Fig. 1. Architecture of YOLOv3 network

Fig. 2. Flow chart of expert system

Fig. 3. Part of simulated dataset, including 18 satellites, each of which has 5 attitudes

Fig. 4. Part of real dataset. (a) Clear; (b) blur of lower degree; (c) blur of higher degree

Fig. 5. Class and state datasets of space targets. (a) reconnaissance-1; (b) reconnaissance-2; (c) communication satellite; (d) KH-11

Fig. 6. Specific dividing method for training set

Fig. 7. Detection results of space target payloads in simulated images. (a)(e)(i) Clear images (T1); (b)(f)(j) blur images (T2) with noise of 9 dB; (c)(g)(k) blur images (T3) with noise of 9 dB; (d)(h)(l) blur images (T3) with noise of 7 dB

Fig. 8. Payload identification results of real data (ISS). (a) 1st frame; (b) 20th frame; (c) 40th frame

Fig. 9. Payload identification results of space targets from website images. (a)(b)(c) ISS; (d) Hubble telescope; (e) Sentinel; (f) unknown space target

Fig. 10. Payload identification results of space target for Gaussian blur. (a)(e) Original high-resolution clear images; (b)(f) clear images; (c)(g) Gaussian blur images from T2'; (d)(h) Gaussian blur images from T3'

Fig. 11. State identification results of simulated space targets. (a) Clear image from T1; (b) blur image from T2; (c) blur image from T3

Fig. 12. State identification results of real data (ISS)