Saved in:
Bibliographic Details
Main Author: Chuc, Man Duc
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.20174
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916811055824896
author Chuc, Man Duc
author_facet Chuc, Man Duc
contents Foundation models are rapidly transforming Earth Observation data mining by enabling generalizable and scalable solutions for key tasks such as scene classification and semantic segmentation. While most efforts in the geospatial domain have focused on developing large models trained from scratch using massive Earth Observation datasets, an alternative strategy that remains underexplored is the reuse and combination of existing pretrained models. In this study, we investigate whether foundation models pretrained on remote sensing and general vision datasets can be effectively combined to improve performance across a diverse set of key Earth Observation tasks. Using the GEO-Bench benchmark, we evaluate several prominent models, including Prithvi, Hiera, and DOFA, on eleven datasets covering a range of spatial resolutions, sensor modalities, and task types. The results show that feature-level ensembling of smaller pretrained models can match or exceed the performance of much larger models, while requiring less training time and computational resources. Moreover, the study highlights the potential of applying knowledge distillation to transfer the strengths of ensembles into more compact models, offering a practical path for deploying foundation models in real-world Earth Observation applications.
format Preprint
id arxiv_https___arxiv_org_abs_2506_20174
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Scalable and Generalizable Earth Observation Data Mining via Foundation Model Composition
Chuc, Man Duc
Computer Vision and Pattern Recognition
Foundation models are rapidly transforming Earth Observation data mining by enabling generalizable and scalable solutions for key tasks such as scene classification and semantic segmentation. While most efforts in the geospatial domain have focused on developing large models trained from scratch using massive Earth Observation datasets, an alternative strategy that remains underexplored is the reuse and combination of existing pretrained models. In this study, we investigate whether foundation models pretrained on remote sensing and general vision datasets can be effectively combined to improve performance across a diverse set of key Earth Observation tasks. Using the GEO-Bench benchmark, we evaluate several prominent models, including Prithvi, Hiera, and DOFA, on eleven datasets covering a range of spatial resolutions, sensor modalities, and task types. The results show that feature-level ensembling of smaller pretrained models can match or exceed the performance of much larger models, while requiring less training time and computational resources. Moreover, the study highlights the potential of applying knowledge distillation to transfer the strengths of ensembles into more compact models, offering a practical path for deploying foundation models in real-world Earth Observation applications.
title Towards Scalable and Generalizable Earth Observation Data Mining via Foundation Model Composition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.20174