Transfer Learning for Onboard Cloud Segmentation in Thermal Earth Observation: From Landsat to a CubeSat Constellation

2025-10-31

AI Summary

To address the challenges of limited onboard hardware resources, single-band thermal infrared (TIR) imagery, and scarce labeled data in CubeSat-based on-board cloud segmentation, this paper proposes a transfer learning framework. We pretrain a lightweight MobileNet-UNet architecture on publicly available Landsat TIR imagery, then jointly fine-tune it using a small set of onboard CubeSat data. The model is further optimized via TensorRT acceleration, enabling full-resolution inference within ≤5 seconds on a Jetson Nano edge device. Evaluated on the FOREST-2 CubeSat dataset, our method achieves a macro-F1 score of 0.877—significantly outperforming purely onboard-trained baselines. Our key contributions are: (i) the first application of cross-platform, multi-source TIR remote sensing data for transfer learning in CubeSat cloud segmentation; and (ii) empirical validation of real-time TIR cloud detection feasibility on low-power edge hardware.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Learning on the Edge & Model Compression
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Cloud, edge and content delivery systems for the Web
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Onboard cloud segmentation is a critical yet underexplored task in thermal Earth observation (EO), particularly for CubeSat missions constrained by limited hardware and spectral information. CubeSats often rely on a single thermal band and lack sufficient labeled data, making conventional cloud masking techniques infeasible. This work addresses these challenges by applying transfer learning to thermal cloud segmentation for the FOREST-2 CubeSat, using a UNet with a lightweight MobileNet encoder. We pretrain the model on the public Landsat-7 Cloud Cover Assessment Dataset and fine-tune it with a small set of mission-specific samples in a joint-training setup, improving the macro F1 from 0.850 to 0.877 over FOREST-2-only baselines. We convert the model to a TensorRT engine and demonstrate full-image inference in under 5 seconds on an NVIDIA Jetson Nano. These results show that leveraging public datasets and lightweight architectures can enable accurate, efficient thermal-only cloud masking on-orbit, supporting real-time decision-making in data-limited EO missions.

Problem

• Transfer learning enables thermal cloud segmentation for CubeSats
• Overcoming limited labeled data and hardware constraints in Earth observation
• Achieving efficient onboard inference for real-time decision-making

Innovation

• Transfer learning from Landsat to CubeSat cloud segmentation
• UNet with lightweight MobileNet encoder for thermal data
• TensorRT conversion enables under 5-second onboard inference