We explore the capabilities of random forest models to classify several types of ships imaged through a satellite-borne C-band SAR with 20m spatial resolution. A number of attribute subsets estimated from the Sentinel 1 images provided by the OpenSARShip public data set are used to train models that are then tested against never-seen-before data. A vast data set has been extracted from OpenSARShip and used to estimate the whole attribute set, composed of 8 naive geometrical features and 8 scattering features. The results are encouraging, as the performances obtained seem to be good when compared to other results from non-deep-learning classifiers reported in the literature. Against previous claims found in the literature, the advantages of adding scattering features to purely geometric ones is here confirmed.
Testing random-forest models trained by Sentinel-1 data from the OpenSARShip data set
Salerno E
2022
Abstract
We explore the capabilities of random forest models to classify several types of ships imaged through a satellite-borne C-band SAR with 20m spatial resolution. A number of attribute subsets estimated from the Sentinel 1 images provided by the OpenSARShip public data set are used to train models that are then tested against never-seen-before data. A vast data set has been extracted from OpenSARShip and used to estimate the whole attribute set, composed of 8 naive geometrical features and 8 scattering features. The results are encouraging, as the performances obtained seem to be good when compared to other results from non-deep-learning classifiers reported in the literature. Against previous claims found in the literature, the advantages of adding scattering features to purely geometric ones is here confirmed.File | Dimensione | Formato | |
---|---|---|---|
prod_466675-doc_183530.pdf
non disponibili
Descrizione: Ranfor_3.pdf
Dimensione
198.24 kB
Formato
Adobe PDF
|
198.24 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.