SpectralEarth: Training Hyperspectral Foundation Models at Scale

Foundation models have triggered a paradigm shift in computer vision and are increasingly being adopted in remote sensing, particularly for multispectral imagery. Yet, their potential in hyperspectral imaging (HSI) remains untapped due to the absence of comprehensive and globally representative hype...

Full description

Saved in:
Bibliographic Details
Main Authors: Nassim Ait Ali Braham, Conrad M. Albrecht, Julien Mairal, Jocelyn Chanussot, Yi Wang, Xiao Xiang Zhu
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11045062/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849718136431968256
author Nassim Ait Ali Braham
Conrad M. Albrecht
Julien Mairal
Jocelyn Chanussot
Yi Wang
Xiao Xiang Zhu
author_facet Nassim Ait Ali Braham
Conrad M. Albrecht
Julien Mairal
Jocelyn Chanussot
Yi Wang
Xiao Xiang Zhu
author_sort Nassim Ait Ali Braham
collection DOAJ
description Foundation models have triggered a paradigm shift in computer vision and are increasingly being adopted in remote sensing, particularly for multispectral imagery. Yet, their potential in hyperspectral imaging (HSI) remains untapped due to the absence of comprehensive and globally representative hyperspectral datasets. To close this gap, we introduce <italic>SpectralEarth</italic>, a large-scale multitemporal dataset designed to pretrain hyperspectral foundation models leveraging data from the environmental mapping and analysis program (EnMAP). SpectralEarth comprises 538 974 image patches covering 415 153 unique locations from 11 636 globally distributed EnMAP scenes spanning two years of archive. In addition, 17.5&#x0025; of these locations include multiple timestamps, enabling multitemporal HSI analysis. Utilizing state-of-the-art self-supervised learning algorithms, we pretrain a series of foundation models on SpectralEarth, integrating a spectral adapter into classical vision backbones to accommodate the unique characteristics of HSI. In tandem, we construct nine downstream datasets for land-cover, crop-type mapping, and tree-species classification, providing benchmarks for model evaluation. Experimental results support the versatility of our models and their generalizability across different tasks and sensors. We also highlight computational efficiency during model fine-tuning.
format Article
id doaj-art-345bf1b4e3bf4ee683ebdbe53f93436f
institution DOAJ
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-345bf1b4e3bf4ee683ebdbe53f93436f2025-08-20T03:12:27ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118167801679710.1109/JSTARS.2025.358145111045062SpectralEarth: Training Hyperspectral Foundation Models at ScaleNassim Ait Ali Braham0https://orcid.org/0009-0001-3346-3373Conrad M. Albrecht1https://orcid.org/0009-0009-2422-7289Julien Mairal2https://orcid.org/0000-0001-6991-2110Jocelyn Chanussot3https://orcid.org/0000-0003-4817-2875Yi Wang4https://orcid.org/0000-0002-3096-6610Xiao Xiang Zhu5https://orcid.org/0000-0001-8107-9096Chair of Data Science in Earth Observation, Technical University of Munich, Munich, GermanyRemote Sensing Technology Institute, German Aerospace Center, Wessling, GermanyUniversity Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, Grenoble, FranceUniversity Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, Grenoble, FranceChair of Data Science in Earth Observation, Technical University of Munich, Munich, GermanyChair of Data Science in Earth Observation, Technical University of Munich, Munich, GermanyFoundation models have triggered a paradigm shift in computer vision and are increasingly being adopted in remote sensing, particularly for multispectral imagery. Yet, their potential in hyperspectral imaging (HSI) remains untapped due to the absence of comprehensive and globally representative hyperspectral datasets. To close this gap, we introduce <italic>SpectralEarth</italic>, a large-scale multitemporal dataset designed to pretrain hyperspectral foundation models leveraging data from the environmental mapping and analysis program (EnMAP). SpectralEarth comprises 538 974 image patches covering 415 153 unique locations from 11 636 globally distributed EnMAP scenes spanning two years of archive. In addition, 17.5&#x0025; of these locations include multiple timestamps, enabling multitemporal HSI analysis. Utilizing state-of-the-art self-supervised learning algorithms, we pretrain a series of foundation models on SpectralEarth, integrating a spectral adapter into classical vision backbones to accommodate the unique characteristics of HSI. In tandem, we construct nine downstream datasets for land-cover, crop-type mapping, and tree-species classification, providing benchmarks for model evaluation. Experimental results support the versatility of our models and their generalizability across different tasks and sensors. We also highlight computational efficiency during model fine-tuning.https://ieeexplore.ieee.org/document/11045062/Hyperspectral imagingSelf-supervised learningFoundation models
spellingShingle Nassim Ait Ali Braham
Conrad M. Albrecht
Julien Mairal
Jocelyn Chanussot
Yi Wang
Xiao Xiang Zhu
SpectralEarth: Training Hyperspectral Foundation Models at Scale
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Hyperspectral imaging
Self-supervised learning
Foundation models
title SpectralEarth: Training Hyperspectral Foundation Models at Scale
title_full SpectralEarth: Training Hyperspectral Foundation Models at Scale
title_fullStr SpectralEarth: Training Hyperspectral Foundation Models at Scale
title_full_unstemmed SpectralEarth: Training Hyperspectral Foundation Models at Scale
title_short SpectralEarth: Training Hyperspectral Foundation Models at Scale
title_sort spectralearth training hyperspectral foundation models at scale
topic Hyperspectral imaging
Self-supervised learning
Foundation models
url https://ieeexplore.ieee.org/document/11045062/
work_keys_str_mv AT nassimaitalibraham spectralearthtraininghyperspectralfoundationmodelsatscale
AT conradmalbrecht spectralearthtraininghyperspectralfoundationmodelsatscale
AT julienmairal spectralearthtraininghyperspectralfoundationmodelsatscale
AT jocelynchanussot spectralearthtraininghyperspectralfoundationmodelsatscale
AT yiwang spectralearthtraininghyperspectralfoundationmodelsatscale
AT xiaoxiangzhu spectralearthtraininghyperspectralfoundationmodelsatscale