Hyperbolic geometry improves unknown object detection in satellite images
Hyperbolic Geometry for Open-World Object Detection in Remote Sensing Imagery
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Detecting new or unknown objects in satellite images is a challenge because objects often have hidden relationships that usual methods miss. The authors use hyperbolic geometry, a type of math that better represents these relationships, to help identify unknown objects more accurately. They built a system that separates objects from backgrounds and measures uncertainty to decide if an object is known or new. Their method also helps computers learn new object types over time without forgetting old ones. Tests on satellite image datasets showed their approach works better than existing methods.
What this means in practice
- •For satellite image analysts: Improve detection of previously unseen objects in satellite images for better monitoring and analysis.
- •For environmental monitoring teams: Support continuous learning of new object categories in environmental remote sensing data without losing prior knowledge.
Authors
Wuzhou Li, Jiawei Zhou, Shenghang Wang, Xiang Li
Abstract
Open-world object detection (OWOD) extends closed-set detection by requiring models to identify unknown objects and incrementally learn them once annotations become available. In remote sensing imagery, object categories often exhibit latent hierarchical relationships that may be inadequately represented in the Euclidean spaces commonly adopted by existing methods, limiting unknown-object recall and incremental-learning performance. To address this issue, we investigate hyperbolic geometry for OWOD in remote sensing imagery and propose HyRS-OWOD. To improve unknown object recall, we design a two-step unknown-object discovery mechanism: a Decoupled Objectness Learning (DOL) module that disentangles foreground perception from semantic information to separate foreground proposals from background regions, followed by a Hyperbolic Uncertainty Learning (HUL) component that leverages the radius of hyperbolic embeddings as an uncertainty-aware cue for known-unknown discrimination. For incremental learning, we develop a Hyperbolic Metric Learning (HML) strategy that enhances inter-class separability, facilitating the incorporation of novel categories while mitigating catastrophic forgetting. Experiments on three remote sensing benchmarks demonstrate consistent improvements in unknown recall and incremental learning over state-of-the-art OWOD methods.