Change detection (step) in timeseries signal

Hello all,

I have a signal where I have to detect particular points where an event happens. I have tried using the usual matlab algos of findchange points to no avail. I have attached images of the signal and at which points I want to detect.

What I would like to detect is to detect the points on the thin stem just before it slides down to the bigger stem. Any pointers will be greatly appreciated.

Answer 1

[Star Strider]

You can likely get close to that point using islocalmax and particularly Flat Maxima Regions. It might be necessary to use smoothdata first, if noise is a problem.

[Daniel Adeniyi]

Thanks for your response. I dont know how can i smooth the thin stems.

The challenge is that i have a bunch of signals like these and the thickness of the upper stem varies and i need to detect the values just before a descent into a new level

[Star Strider]

My pleasure!

The idea is not to smooth the stems, but to smooth the flat sections so that islocalmax does not have any problem in detecting the beginning of the flat section near the stems. The point you want is likely one index position less that the beginning of the flat section.