Poly2Vec: Polymorphic Fourier-Based Encoding of Geospatial Objects for GeoAI Applications

Encoding geospatial objects is fundamental for geospatial artificial intelligence (GeoAI) applications, which leverage machine learning (ML) models to analyze spatial information. Common approaches transform each object into known formats, like image and text, for compatibility with ML models. However, this process often discards crucial spatial information, such as the object's position relative to the entire space, reducing downstream task effectiveness. Alternative encoding methods that preserve some spatial properties are often devised for specific data objects (e.g., point encoders), making them unsuitable for tasks that involve different data types (i.e., points, polylines, and polygons). To this end, we propose Poly2Vec, a polymorphic Fourier-based encoding approach that unifies the representation of geospatial objects, while preserving the essential spatial properties. Poly2Vec incorporates a learned fusion module that adaptively integrates the magnitude and phase of the Fourier transform for different tasks and geometries.We evaluate Poly2Vec on five diverse tasks, organized into two categories. The first empirically demonstrates that Poly2Vec consistently outperforms object-specific baselines in preserving three key spatial relationships: topology, direction, and distance. The second shows that integrating Poly2Vec into a state-of-the-art GeoAI workflow improves the performance in two popular tasks: population prediction and land use inference.

Machine learning is increasingly used to understand and reason about the world around us, including geographic data like buildings, roads, and parks. But unlike text or images, geospatial data comes in many forms: a store might be a point, a street a line, and a park a polygon. Most machine learning models aren't designed to work with this kind of diverse spatial data directly.To make it work, a common approach is to convert map objects into other formats, like images. But this conversion can distort important spatial relationships, making models less accurate. Other methods try to encode spatial features more directly, but only for one object type at a time, such as points. That makes them hard to generalize across tasks involving mixed geometries.We introduce Poly2Vec, a new method that encodes all types of geospatial object (points, polylines, and polygons) into a unified format that machine learning models can use, without losing key spatial information. Poly2Vec uses the Fourier transform to capture important spatial properties like the shape and location of each object, and adapts to different tasks.Plugging Poly2Vec into existing workflows improves performance on real-world applications like population prediction and land use classification, demonstrating its versatility and effectiveness in GeoAI pipelines.