This letter reports an experimental study aimed at establishing the questionable usefulness of scattering attributes for ship classification from moderate-resolution SAR images. About 2700 example images representing four ship types have been extracted from the OpenSARShip annotated data set and used to form the training and test sets for random forest models. After importance ranking and cross-validation, different subsets of both geometric and scattering attributes were selected from a fixed training set and used to train the classifier. The results from the validation using the test sets show that the scattering attributes give a significant contribution in terms of overall classification accuracy.
Using low-resolution SAR scattering features for ship classification
Salerno E
2022
Abstract
This letter reports an experimental study aimed at establishing the questionable usefulness of scattering attributes for ship classification from moderate-resolution SAR images. About 2700 example images representing four ship types have been extracted from the OpenSARShip annotated data set and used to form the training and test sets for random forest models. After importance ranking and cross-validation, different subsets of both geometric and scattering attributes were selected from a fixed training set and used to train the classifier. The results from the validation using the test sets show that the scattering attributes give a significant contribution in terms of overall classification accuracy.File | Dimensione | Formato | |
---|---|---|---|
prod_468769-doc_189566.pdf
accesso aperto
Descrizione: postprint - Using_low-resolution_SAR_scattering_features_for_ship_classification
Tipologia:
Versione Editoriale (PDF)
Dimensione
273.82 kB
Formato
Adobe PDF
|
273.82 kB | Adobe PDF | Visualizza/Apri |
prod_468769-doc_192563.pdf
accesso aperto
Descrizione: Using_low-resolution_SAR_scattering_features_for_ship_classification
Tipologia:
Versione Editoriale (PDF)
Dimensione
621.18 kB
Formato
Adobe PDF
|
621.18 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.