Plan, Decouple, Assimilate: Physics-Aware Object Insertion in Remote Sensing Imagery

Object insertion has emerged as a promising augmentation paradigm for the label scarcity and long-tailed distributions in remote sensing, generating training samples by synthesizing target instances onto real backgrounds. However, existing methods suffer from three critical issues: (i) semantic placement inconsistency, (ii) radiometric inconsistency with illumination and atmospheric conditions, and (iii) textural discontinuity. To address these, we propose a physics-aware method, "Plan, Decouple, Assimilate" (PDA), for generating high-fidelity training samples. In the planning stage, the Planning (P) module automatically generates geometrically valid bounding boxes. In the generation stage, a dual-module design synthesizes the target instance: the Decoupling (D) module employs Asymmetric Spectral Adaptation to disentangle structural identity from environmental illumination, while the Assimilation (A) module uses Neighborhood-Aware Texture Assimilation to harmonize the local manifold. By integrating these modules, PDA enforces multi-level consistency from global geometry to local micro-textures. Extensive experiments verify that PDA outperforms state-of-the-art methods in generative quality, reducing whole-image FID by 15.7% over the strongest baseline, and substantially improves downstream detection, boosting average mAP50 by +17.07 points over the real data.

Satellite and aerial images help us monitor everything from ship traffic and airport activity to critical infrastructure. However, training AI to recognize rare objects, such as uncommon aircraft, is difficult because real examples are scarce, often restricted, and expensive to collect. A common workaround is to paste images of target objects onto real backgrounds to create synthetic training data. Unfortunately, the results often look fake. Objects appear in impossible places like aircraft in the middle of forests, their shadows point the wrong way relative to the sun, and they look "stuck on" rather than naturally photographed. We developed PDA, a method that respects the physics of how satellites actually capture images. It first plans where an object can realistically go based on the scene's geography. It then adjusts the object's lighting and shadows to match the sun's position in the original photo. Finally, it blends the object's texture so it looks like part of the same image rather than a sticker pasted on top. The resulting synthetic images look substantially more realistic and dramatically improve AI detection accuracy, boosting performance by 17 percentage points over training with real data alone. This helps build reliable detection systems even when real training examples are scarce.