Abstract
Abstract. Fully automatic large-scale land cover mapping belongs to the core challenges addressed by the remote sensing community. Usually, the basis of this task is formed by (supervised) machine learning models. However, in spite of recent growth in the availability of satellite observations, accurate training data remains comparably scarce. On the other hand, numerous global land cover products exist and can be accessed often free-of-charge. Unfortunately, these maps are typically of a much lower resolution than modern day satellite imagery. Besides, they always come with a significant amount of noise, as they cannot be considered ground truth, but are products of previous (semi-)automatic prediction tasks. Therefore, this paper seeks to make a case for the application of weakly supervised learning strategies to get the most out of available data sources and achieve progress in high-resolution large-scale land cover mapping. Challenges and opportunities are discussed based on the SEN12MS dataset, for which also some baseline results are shown. These baselines indicate that there is still a lot of potential for dedicated approaches designed to deal with remote sensing-specific forms of weak supervision.
Highlights
The problem of automatic land cover mapping from remote sensing imagery is traditionally cast as a machine learning task, especially when applied to large study areas (Cihlar, 2000)
We focus on the freely available global imagery provided by the Sentinel-1 and Sentinel-2 missions of the European Copernicus program (Torres et al, 2012, Drusch et al, 2012) and a simplified version of the land cover classification scheme of the International Geosphere-Biosphere Programme (IGBP) (Loveland, Belward, 1997), which is reflected by the SEN12MS dataset (Schmitt et al, 2019) and the 2020 IEEE-GRSS Data Fusion Contest (DFC2020) (Yokoya et al, 2020)
While land cover maps are traditionally assessed via the overall accuracy (OA) measure, we propose to use the less optimistic average accuracy (AA) for comparison, as it gives less weight to large classes, which are rather simple to classify, e.g. Forest and Water
Summary
The problem of automatic land cover mapping from remote sensing imagery is traditionally cast as a (supervised) machine learning task, especially when applied to large study areas (Cihlar, 2000). Manifold large-scale land cover datasets already exist, all of which are the result of (semi-)automated processes themselves This introduces weakly supervised learning as a promising strategy to train well-generalizing models on available data – even if the labels come with a significant error bar or at comparably low resolutions. We discuss the problem of weakly supervised learning of models for land cover prediction from satellite data For this purpose, we focus on the freely available global imagery provided by the Sentinel-1 and Sentinel-2 missions of the European Copernicus program (Torres et al, 2012, Drusch et al, 2012) and a simplified version of the land cover classification scheme of the International Geosphere-Biosphere Programme (IGBP) (Loveland, Belward, 1997), which is reflected by the SEN12MS dataset (Schmitt et al, 2019) and the 2020 IEEE-GRSS Data Fusion Contest (DFC2020) (Yokoya et al, 2020). Besides a description of the challenge and how SEN12MS and DFC2020 are addressing it, baseline results using off-the-shelf deep learning models are provided to highlight the importance of dedicated research
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
