Researchers from the School of Resource and Environmental Science at Wuhan University have developed a new artificial intelligence framework that can reconstruct partially hidden objects in satellite imagery with high accuracy. The method, detailed in the Journal of Remote Sensing, addresses a critical challenge in geospatial AI: how to infer complete objects when only fragments are visible due to cloud cover, overlapping objects, or imaging angles.
Satellite imagery is widely used in disaster response, urban planning, environmental monitoring, automated mapping, and security analysis. However, ground objects in remote sensing images are frequently obscured, causing recognition models to misclassify objects, detectors to miss full targets, and mapping workflows to generate fragmented geometry. Existing image inpainting methods often produce visually plausible results but may distort object structure or hallucinate incorrect content.
The study introduces Remote Sensing Amodal Completion (RSAC) as a dedicated task for reconstructing complete ground objects from partial observations. The proposed Dual-Adaptive Diffusion-Based Framework adapts Stable Diffusion to the remote sensing domain through Low-Rank Adaptation (LoRA), while a four-channel ControlNet uses image and mask information to guide structural completion. A prior-enhanced initialization strategy preserves low-frequency information from the visible object rather than starting from random noise, improving physical consistency.
In comparative experiments, the proposed method achieved 100% valid-output coverage, an Intersection over Union (IoU) of 0.853, an amodal completion IoU (ACIoU) of 0.688, a mean squared error (MSE) of 11.822, a peak signal-to-noise ratio (PSNR) of 24.799 dB, and a structural similarity index (SSIM) of 0.930. These results outperformed baseline methods including Stable Diffusion Inpainting, LaMa, BrushNet, and Open-World Amodal Appearance Completion (OWAAC), which showed problems such as distorted geometry, unrealistic backgrounds, and weak foreground separation.
The researchers built a dedicated RSAC dataset containing 1,770 annotated instances across 10 categories of typical remote sensing objects, including planes, ships, large vehicles, storage tanks, roundabouts, tennis courts, basketball courts, baseball diamonds, soccer ball fields, and ground track fields. The dataset included 1,235 training images and 535 testing images.
The framework also helped restore semantic identity for vision-language models (VLMs), improved downstream object detection, and supported layered 2.5D scene understanding. The team emphasized that the goal was not simply to make incomplete images look visually complete, but to help machines infer what an object is and how it should be structured.
This technology could support more reliable geospatial intelligence in scenarios where objects are frequently obscured, such as post-disaster assessment, infrastructure mapping, automated cartography, facility reconstruction, and urban monitoring. By restoring complete object morphology from partial observations, RSAC may also improve training data for detection models and help AI systems interpret satellite imagery more like human analysts.
The research was supported by the National Natural Science Foundation of China under grant numbers 42422109 and 42371366. Future studies may extend the framework to more object categories, dynamic drone perspectives, full three-dimensional reconstruction, and multitemporal or multimodal remote sensing data.
