IEEE Geoscience and Remote Sensing Magazine - December 2021 - 274

TECHNICAL COMMITTEES
NAOTO YOKOYA, PEDRAM GHAMISI, RONNY HÄNSCH, COLIN PRIEUR, HANA MALHA,
JOCELYN CHANUSSOT, CALEB ROBINSON, KOLYA MALKIN, AND NEBOJSA JOJIC
Report on the 2021 IEEE GRSS Data Fusion
Contest-Geospatial Artificial Intelligence
for Social Good
T
he Image Analysis and
Data Fusion Technical
Committee (IADF TC) of
the IEEE Geoscience and
Remote Sensing Society
(GRSS) has been organizing
the annual Data Fusion
Contest (DFC) since
2006. The contest promotes
the development
of methods for extracting
geospatial information
from large-scale, multisensor,
multimodal, and multitemporal data. It aims to
propose new problem settings that are challenging to
address with existing techniques and to establish new
benchmarks for scientific challenges in remote sensing
image analysis [1]-[5].
THE 2021 DATA FUSION CONTEST
The 2021 IEEE GRSS DFC promoted interdisciplinary research
on geospatial artificial intelligence (AI) for social
good. The ultimate goal of the contest is to build models
to understand the state and changes in the manmade
and natural environment using multisensor and multitemporal
remote sensing data for sustainable development.
This contest was designed as a benchmarking
competition following previous editions [1], [2], [4], [6],
[7]. The 2021 DFC had two tracks running in parallel:
1) Track DSE: detection of settlements without electricity
2) Track MSD: multitemporal semantic change detection.
Track DSE, co-organized by Hewlett Packard Enterprise,
SolarAid, and Data Science Experts, aimed to
automatically detect human settlements without access
to electricity using multimodal, multiresolution, and
Digital Object Identifier 10.1109/MGRS.2021.3121628
Date of current version: 14 January 2022
274
multitemporal satellite remote
sensing data. As input
data, we used Sentinel-1 SAR
data, Sentinel-2, Landsat-8,
and Suomi Visible Infrared
Imaging Radiometer Suite
nighttime images . The
original ground sampling
distance (GSD) ranged
from 10 m to 750 m, but all
images were resampled at
10 m. Semantic labels of
four classes (i.e., settlements
with and without electricity, no settlements with
and without electricity) were provided at a GSD of
500 m for the training data. Participants submitted binary
detection maps of settlements without electricity
with a GSD of 500 m. The classification accuracy was
assessed with the F1 score. The main challenge of Track
DSE was to develop robust and efficient methods for extracting
high-level semantic information from heterogeneous
data.
Track MSD, co-organized by Microsoft AI for
Earth, focused on the automated detection and
classification of land cover change from multitemporal,
multiresolution, and multispectral imagery.
The challenge of Track MSD was to create high-resolution
land cover maps for two time periods using
only low-resolution and noisy land cover labels for
training. Participants were provided with 1) 1-m
multispectral aerial imagery for 2013 and 2017 from
the U.S. Department of Agriculture's National Agriculture
Imagery Program data, 2) 30-m multispectral
satellite imagery (Landsat-8) for five time points
between 2013 and 2017, and 3) 30-m noisy lowresolution
land cover labels for 2013 and 2016 from
the U.S. Geological Survey's National Land Cover
Database data over Maryland. Participants created
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE DECEMBER 2021
IEEE Geoscience and Remote Sensing Magazine - December 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2021
