Support Vector Machines for burnt area discrimination

Olivier Zammit (1) , Xavier Descombes (1) , Josiane Zerubia (1)

1 ARIANA - Inverse problems in earth monitoring

This report addresses the problem of burnt area discrimination using remote sensing images. The detection is based on a single post-fire image acquired by SPOT 5 satellite. To delineate the burnt areas, we use a recent classification method called Support Vectors Machines (SVM). This approach is compared to more conventional classifiers such as K-means or K-nearest neighbours which are widely used in image processing. We also proposed a new automatic classification approach combining K-means and SVM. The results given by the different methods are finally compared to ground truths on various burnt areas.

Mots clés: forest fire; burnt areas; remote sensing; high resolution satellite images; Support Vectors Machines; K-means; K-nearest neighbours; classification

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Olivier Zammit, Xavier Descombes, Josiane Zerubia. Support Vector Machines for burnt area discrimination. [Research Report] RR-6343, INRIA. 2007, pp.37. ⟨inria-00185101v2⟩