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

TL;DR

TerraMAE introduces an adaptive masked autoencoder framework tailored for hyperspectral satellite imagery, effectively capturing spatial-spectral features and improving performance in geospatial analysis tasks.

Contribution

The paper presents TerraMAE, a novel hyperspectral autoencoder with adaptive channel grouping and enhanced loss, advancing self-supervised learning for complex spectral-spatial data.

Findings

Superior reconstruction of hyperspectral images.

Enhanced performance on crop, land cover, and soil tasks.

Effective spectral similarity capture through adaptive grouping.

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.