Remote Sensing Small Object Detection

The field studies how to locate small targets—such as vehicles, ships, and wildlife—in aerial and satellite images where objects may occupy only a few pixels and appear in dense or cluttered scenes. Common approaches use feature pyramids, attention, and context modeling to combine fine spatial detail with information from larger regions; benchmarks include VisDrone, DOTA, AI-TOD, and xView. Other methods process high-resolution images with overlapping slices or selectively allocated tokens and queries, while SAR, infrared, and visible-infrared settings require handling noise or combining different sensor cues. A central challenge is improving localization and transfer across sensors, resolutions, and domains without making inference too costly or losing the weak visual details of tiny objects.