The Sentinel-1 and Sentinel-2 images should now be downloaded and ready to use. We will proceed to the creation of a dataset that can be used (i) to train the model described in section 14.3, and also (ii) to evaluate its performance in terms of root mean squared error.
The Sentinel-1 image has 16 bits signed encoding and the Sentinel-2 images have 16 bits unsigned encoding, i.e. integer numbers. We need to normalize the pixels to have absolute values inside the [0, 1] range. As in sections 5.1, 8.3 and 12.12 we use the BandMathX application of the Orfeo ToolBox to perform the band concatenation and the normalization of values.
We process only the 10-m-resolution bands, ...
Get Deep Learning for Remote Sensing Images with Open Source Software now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
