TerraMAE: Learning Spatial-Spectral Representations from Hyperspectral Earth Observation Data via Adaptive Masked Autoencoders

2025-08-09

AI Summary

Modeling spatial-spectral correlations in hyperspectral remote sensing imagery (HSI) with >200 spectral bands remains challenging. To address this, we propose a self-supervised representation learning framework based on a masked autoencoder. Our method introduces two key innovations: (1) an adaptive channel grouping mechanism grounded in the statistical properties of surface reflectance, explicitly capturing spectral similarity; and (2) a multi-dimensional reconstruction loss that jointly enforces spatial structural fidelity and spectral consistency. This design significantly improves high-fidelity HSI reconstruction and enhances joint spatial-spectral feature representation. Extensive experiments demonstrate state-of-the-art performance on three downstream tasks—crop identification, land-cover classification, and soil texture prediction—validating both the effectiveness and generalizability of the learned representations.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Representation Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Hyperspectral satellite imagery offers sub-30 m views of Earth in hundreds of contiguous spectral bands, enabling fine-grained mapping of soils, crops, and land cover. While self-supervised Masked Autoencoders excel on RGB and low-band multispectral data, they struggle to exploit the intricate spatial-spectral correlations in 200+ band hyperspectral images. We introduce TerraMAE, a novel HSI encoding framework specifically designed to learn highly representative spatial-spectral embeddings for diverse geospatial analyses. TerraMAE features an adaptive channel grouping strategy, based on statistical reflectance properties to capture spectral similarities, and an enhanced reconstruction loss function that incorporates spatial and spectral quality metrics. We demonstrate TerraMAE's effectiveness through superior spatial-spectral information preservation in high-fidelity image reconstruction. Furthermore, we validate its practical utility and the quality of its learned representations through strong performance on three key downstream geospatial tasks: crop identification, land cover classification, and soil texture prediction.

Problem

• Exploiting spatial-spectral correlations in hyperspectral images
• Learning representative embeddings for geospatial analyses
• Improving performance on crop, land cover, and soil tasks

Innovation

• Adaptive channel grouping for spectral similarities
• Enhanced reconstruction loss with quality metrics
• Spatial-spectral embedding for geospatial analyses