Geospatial Foundational Embedder: Top-1 Winning Solution on EarthVision Embed2Scale Challenge (CVPR 2025)

2025-09-03

AI Summary

To address the limited cross-resolution representation capability of remote sensing data, this paper proposes a geospatial foundation model that generates universal vector embeddings from the SSL4EO-S12 hyperspectral data cube. Methodologically, we design a multi-scale feature fusion architecture integrating two complementary self-supervised pretraining strategies—contrastive learning and masked reconstruction—and incorporate a Transformer encoder with hierarchical pooling to achieve robust modeling of heterogeneous, multi-resolution remote sensing inputs. Evaluated on the EarthVision Embed2Scale Challenge, our model ranks first, substantially outperforming existing approaches. It demonstrates strong generalization and transferability across diverse downstream tasks, including classification and regression. This work establishes a scalable, extensible paradigm for geospatial foundation models, advancing the representation learning of Earth observation data across spatial and spectral scales.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Deep Neural Architectures and Foundation Models
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

EarthVision Embed2Scale challenge (CVPR 2025) aims to develop foundational geospatial models to embed SSL4EO-S12 hyperspectral geospatial data cubes into embedding vectors that faciliatetes various downstream tasks, e.g., classification, regression, etc. In this technical report, we introduce our proposed method for the Top-1 winning solution on the Embed2Scale Challenge.

Problem

• Develops foundational geospatial models for hyperspectral data
• Embeds SSL4EO-S12 data cubes into vector representations
• Facilitates downstream tasks like classification and regression

Innovation

• Develops foundational geospatial embedding models
• Processes hyperspectral data cubes into vectors
• Enables classification and regression downstream tasks