2023 -0400
@@ -266,7 +266,11 @@
            ceil (min(idx(i)+minD/2,np))).';
     pp      = zeros (1,3);
     # If current peak is not local maxima, then fit parabola to neighbor
-    if any (data(ind) > data(idx(i)))
+    if any (data(idx(i)-1) == data(idx(i)))
+      # sample on left same as peak
+      xm    = 0;
+      pp    = ones (1,3);
+    elseif any (data(ind) > data(idx(i)))
       pp = polyfit (ind, data(ind), 2);
       xm = -pp(2)^2 / (2*pp(1));   # position of extrema
       H  = polyval (pp, xm);      # value at extrema


It passes the current tests as well as the fail of this bug


    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?63987>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/