Fusion of S1 and S2 using PCA analysis

Microwave Toolbox
#1

Hello, guys!

I’ve been reading the topics about synergic use of S1 and S2 data for land use and land cover analysis. Futhermore, I read the new tutorial by Sir. Andreas Braun, for combination of S1 data (http://step.esa.int/docs/tutorials/S1TBX Synergetic use of S1 (SAR) and S2 (optical) data Tutorial.pdf).

I intend to performe a fusion of S1 and S2 dataset for forest degradation/LULC mapping in amazon region in my final paper for Radar discipline. I’m master’s student.

The point is, can I perform a reverse PCA on SNAP for literaly fusion and not just combine the images in a stack or collocation? I intend to use a supervised classifier and assess the accurcy of fusioned and combined data, just like in this paper http://dx.doi.org/10.1080/15481603.2013.805589

I just know the tool for perform the PCA on stack data and obtain the components. Is it correct to apply the classifier on PCA bands?

Aditional information: I’ll use a SLC image and pre-processing it (apply orbit file, radiometric calibration, thermal noise removal, multilooking, terrain correction and linear to dB) and apply a atmospheric correction and resampling on S2 image.

Thank you for your attention guys!

#2

Combining the bands in a stack is the technical precondition for the PCA in SNAP, but not quite part of the actual fusion (although there are different definitions of what a fusion od SAR and optical data is). The PCA reduces the dimensionality of the stack so that the output is often a combination of both products within the components. Using them as input for a classification can make sense to reduce the amount of data volume and processing time. Others classifiers which are based on permutation, repetition and thresholds, such as random forest, are less benefiting from this, because they rely on large numbers of input features to be effective.

1 Like