Some research on weakly-supervised learning for automatic geohazards recognition

Geohazards recognition is widely used in natural disaster risk management. Manual geohazards recognition is expensive and inefficient. We will talk about why and how to develop deep neural networks for automatic geohazards recognition. First, the challenges and opportunities in identifying geohazards from remote sensing images will be discussed. Second, a deep multiple instance learning model for automatic recognition of landslide hazards will be presented. Third, we will also talk about some results on contrastive learning and transformer for automatic feature extraction of remote sensing images.