Geo2Vec: Shape- and Distance-Aware Neural Representation of Geospatial Entities

2025-08-26

AI Summary

Existing spatial representation learning methods suffer from three key limitations: (1) reliance on single-type modeling, (2) requirement of entity decomposition to enable Fourier transforms—introducing substantial computational overhead—and (3) dependence on uniform sampling due to geometric misalignment, which degrades fine-grained features such as boundaries. To address these, Geo2Vec introduces a neural representation framework grounded in Signed Distance Fields (SDF), the first to apply SDFs to geographic entity modeling. It enables adaptive sampling and encoding of heterogeneous geospatial primitives—points, polylines, and polygons—directly in their native coordinate space, preserving geometric shape and spatial relationships. A novel rotation-invariant positional encoding is proposed to enhance high-frequency geometric detail modeling and embedding robustness. Experiments on real-world geographic AI tasks demonstrate that Geo2Vec significantly outperforms state-of-the-art baselines in accuracy, efficiency, and joint characterization of shape, position, topology, and metric distance relationships.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Representation Learning

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
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Spatial representation learning is essential for GeoAI applications such as urban analytics, enabling the encoding of shapes, locations, and spatial relationships (topological and distance-based) of geo-entities like points, polylines, and polygons. Existing methods either target a single geo-entity type or, like Poly2Vec, decompose entities into simpler components to enable Fourier transformation, introducing high computational cost. Moreover, since the transformed space lacks geometric alignment, these methods rely on uniform, non-adaptive sampling, which blurs fine-grained features like edges and boundaries. To address these limitations, we introduce Geo2Vec, a novel method inspired by signed distance fields (SDF) that operates directly in the original space. Geo2Vec adaptively samples points and encodes their signed distances (positive outside, negative inside), capturing geometry without decomposition. A neural network trained to approximate the SDF produces compact, geometry-aware, and unified representations for all geo-entity types. Additionally, we propose a rotation-invariant positional encoding to model high-frequency spatial variations and construct a structured and robust embedding space for downstream GeoAI models. Empirical results show that Geo2Vec consistently outperforms existing methods in representing shape and location, capturing topological and distance relationships, and achieving greater efficiency in real-world GeoAI applications. Code and Data can be found at: https://github.com/chuchen2017/GeoNeuralRepresentation.

Problem

• Representing diverse geospatial entities with unified neural embeddings
• Capturing geometric details without decomposition or high computational cost
• Modeling spatial relationships and variations robustly for GeoAI applications

Innovation

• Uses signed distance fields for geometry-aware encoding
• Adaptive sampling without decomposition for efficiency
• Rotation-invariant encoding for robust embedding space