Self-Supervised Multisensor Change Detection

Most change detection (CD) methods assume that prechange and postchange images are acquired by the same sensor. However, in many real-life scenarios, e.g., natural disasters, it is more practical to use latest available before after occurrence of incidence, which may be using different sensors. In particular, we interested combination optical synthetic aperture radar (SAR) SAR appear vastly from even when capturing scene. Adding this, CD often constrained only target image-pair, no labeled data, additional unlabeled data. Such constraints limit scope traditional supervised machine learning unsupervised generative approaches for multisensor CD. The recent rapid development self-supervised has shown some them can work with few images. Motivated this work, propose a method bitemporal used training network fashion deep clustering contrastive learning. proposed evaluated on four multimodal scenes showing change, benefits our approach demonstrated. Code at https://gitlab.lrz.de/ai4eo/cd/-/tree/main/sarOpticalMultisensorTgrs2021 .