This paper addresses oil spill detection from remotely sensed optical images. In particular, it focuses on the automatic classification of regions of interest (ROIs) in two classes, namely oil spills or look-alikes. Candidate regions and the corresponding boundaries have been manually identified from full resolution Moderate Resolution Imaging Spectroradiometer images, related to the Mediterranean Sea over the years 2008 and 2009. Then, a set of features has been extracted from each ROI, allowing to formulate the oil spill detection problem as a two-class classification task on the provided regions (i.e. using a supervised learning strategy). Since ROI classification is challenging, some desired characteristics for the classification algorithm are first identified, such as accuracy, robustness, etc. Then, a solution (called SVME) is provided: it is based on an ensemble of incremental/decremental cost-oriented Support Vector Machines, aggregated with the Receiving Operating Characteristic (ROC) convex hull method in the ROC space. Such a solution addresses all the desired characteristics. Finally, the results obtained on the collected dataset are shown. The importance of this study is the devising of a powerful classification technique that may have an impact on optical oil spill detection from space, especially if fused with satellite synthetic aperture radar data. Moreover, it is shown how the proposed system can be used as a decision support tool, to help a junior operator in making more reliable detections.

SVME: an ensemble of support vector machines for detecting oil spills from full resolution MODIS images

COCOCCIONI, MARCO;
2012-01-01

Abstract

This paper addresses oil spill detection from remotely sensed optical images. In particular, it focuses on the automatic classification of regions of interest (ROIs) in two classes, namely oil spills or look-alikes. Candidate regions and the corresponding boundaries have been manually identified from full resolution Moderate Resolution Imaging Spectroradiometer images, related to the Mediterranean Sea over the years 2008 and 2009. Then, a set of features has been extracted from each ROI, allowing to formulate the oil spill detection problem as a two-class classification task on the provided regions (i.e. using a supervised learning strategy). Since ROI classification is challenging, some desired characteristics for the classification algorithm are first identified, such as accuracy, robustness, etc. Then, a solution (called SVME) is provided: it is based on an ensemble of incremental/decremental cost-oriented Support Vector Machines, aggregated with the Receiving Operating Characteristic (ROC) convex hull method in the ROC space. Such a solution addresses all the desired characteristics. Finally, the results obtained on the collected dataset are shown. The importance of this study is the devising of a powerful classification technique that may have an impact on optical oil spill detection from space, especially if fused with satellite synthetic aperture radar data. Moreover, it is shown how the proposed system can be used as a decision support tool, to help a junior operator in making more reliable detections.
2012
Cococcioni, Marco; Corucci, Linda; Masini, Andrea; Nardelli, Fabio
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/154717
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 23
  • ???jsp.display-item.citation.isi??? 20
social impact