Table 1. Experimental comparison results of average accuracy of different detection methods on the ground. The table lists the test results of two public data sets RSOD and NWPU VHR-10. Judging from the single type detection results, our model performed well. From the average accuracy it seems that our model is better than other models.

RSOD	AP1	AP2	AP3	AP4							mAP
YOLOv5n	97.8	98.2	83.3	47.1							81.6
YOLOv5s	98.2	98.6	82.3	44.8							83.5
YOLOv3-Tiny	96.3	96.3	76.8	65.9							83.9
AS-YOLOv5n	98.2	97.5	86.8	57.2							84.9

The types in the ROSD data set include: AP1 (aircraft), AP2 (oiltank), AP3 (overpass), AP4 (playground).
The types in the NWPU VHR-10 data set include: AP1 (airplane), AP2 (ship), AP3 (storage tank), AP4(baseball diamond), AP5 (tennis court), AP6 (basketball court), AP7 (ground track field), AP8 (harbor), AP9 (bridge), AP10 (vehicle).