[antig]

Hi

I calculated no linear result. Used smart 5. Looking the result I have rised question.
I have two calculated results pixel side by side (pictures). unwrap1 has a peak down after Jul 17. after that graph rase and are quite stable. Graph at unwrap2 near july quite stable down direction and after that stabilization. One of them should be wrong. I suggest that in uwrap1 is poorly unwraped because if mirror graph after july 17 it would be almost same as unwrap2. Additionaly it fits better to the large pictue. Question is. How improve result? Cohherece are good both of them. Pixels are side by side. That is the sign that non linear modeling is good?

• This topic was modified 5 years, 7 months ago by antig.

Attachments:

You must be logged in to view attached files.

[periz]

Hi Antig,
the non-linear model is a time filter (the movement is smoothed out of the displacement phase). This implies a series of things: the result is more flexible and will match more the displacement phase. If the displacement has a non-linear smooth behavior, the model works great. If the displacement is noisy, the model will be more affected by noise. And this might lead to possible phase unwrapping errors (like in your case. by the way, which one is correct should be decided based on local interpretation). Since this model follows closer the displacement phase (whether clean or noisy), the ultimate phase residuals (displacement phase – model) will be smaller, which means, higher coherence. That is to say, higher coherence does not mean in this case that the model is better. It simply means that the model follows better the displacement phase.
How to improve things when there is an uncertainty (when the situation is particularly noisy)? You can take a bigger smart number (=longer smoothing window), or you switch to linear estimation. In most cases the linear estimation works fine and you have signals when something non linear happens. At those exceptions you can carry out a finer analysis with a non linear model.

[Author]

Posts