Ok, so I will be very precise about the query. I have classified some satellite images as having multiple bands, like more than 20 bands. These bands include the usual multispectral bands as well as some spectral indices generated from these bands. I have stacked them all together and used for the classification purpose. I have used a random forest classifier. Now I want to know the contribution of each band from the stacked image in the classification process. How do I calculate that? Any guidance would be helpful.
Thanks in advance.
Please, use principle component analysis - PCA to find the contribution of each band.
Sher Bahadur Gurung Thank you Sir for your kind response. Can you please explain it a bit more? I want to know, how much contribution does each band has in the classification process.
* In the scikit-learn machine learning Python library, there are two options:
1) "feature_importances_" witch is an attribute to ensemble classifiers (e.g., RandomForestClassifier: https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier.feature_importances_);
2) "permutation_importance" (https://scikit-learn.org/stable/modules/generated/sklearn.inspection.permutation_importance.html#sklearn.inspection.permutation_importance).
* In Google Earth Engine, you can use: var variable_importance = ee.Feature(null, ee.Dictionary(classifier.explain()).get('importance'));
Thanks, everyone! I would like to mention here, that I performed the classification using the Orfeo Toolbox of QGIS 3.16. Can you please give some idea to use that software to know the contributions of the band? Alessandro Ferrarini and Abdelrazek Elnashar
Variable importance in TrainVectorClassifier is not possible: https://forum.orfeo-toolbox.org/t/random-forests-variable-importance-in-otb/322
Swadhina Koley thanks for your question. The answers also interest me.
- Badges
- Science topic
- Similar topics
- Computer Science
- Research Topics