HETEROGENEOUS IMAGE CLASSIFICATION WITH MULTI-STAGE CONDITIONAL ADVERSARIAL DOMAIN ADAPTATION BETWEEN SAR AND OPTICAL IMAGERY

In this paper, we deal with the problem of heterogeneous image classifiers transferring between SAR and optical imagery through a novel multi-stage domain adaptation technique. The problem of transferring the optical image classifier to SAR and vice-versa is of practical importance because it allows us to leverage plenty of labelled data in the source domain for the target domain task, but gains little attention. Because there is a drastic distribution-gap between both the optical and SAR imaging modalities, it is non-trivial to apply domain adaption directly. We propose a multi-stage adversarial feature alignment procedure that firstly performs global adversarial feature alignment and then a class-conditional adversarial feature alignment is conducted to further enable class-discriminative feature adaption. The proposed method is validated on the SEN12MS dataset, and some discussions are provided about heterogeneous domain adaption between SAR and optical imagery.