Available sensor transformations

The classifiers included in the SpecHomo package

The SpecHomo package provides a set of classifiers that can be used for spectral homogenization of various sensor combinations (see here). These classifiers have been trained based on simulated multispectral remote sensing images created by spectral convolution from hyperspectral airborne datasets. These hyperspectral datasets were chosen with the intention to include as many different land-cover types as possible in order to maximize spectral variability. This makes these classifiers suitable for many application fields. Details on the underlying datasets and the used algorithms to create the classifiers can be found in Scheffler et al. 2020.

Which sensor transformations are available?

SpecHomo enables you to perform spectral harmonization for various sensors - in different combinations and with different band compositions. To get a list of all the available transformations, you may use the function list_available_transformations.

Note

You can directly copy/paste the outputs of this function to the method you need for running the homogenization: spechomo.prediction.SpectralHomogenizer.predict_by_machine_learner(). The column names of the returned table are consistent with this method.

Here is the code to get the full list of available transformations:

>>> from spechomo.utils import list_available_transformations
>>> list_available_transformations()

The output may also be directly filtered using the input parameters of the function. Here is an example:

>>> from spechomo.utils import list_available_transformations
>>> list_available_transformations(method='LR', src_sat='Landsat-8', n_clusters=50)

Note

If you have trained your own classifiers for homogenization, you may also pass a different classifier root directory to list_available_transformations. Use the parameter classifier_rootDir for this.