Weakly Supervised Learning for Earth Observation

Abstract

Earth observation provides a rich source of remotely sensed information for a variety of applications such as global land cover monitoring, environmental impact due to natural disasters, and transformation of urban spaces. Although advances in deep learning provide an agile vehicle to leverage petabytes of earth observation data, it remains a challenge to generate accurate annotation for supervised learning schemes. The presentation taps into our recent works towards solutions. Specifically, we - demonstrate the value of rule-based, auto-generated labels from airborne laser measurements [1,2], and - elaborate on optical-radar sensor feature representation generation through the framework of self-supervised learning for models that require little amount of data labels. [1] C. Albrecht, F. Marianno, and L. Klein, IEEE Big Data conference (2021). [2] C. Albrecht, C. Liu, Y. Wang, L. Klein, X. Zhu, IGARSS conference (2022). [2] Y. Wang, C. Albrecht, and X. Zhu, IGARSS conference (2022).