A Cycle GAN Approach for Heterogeneous Domain Adaptation in Land Use Classification

Claire Voreiter (1) , Jean-Christophe Burnel (1) , Pierre Lassalle (2) , Marc Spigai (3) , Romain Hugues (3) , Nicolas Courty (4, 1)

1 OBELIX - Observation de l’environnement par imagerie complexe
2 CNES - Centre National d'Études Spatiales [Toulouse]
3 TAS - Thales Alenia Space [Toulouse]
4 FT R&D - France Télécom Recherche & Développement

In the field of remote sensing and more specifically in Earth Observation, new data are available every day, coming from different sensors. Leveraging on those data in classification tasks comes at the price of intense labelling tasks that are not realistic in operational settings. While domain adaptation could be useful to counterbalance this problem, most of the usual methods assume that the data to adapt are comparable (they belong to the same metric space), which is not the case when multiple sensors are at stake. Heterogeneous domain adaptation methods are a particular solution to this problem. We present a novel method to deal with such cases, based on a modified cycleGAN version that incorporates classification losses and a metric space alignment term. We demonstrate its power on a land use classification tasks, with images from both Google Earth and Sentinel-2.

Mots clés: Land Use Classification; Cycle GAN; Heterogeneous Domain Adaptation

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

Claire Voreiter, Jean-Christophe Burnel, Pierre Lassalle, Marc Spigai, Romain Hugues, et al.. A Cycle GAN Approach for Heterogeneous Domain Adaptation in Land Use Classification. 2020. ⟨hal-02537135⟩