IN FOCUS: Data Fusion Contest Promotes Innovative Solutions

By Joanne Van Voorhis

IEEE GRSS is committed to supporting events and activities that both engage members and encourage collaboration to further its fields of focus. A perfect example is the annual Data Fusion Contest, which was initiated almost two decades ago by the GRSS Image Analysis and Data Fusion Technical Committee (IADF TC) to promote interdisciplinary research on geospatial artificial intelligence for social good. The IADF TC is an international network of scientists working on earth observation, geospatial data fusion, and algorithms for image analysis.

The annual IADF Data Fusion contest fosters greater engagement at the intersection of compelling applications of geospatial imaging and machine learning/computer vision. The contest provides standardized evaluation datasets, promotes best practices, and encourages innovation in combining multi-sensor Earth Observation data. Contest participants have a unique opportunity to test out their ideas on emerging and rich datasets and go beyond benchmarks. In fact, these data have often become established benchmarks within the GRSS and broader research community in subsequent years. The first contest in 2006 focused on pansharpening algorithms. In subsequent years, it expanded to various topics like land cover mapping, building extraction, flood detection, etc. Interestingly, the contest has often focused on covering broader and more socially relevant topics like detecting settlements without electricity (DFC 2021) using multi-modal and multi-temporal remote sensing data.

Focus on SAR and Optical EO Data

The 2025 IEEE GRSS Data Fusion Contest is organized by IADF TC in collaboration with the University of Tokyo, RIKEN, and ETH Zurich, and fosters the development of innovative solutions for all-weather land-cover and building damage mapping using multimodal SAR and optical EO data at submeter resolution. The contest includes two tracks focusing on land cover types and building damage, respectively, and offers two main technical challenges: effective integration of multimodal data and the handling of noisy labels.This contest is advancing the design and validation of emerging deep learning architectures (models) for robust analysis of multimodal SAR and optical EO data. Image analysis tasks such as these are inherently challenging, and advances to models that can better tackle them can have far-reaching positive impacts.

Real World Applications

With rapid advances in small Synthetic Aperture Radar (SAR) satellite technology, Earth Observation now provides submeter-resolution all-weather mapping with increasing temporal resolution. But while optical data offers intuitive visuals and fine detail, it is frequently limited by weather and lighting conditions. Alternatively, SAR has the ability to penetrate cloud cover and provide consistent imagery in challenging weather and various lighting conditions. These capabilities can provide more efficient and frequent monitoring which can make a huge difference when responding to disasters or working in rapidly changing environments. However, effectively exploiting the complementary properties of SAR and optical data to solve complex remote sensing image analysis problems remains a significant technical challenge. The capability to reliably analyze such imagery per modality and in tandem can significantly advance the tools available to first-responders as well as policy makers to quantify, respond and plan for disasters.

Collective Group Effort

Dr. Saurabh Prasad, Director of the Machine Learning and Signal Processing Lab at the University of Houston, and IADF TC Co-Chair, explains how the contest is truly a collaborative effort. “The challenge takes a collective and coordinated group effort to ensure smooth operation and impactful challenges.” The committee receives proposals from the community outlining their interest in hosting such challenges which are reviewed until a partner team is selected to work closely with the TC organizing the contest. “A large part of the effort involves curating extensive datasets with labels,” Dr. Prasad explains. “The challenge is hosted on a portal where users can download the data and post their results. Winning teams are ranked based on metrics pertinent to the challenge at hand and following a check to ensure compliance with contest rules. The selected teams have an opportunity to document details of their algorithmic approaches in a collaborative journal article.” Prasad notes that there are many tasks managed by dedicated volunteers to ensure the contest runs smoothly each year. “Each year it is impressive to see the IADF TC serve as a melting pot of volunteers who work with the committee to deliver something so rewarding to the community,” he adds.

2025 Winners Recently Announced

This year’s contest was held with a six-week timeline starting in mid-January with the release of training and validation data, a deadline for a short description of an approach at the end of February, and the release of test data in early March. This year’s winners will present papers this August at the GRSS flagship conference IGARSS 2025 in Australia, and have an opportunity to co-author a journal paper summarizing the DFC25 outcomes, which will be submitted with open access to IEEE JSTARS. Details about this year’s winners are available on the GRSS website.

Benefits to Planners and Participants

Part of the reason GRSS supports the effort is that all participants – not just the winners — benefit from the contest. Dr. Prasad explains that “contest participants and winners benefit from engaging with the broader community of like-minded scholars working in areas of mutual interest. This engagement promotes greater cross-fertilization of ideas. Contest winners get to showcase the algorithms and approaches they developed to a broad audience and get visibility to an audience deeply interested in such developments.”

The contest offers students and early-career researchers a chance to hone skills in data fusion, machine learning, remote sensing, and image analysis, preparing them for research or industry roles. It connects them with a global community of geoscience experts, fostering collaboration and potential mentorship. Plus, because it is open to non-IEEE members, it promotes broad participation and inclusivity. “I think everyone benefits,” Prasad adds, “the volunteers who contribute their time and effort to make the contest possible — and the participants who engage with the challenge.”

Links:

• Find out more about the 2025 Data Fusion Contest
• Explore data and results from prior Data Fusion Contests 
• Find out more about the work of the IADF TC 
• Join IADF TC free of cost