Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Gold Exploration using Representations from a Multispectral Autoencoder

Argyro Tsandalidou, Konstantinos Dogeas, Eleftheria Tetoula Tsonga, Elisavet Parselia, Georgios Tsimiklis, George Arvanitakis

TL;DR

The paper addresses gold prospectivity mapping from satellite imagery under limited labeled data by leveraging representations learned from unlabeled multispectral Sentinel-2 data. It introduces Isometric, a frozen Masked Autoencoder-based foundation model trained on the large FalconSpace-S2v1.0 dataset to produce spectral-spatial embeddings, which feed a lightweight XGBoost classifier for gold vs non-gold detection. Compared to using raw multispectral inputs and to the state-of-the-art SpectralGPT, the representation-based approach yields notable gains in both patch-level and image-level accuracy, evidencing robust transfer of mineralogical patterns. The findings demonstrate the potential of foundation-model representations to enable scalable, globally applicable mineral exploration from space, with the encoder acting as a reusable feature extractor for limited-label tasks; future work envisions larger datasets and data fusion with SAR and hyperspectral modalities plus multi-temporal analysis.

Abstract

Satellite imagery is employed for large-scale prospectivity mapping due to the high cost and typically limited availability of on-site mineral exploration data. In this work, we present a proof-of-concept framework that leverages generative representations learned from multispectral Sentinel-2 imagery to identify gold-bearing regions from space. An autoencoder foundation model, called Isometric, which is pretrained on the large-scale FalconSpace-S2 v1.0 dataset, produces information-dense spectral-spatial representations that serve as inputs to a lightweight XGBoost classifier. We compare this representation-based approach with a raw spectral input baseline using a dataset of 63 Sentinel-2 images from known gold and non-gold locations. The proposed method improves patch-level accuracy from 0.51 to 0.68 and image-level accuracy from 0.55 to 0.73, demonstrating that generative embeddings capture transferable mineralogical patterns even with limited labeled data. These results highlight the potential of foundation-model representations to make mineral exploration more efficient, scalable, and globally applicable.

Gold Exploration using Representations from a Multispectral Autoencoder

TL;DR

The paper addresses gold prospectivity mapping from satellite imagery under limited labeled data by leveraging representations learned from unlabeled multispectral Sentinel-2 data. It introduces Isometric, a frozen Masked Autoencoder-based foundation model trained on the large FalconSpace-S2v1.0 dataset to produce spectral-spatial embeddings, which feed a lightweight XGBoost classifier for gold vs non-gold detection. Compared to using raw multispectral inputs and to the state-of-the-art SpectralGPT, the representation-based approach yields notable gains in both patch-level and image-level accuracy, evidencing robust transfer of mineralogical patterns. The findings demonstrate the potential of foundation-model representations to enable scalable, globally applicable mineral exploration from space, with the encoder acting as a reusable feature extractor for limited-label tasks; future work envisions larger datasets and data fusion with SAR and hyperspectral modalities plus multi-temporal analysis.

Abstract

Satellite imagery is employed for large-scale prospectivity mapping due to the high cost and typically limited availability of on-site mineral exploration data. In this work, we present a proof-of-concept framework that leverages generative representations learned from multispectral Sentinel-2 imagery to identify gold-bearing regions from space. An autoencoder foundation model, called Isometric, which is pretrained on the large-scale FalconSpace-S2 v1.0 dataset, produces information-dense spectral-spatial representations that serve as inputs to a lightweight XGBoost classifier. We compare this representation-based approach with a raw spectral input baseline using a dataset of 63 Sentinel-2 images from known gold and non-gold locations. The proposed method improves patch-level accuracy from 0.51 to 0.68 and image-level accuracy from 0.55 to 0.73, demonstrating that generative embeddings capture transferable mineralogical patterns even with limited labeled data. These results highlight the potential of foundation-model representations to make mineral exploration more efficient, scalable, and globally applicable.
Paper Structure (8 sections, 2 figures, 3 tables)

This paper contains 8 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Our contribution
  3. Model
  4. Generative AI Model
  5. Reconstruction Performance:
  6. Mineral Exploration Model
  7. Results
  8. Conclusion

Figures (2)

  • Figure 1: ROC curves illustrating the classification performance of the Raw, Spectral, and Isometric approaches.
  • Figure 2: Gold vs Non-Gold. The three images of subfigure \ref{['fig:results_mineral_left']} do not contain gold, while the three images of subfigure \ref{['fig:results_mineral_right']} do. Using the representations of the Isometric model (our approach), XGBoost classifies correctly all six images, while using the raw input, XGBoost mislabel two out of three samples that contain gold (only the middle image of subfigure \ref{['fig:results_mineral_right']} is correctly predicted using raw data).