Aggregated primary detectors for generic change detection in satellite images

Vincent Jean Michel Vidal (1, 2) , Matthieu Limbert (1) , Tugdual Ceillier (1) , Lionel Moisan (2)

1 SAS - Earthcube
2 MAP5 - UMR 8145 - Mathématiques Appliquées Paris 5

Detecting changes between two satellite images of the same scene generally requires an accurate (and thus often uneasy to obtain) model discriminating relevant changes from irrelevant ones. We here present a generic method, based on the definition of four different a-contrario detection models (associated to arbitrary features), whose aggregation is then trained from specific examples with gradient boosting. The results we present are encouraging, and in particular the low false positive rate is noticeable.

Mots clés: change detection; A-contrario methodology; Very high resolution satellite imagery; supervised learning; image comparison

Domaines: Traitement des images [eess.IV]; Intelligence artificielle [cs.AI]

Vincent Jean Michel Vidal, Matthieu Limbert, Tugdual Ceillier, Lionel Moisan. Aggregated primary detectors for generic change detection in satellite images. 2019 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Jul 2019, Yokohama, Japan. ⟨hal-02351557⟩