What is it about?
Earth observation satellites and drones collect huge amounts of images of our planet. This data allows for monitoring the environment, tracking ships and aircraft, detecting disasters, and studying climate change. However, sending all raw data back to Earth is difficult because communication bandwidth from space is limited. One promising solution is to process the data directly onboard using Artificial Intelligence (AI). This enables satellites to analyze images in real time and transmit only the most relevant information. In this work, we review academic publications that combine machine learning, remote sensing, and FPGA hardware—reconfigurable chips that can run AI efficiently in constrained environments such as satellites or drones. Following a systematic review methodology, we analyze dozens of studies that deploy AI models on FPGA devices for Earth observation tasks. We organize this research into two taxonomies to help interested researchers find work relevant to them. The first taxonomy is based on the types of Earth observation applications, while the second focuses on the hardware design strategies employed. Each view provides unique insights into the research landscape. Our analysis highlights current trends, common challenges, and emerging solutions for running AI directly on remote sensing platforms.
Featured Image
Photo by Kevin Stadnyk on Unsplash
Why is it important?
Earth observation missions are evolving rapidly. The rise of small satellites and drones means that more sensors are collecting higher-resolution data than ever before. At the same time, the ability to transmit data to Earth has not increased at the same pace. This creates a growing bottleneck. Running AI directly on the sensor platform, also known as AI at the edge, can help solve this problem. Instead of sending all images to Earth, satellites can analyze them in orbit and transmit only useful results, such as detected objects, environmental changes, or compressed information. FPGAs are particularly well-suited for this task because they offer a balance between performance, energy efficiency, and flexibility. However, deploying machine learning models on such hardware remains challenging. Our review is the first to systematically analyze the intersection of machine learning, remote sensing, and FPGA hardware. By summarizing existing work, identifying research gaps, and providing guidelines for future research, we aim to help scientists and engineers design the next generation of intelligent Earth observation systems.
Perspectives
Writing this survey was a rewarding experience because it brought together three research areas that often evolve separately: machine learning, earth observation, and hardware design. Being more familiar with machine learning than remote sensing and FPGA designing it was enriching to view problems and specific questions from different perspectives. On top of that, I enjoyed writing a systematic review following the PRISMA 2020 guidelines. Following such a structured and carefully designed method helps staynig organized and analyzing articles and experiments as fairly as possible. Overall, it was a very enriching experience that, even though it takes a considerable amount of work, I would recommend to everyone during a PhD.
Cedric Leonard
German Aerospace Center (DLR)
Read the Original
This page is a summary of: FPGA-Enabled Machine Learning Applications in Earth Observation: A Systematic Review, ACM Computing Surveys, March 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3800686.
You can read the full text:
Resources
Contributors
The following have contributed to this page
