Random Forest Classifier - Feature error by applying a RFC

Hello SNAP development team,

I would like apply a SNAP trained-and-saved random forest classifier on another product.

Thereby I get the following error mesage in this process: Do not found the features.
But the product contains the features and the band names are also equivalent to the names of the features in the classifier.

Does anybody have an idea? Thanks in advance.

Kind regards,

In the xml file that is saved for the classifier (it can be found in “.snap\auxdata\classifiers\RandomForest”), what are the feature names inside the tag “featureNames”? What are the band names in the product that you are trying to classify with the classifier loaded from file?

Please find uploaded the xml file and the product dim file for further assessment!
(band names and feature names are equal)

S2TestClassifier.xml (1.3 KB)
PixEx_S2_MSI_NN_ALL_20171006_Test.dim (20.8 KB)

The featureNames do match the band names in the product. However, I notice some irregularities in S2TestClassifier.xml. I assume the classification is “Train on Vectors”? There are six values in sortedClasses but only one value in trainingBands. sortedClasses should match trainingBands. E.g., if there are only 3 training vectors: dark, middle and light, then in sortedClasses, there would be three values: 0.0, 1.0, 2.0 and in trainingBands, there would be three values dark, middle and light.

How did you create the classifier? What are the Training vectors listed in the RF classifier UI? Which ones did you choose?

I have used - train on raster.

Please find uploaded the xml file of the classifier, a snapshot of the SNAP for the training and for the usage of the classifier.

S2_NN_Classifier.xml (1.3 KB)

Thank you for the information. I will look into it.

Hello @CWong - are there any news regarding this issue?