EarthSynth: Generating Informative Earth Observation with Diffusion Models

2025-05-17

AI Summary

Remote sensing image interpretation is severely constrained by the scarcity of annotated data. To address this, we propose EarthSynth—the first diffusion-based foundation model for multi-task remote sensing generation, enabling cross-satellite, multi-class, label-conditioned image synthesis. Methodologically: (1) we establish a novel multi-task generative paradigm tailored to remote sensing; (2) we introduce counterfactual compositional training to enhance class controllability and synthetic data diversity; and (3) we design the R-Filter—a rule-based sample selection mechanism—to retain high-information synthetic images. EarthSynth is pre-trained on EarthSynth-180K, a large-scale, diverse remote sensing dataset. Extensive evaluation demonstrates substantial improvements in few-shot and zero-shot performance across scene classification, object detection, and semantic segmentation. By providing scalable, label-aware synthetic data, EarthSynth establishes a generative data infrastructure for open-world remote sensing interpretation.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Remote sensing image (RSI) interpretation typically faces challenges due to the scarcity of labeled data, which limits the performance of RSI interpretation tasks. To tackle this challenge, we propose EarthSynth, a diffusion-based generative foundation model that enables synthesizing multi-category, cross-satellite labeled Earth observation for downstream RSI interpretation tasks. To the best of our knowledge, EarthSynth is the first to explore multi-task generation for remote sensing. EarthSynth, trained on the EarthSynth-180K dataset, employs the Counterfactual Composition training strategy to improve training data diversity and enhance category control. Furthermore, a rule-based method of R-Filter is proposed to filter more informative synthetic data for downstream tasks. We evaluate our EarthSynth on scene classification, object detection, and semantic segmentation in open-world scenarios, offering a practical solution for advancing RSI interpretation.

Problem

• Addresses scarcity of labeled remote sensing images
• Proposes diffusion model for multi-category Earth observation synthesis
• Enhances downstream tasks via synthetic data filtering

Innovation

• Diffusion-based generative model for Earth observation
• Counterfactual Composition training for diversity
• Rule-based R-Filter for informative synthetic data