Answers are grounded in this paper and cite their sources.
In the field of remote sensing, rotating scan optical satellites are an important innovation, offering the unique advantage of capturing extensive ground coverage areas with high precision. However, unlike traditional remote sensing images, the inherent characteristics of rotating scan remote sensing images make fast object detection and localization a more challenging task because of: 1) the ultra-wide swath results in extremely high image resolution; 2) changes in the rotation angle cause variations in resolution; 3) the large variation in object size; and 4) side-looking imaging introduces greater geometric distortions. A common solution is to divide the large remote sensing image into smaller (uniform) patches and then apply object detection to each small patch. In this paper, we investigate image cropping strategies to address these challenges. Specifically, we propose an adaptive image cropping method (AICM) for rotating scan remote sensing images. Starting from the resolution changes caused by variations in the rotation angle, we establish a ground sampling distance model that varies with the rotation angle, achieving adaptive image cropping under different ground sampling distances. Verification on ultra-wide swath rotating scan remote sensing images shows that, compared with the traditional fixed stride cropping method, AICM improves the average inference speed by 7.39% and increases the mAP by 7.54%.
No takes yet. Share an insight, caveat, or question.
Deng et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: