Global Land Cover Mapping with Weak Supervision: Outcome of the 2020 IEEE GRSS Data Fusion Contest
Caleb Robinson, Kolya Malkin, Nebojsa Jojic, Huijun Chen, Rongjun Qin, Changlin Xiao, Michael Schmitt, Pedram Ghamisi, Ronny Hansch, and Naoto Yokoya
This paper presents the results of the annual Data Fusion Contest organized by the Image Analysis and Data Fusion Technical Committee of the IEEE Geoscience and Remote Sensing Society.
Global land cover maps allow monitoring the Earth’s surface and tracking global phenomena, such as climate change and natural disasters. In the field of remote sensing, low resolution global maps are regularly updated and openly available. These data are collected by different sensors and are of different type and resolution. Obtaining an accurate land cover classification by processing these heterogeneous multi-sensor data remains a fundamental challenge. The 2020 Data Fusion Contest addresses the problem of automatic land cover mapping from global satellite data with weak supervision.
In the 2020 contest, we used SEN12MS dataset for training land cover classification models and a simplified version of the International Geosphere-Biosphere Programme (IGBP) classification scheme. For testing and validation, semi-manually derived high-resolution land cover maps not included in SEN12MS were provided without geolocation information. The goal of the competition was to predict land cover labels at the ground sampling distance of 10 m. The contest included two competition tracks: in track 1, only low-resolution noise labels were used for training; in track 2, a limited number of high-resolution clean labels were provided. We report the competition results and discuss the highest-ranked approaches.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021-03-04.