Towards Onboard Continuous Change Detection for Floods

TL;DR

This paper introduces HiT, a memory-efficient Transformer-based onboard flood detection system for small satellites, enabling real-time hazard monitoring with minimal data storage and high processing speed.

Contribution

The paper presents the HiT mechanism that maintains historical context in Transformer models while drastically reducing data storage, suitable for onboard satellite flood detection.

Findings

Maintains detection accuracy comparable to bi-temporal baseline.

Achieves 43 FPS on Jetson Orin Nano hardware.

Reduces data storage by over 99% of original image size.

Abstract

Natural disaster monitoring through continuous satellite observation requires processing multi-temporal data under strict operational constraints. This paper addresses flood detection, a critical application for hazard management, by developing an onboard change detection system that operates within the memory and computational limits of small satellites. We propose History Injection mechanism for Transformer models (HiT), that maintains historical context from previous observations while reducing data storage by over 99\% of original image size. Moreover, testing on the STTORM-CD flood dataset confirms that the HiT mechanism within the Prithvi-tiny foundation model maintains detection accuracy compared to the bi-temporal baseline. The proposed HiT-Prithvi model achieved 43 FPS on Jetson Orin Nano, a representative onboard hardware used in nanosats. This work establishes a practical framework for satellite-based continuous monitoring of natural disasters, supporting real-time hazard assessment without dependency on ground-based processing infrastructure. Architecture as well as model checkpoints is available at https://github.com/zaitra/HiT-change-detection .