 The proposed COSR algorithm is a one-stage anchor-free rotated object detector that can be easily deployed on most platforms and uses a well-defined label assignment strategy for aerial image objects. It achieves a map of 79.25, 75.41, and 90.13 on the Dota V1.0, Dota V1.5, and HRSC 2016 datasets, respectively. The small model achieves a map of 74.05 at a speed of 23.7 FPS on an RTX 2080T GPU, while the lightweight COSR model converted to tensor-RT format achieves a map of 73.93 on Dota V1.0 at a speed of 17.76 FPS on a Jetson AGX Xavier device with a single scale. This article was authored by Zhonghua Li, Biao Hou, Xidong Wu, and others.