In a recent article about the Future of Neuromorphic AI in Electronic Warfare, Steven Harbour from Parallax Advanced Research wrote that his company were “at the forefront of advancing third-generation AI algorithms, partnering with Intel and Brainchip to develop scalable neuromorphic hardware.”
But it is not only the use of NC in EW they are researching:
“Spiking neural networks are shaping the future of Arctic monitoring”.
A recent collaboration between Parallax Advanced Research, Ohio Aerospace Institute and the University of Dayton Vision Lab resulted in “the first-ever application of a Spiking U-Net architecture for pixelwise classification of Arctic imagery”.
“This innovation is crucial for Arctic missions, where satellite and UAV platforms must operate under extreme conditions with limited energy and bandwidth. By integrating spiking models into the traditionally dense U-Net architecture, our researchers have opened a new frontier in efficient, scalable, and real-time remote sensing.
(…) Accurate segmentation of open water, snow, and meltponds is critical for understanding and modeling Arctic climate dynamics. Meltponds, in particular, lower surface albedo and accelerate ice melt, creating a positive feedback loop that influences global sea-level rise. Monitoring these features in real-time supports navigation safety, wildlife conservation, satellite calibration, and, importantly, global climate models.
Our method enables onboard, low-power processing of Arctic imagery, paving the way for deployment on cubesats, long-endurance UAVs, and polar monitoring missions. This capability is essential for providing time-critical data that informs national and international policy efforts aimed at Arctic preservation and strategic environmental security.”
Replacing “standard convolutional neurons with Integrate-and-Fire (IF) and
Leaky Integrate-and-Fire (LIF) neurons” suggests that future deployment of their model on neuromorphic hardware will involve Loihi, which is not surprising, given Steve Harbour’s long collaboration history with Intel during his years at Southwest Research Institute. He is one among a growing number of neuromorphic researchers who see merit in both Loihi and Akida.
🌍 Pioneering Energy-Efficient Arctic Remote Sensing with Spiking Neural Networks ❄️🚀 We’re thrilled to share the groundbreaking work coming out of Parallax Advanced Research, Ohio Aerospace Institute, and the University of Dayton Vision Lab! Together, we’ve achieved a world first: applying a...
www.linkedin.com
View attachment 86939
Pioneering Energy-Efficient Arctic Remote Sensing with Spiking Neural Networks
Published on
May 29, 2025
At
Parallax Advanced Research and the
Ohio Aerospace Institute (OAI), we are committed to pushing the boundaries of basic and applied science in ways that transform future aerospace and defense capabilities. Our latest collaboration with the
University of Dayton Vision Lab exemplifies this commitment, with a groundbreaking achievement: the first-ever application of a Spiking U-Net architecture for pixelwise classification of Arctic imagery.
Why Spiking Neural Networks for the Arctic?
Conventional deep learning models such as Convolutional Neural Networks (CNNs) have demonstrated high accuracy in image segmentation tasks. However, their compute-intensive nature and high energy demands make them ill-suited for resource-constrained environments like the Arctic. In contrast, spiking neural networks (SNNs) leverage sparse, event-driven computation inspired by biological neurons, drastically reducing power consumption while maintaining analytical precision.
Caption: Dr. Harbour, director of AI Hardware Research, Parallax Advanced Research Center of Excellence and Lead Parallax Research Scientist on this research project.
This innovation is crucial for Arctic missions, where satellite and UAV platforms must operate under extreme conditions with limited energy and bandwidth. By integrating spiking models into the traditionally dense U-Net architecture, our researchers have opened a new frontier in efficient, scalable, and real-time remote sensing.
Advancing Pixelwise Classification: The Spiking U-Net
Our Spiking U-Net preserves the U-Net's powerful encoder-decoder framework and critical skip connections for spatial precision. However, we replace standard convolutional neurons with Integrate-and-Fire (IF) and Leaky Integrate-and-Fire (LIF) neurons, enabling asynchronous, temporally aware processing. In this model, neurons accumulate input over time and "fire" once a threshold is met, mimicking biological synapses.
This design not only cuts energy consumption dramatically but also enhances the model's robustness to the noisy and dynamic conditions characteristic of Arctic data. This marks the first time a U-Net has been successfully adapted into a fully spiking architecture for high-resolution environmental monitoring.
Climate Science and Strategic Monitoring
Accurate segmentation of open water, snow, and meltponds is critical for understanding and modeling Arctic climate dynamics. Meltponds, in particular, lower surface albedo and accelerate ice melt, creating a positive feedback loop that influences global sea-level rise. Monitoring these features in real-time supports navigation safety, wildlife conservation, satellite calibration, and, importantly, global climate models.
Our method enables onboard, low-power processing of Arctic imagery, paving the way for deployment on cubesats, long-endurance UAVs, and polar monitoring missions. This capability is essential for providing time-critical data that informs national and international policy efforts aimed at Arctic preservation and strategic environmental security.
See Fig. 1. Melt ponds on the arctic sea ice by NASA:
- High-resolution imagery provides extensive spatial detail for pixel-wise analysis.
- An autonomous and energy efficient method for meltpond detection is useful for environmental monitoring.
- Would allow for timely calculations of important metrics such as the melting rate of meltponds.
- Additional information about open water and sea ice would be useful for climatic studies.
- Allows researchers to better track the overall activities in the Arctic region.
View attachment 86940
Fig. 1 Melt ponds on the arctic sea ice by NASA
See Fig. 2. Melt ponds on the arctic sea ice 1984 to 2016:
- Rapid climate change is drastically transforming polar environments.
- The Arctic is particularly sensitive, influencing global sea levels and ecosystems
- Melting of sea ice (formation of meltponds), marine life.
- Meltponds have lower albedo causing greater absorption of solar radiations.
- Results in positive feedback loop accelerating the rate of melting of sea ice.
- Accurate monitoring is crucial to track changes in vital classes (e.g., snow, meltponds, open water) and ecological shifts.
View attachment 86941
View attachment 86942
Fig. 2. Melt ponds on the arctic sea ice 1984 to 2016:
A Partnership Forging New Ground
The synergy between the University of Dayton's expertise in computer vision and Parallax/OAI’s strengths in neuromorphic and bio-inspired computation has been a cornerstone of this achievement. Together, we created an interdisciplinary ecosystem capable of pushing SNNs from theoretical constructs into operational remote sensing workflows, with clear implications for both defense and environmental research.
This milestone also represents a historic first for the University of Dayton Vision Lab: the deployment of SNNs in practical, real-world imagery analysis.
Overcoming Technical Barriers
Transitioning traditional convolutional architectures to support time-sensitive spiking neurons presented significant challenges. We successfully addressed these by modifying the U-Net decoder to incorporate LIF/IF neurons and implementing careful training protocols using Norse within the PyTorch framework. These modifications ensured model stability and effective learning, even under the sparsity inherent to spike-based updates.
Looking Ahead
Building on this success, future research will extend our Spiking U-Net to:
- Temporal sequences for dynamic meltpond evolution tracking.
- Multi-spectral data integration to enhance classification richness.
- Neuromorphic hardware deployment to validate real-world energy savings.
- Field deployment on Arctic UAVs and edge compute systems for in-situ monitoring.
Parallax/OAI continue to lead innovations at the intersection of neuromorphic computing, remote sensing, and environmental security. We welcome collaboration with academic institutions, government agencies, and industry partners who share our vision for pioneering resilient, energy-efficient technologies that meet tomorrow's defense and aerospace challenges. To discuss partnership opportunities or learn more about our cutting-edge initiatives, please
contact our research development team today.
###
About Parallax Advanced Research & The Ohio Aerospace Institute (OAI)
Parallax is a 501(c)(3) private nonprofit research institute that tackles global challenges through strategic partnerships with government, industry, and academia. It accelerates innovation, addresses critical global issues, and develops groundbreaking ideas with its partners. With offices in Ohio and Virginia, Parallax aims to deliver new solutions and speed them to market. In 2023, Parallax and OAI formed a collaborative affiliation to drive innovation and technological advancements in Ohio and for the nation. OAI plays a pivotal role in advancing the aerospace industry in Ohio and the nation by fostering collaborations between universities, aerospace industries, and government organizations, and managing aerospace research, education, and workforce development projects.
Tags
DaytonDefenseentrepreneurInnovationResearchSciencesmall business eventsstartupsState of OhioTechnologyUAS Industry
MORE TAGS
The Vision Lab is a Center of Excellence that develops new algorithms and architectures for real-time applications in the areas of signal processing, image processing, computer vision, pattern recognition, artificial neural networks and bio-mimetic object-vision recognition.
udayton.edu
View attachment 86962
semiengineering.com
tspasemiconductor.substack.com
