Energy-Efficient Spike-driven Transformer for Satellite-Streak Segmentation in Space Surveillance

Authors: Iulia Maria Filimon, Radu Danescu

Presentation type: Flash talk at SNUFA 2026 online workshop (4-5 Nov 2026)

Abstract

Recent advances in machine learning and generative AI have increased the need for computational power and energyefficient processing. Neuromorphic computing offers an alternative by using brain-inspired Spiking Neural Networks (SNN) models and event-driven neuromorphic hardware. In this work, that contributes to space surveillance applications, we present a feasibility study on applying a Spike- Driven Transformer to satellite streak segmentation. We choose a transformer deliberately: a large backbone pretrained on broad and varied imagery is more likely to respond sensibly to the unfamiliar sensors and observing conditions a surveillance system meets in practice, but models of this size are not usually considered for edge deployment. We therefore ask whether spikedriven execution can carry that capacity at an edge power budget. We fine-tune a Spike-Driven Transformer backbone to segment satellite streaks and we estimate its energy consumption on the BrainChip Akida neuromorphic chip after evaluating the model in a label-free cross-sensor setting. The SNN-based model achieves competitive performance relative to more compact UNet baselines and energy consumption on a neuromorphic chip is projected, rather than hardware-measured, at 33.0mJ per 512 × 512 image inference.

Are you the author and do you want to add some related links to this page? If so, check out our guide to adding related links and videos.

Are you the author and do you want to add your embedded video to this page? If so, check out our guide to adding related links and videos.