A research team from Wuhan University has introduced a novel artificial intelligence framework designed to restore partially hidden objects in satellite imagery, addressing a critical limitation in geospatial analysis. The method, detailed in the Journal of Remote Sensing, moves beyond conventional image inpainting by inferring complete object shape, surface texture, and semantic identity from fragmented visual data.
Satellite imagery is essential for disaster response, urban planning, environmental monitoring, and security analysis. However, objects in remote sensing images are frequently obscured by cloud cover, overlapping features, or limited viewing angles. These incomplete views can lead to misclassification, missed detections, and inaccurate mapping. Existing inpainting methods often produce visually plausible results but may distort object structure or generate contextually incorrect content.
The proposed framework, called Remote Sensing Amodal Completion (RSAC), adapts Stable Diffusion to the remote sensing domain using Low-Rank Adaptation (LoRA) and a four-channel ControlNet that guides structural completion. A prior-enhanced initialization strategy preserves low-frequency information from visible object parts, improving physical consistency. Compared to methods like LaMa, BrushNet, and OWAAC, the new approach achieved superior object geometry, clearer boundaries, and realistic texture continuity.
The team built a dedicated dataset of 1,770 annotated instances across ten categories, including planes, ships, large vehicles, and sports fields. In tests, the framework achieved 100% valid-output coverage, an IoU of 0.853, an ACIoU of 0.688, and an SSIM of 0.930, outperforming baseline methods that showed distorted geometry or weak foreground separation. The framework also improved downstream object detection and supported layered 2.5D scene understanding.
According to the researchers, the goal is to help machines infer what an object is and how it should be structured, not just make images look complete. This technology could enhance geospatial intelligence in post-disaster assessment, infrastructure mapping, automated cartography, and urban monitoring. Future work may extend the framework to more object categories, dynamic drone perspectives, and multimodal remote sensing data.
For further details, the study is available at https://doi.org/10.34133/remotesensing.1035. The research was supported by the National Natural Science Foundation of China under grants 42422109 and 42371366.
