HarmoniDiff-RS: Training-Free Diffusion Harmonization for Satellite Image Composition

2026-04-21

AI Summary

This work addresses the visual inconsistency in synthetic remote sensing imagery caused by radiometric mismatch between source images and target scenes. The authors propose a training-free diffusion-based framework that leverages a pre-trained diffusion model, introducing a latent mean-shift alignment to harmonize the radiometric distributions of source and target domains. A timestep-aware latent fusion strategy is further designed to preserve semantic content while achieving high-quality image harmonization. The study makes three key contributions: it is the first to apply a training-free diffusion mechanism to remote sensing image harmonization, introduces two novel technical components for effective radiometric alignment, and constructs RSIC-H—the first benchmark dataset dedicated to remote sensing image coordination. Experimental results demonstrate that the generated images exhibit superior visual coherence and semantic fidelity, effectively supporting applications such as remote sensing data augmentation and disaster simulation.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Semi-Supervised Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Satellite image composition plays a critical role in remote sensing applications such as data augmentation, disaste simulation, and urban planning. We propose HarmoniDiff-RS, a training-free diffusion-based framework for harmonizing composite satellite images under diverse domain conditions. Our method aligns the source and target domains through a Latent Mean Shift operation that transfers radiometric characteristics between them. To balance harmonization and content preservation, we introduce a Timestep-wise Latent Fusion strategy by leveraging early inverted latents for high harmonization and late latents for semantic consistency to generate a set of composite candidates. A lightweight harmony classifier is trained to further automatically select the most coherent result among them. We also construct RSIC-H, a benchmark dataset for satellite image harmonization derived from fMoW, providing 500 paired composition samples. Experiments demonstrate that our method effectively performs satellite image composition, showing strong potential for scalable remote-sensing synthesis and simulation tasks. Code is available at: https://github.com/XiaoqiZhuang/HarmoniDiff-RS.

Problem

• satellite image composition
• image harmonization
• domain alignment
• remote sensing
• diffusion models

Innovation

• training-free diffusion
• latent mean shift
• timestep-wise latent fusion
• satellite image harmonization
• remote sensing composition