Many Pixels, One Composition: Self-Supervised Spectra for Specimen-Level Prediction

概要

Can hyperspectral (HS) imaging be used to predict the elemental composition of physical specimens found on the Earth's surface? We approach this problem as a supervised regression task, using an open dataset of specimens, each comprising a HS cube and elemental characterization. In particular, we train an encoder in a self-supervised fashion that operates on pixel-level; then freeze the encoder and train a lightweight attention-based regression head in a supervised manner, by pairing many representations of a given specimen with its corresponding elemental label. In test specimens, our approach significantly outperforms traditional spectral matching and unmixing baselines. It also outperforms supervised learning models based on the mean spectrum (of similar model capacity), while full model fine-tuning achieves the strongest performance, at substantially higher model capacity.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
Many Pixels, One Composition: Self-Supervised Spectra for Specimen-Level Prediction

# 概要
Can hyperspectral (HS) imaging be used to predict the elemental composition of physical specimens found on the Earth's surface? We approach this problem as a supervised regression task, using an open…

# 原典
https://arxiv.org/abs/2611.06096

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。