Fig. 1. Processing flow of UAV group infrared image simulation method based on image derivation

Fig. 2. Overall processing flow of infrared UAV image blending method based on ED-GAN

Fig. 3. Template images and magnified image of infrared UAV

Fig. 4. Structure diagram of ED-GAN

Fig. 5. Image from image degradation-recovery dataset

Fig. 6. Multi-task pre-training process supervised by ED-GAN

Fig. 7. Unsupervised model migration in ED-GAN

Fig. 8. Sketch of shape and infrared image of small UAV. (a) UAV profile; (b) infrared images

Fig. 9. Schematic comparison of output images of evolution generator with original images

Fig. 10. Schematic comparison of output images of degraded generator with original images

Fig. 11. Qualitative experimental results of proposed method under different complex backgrounds and their partial enlarged images. (a) Mountain background; (b) woodland background; (c) sky and cloud background; (d) building group background

Fig. 12. Qualitative results of proposed method and other mixed image methods. (a) Copy and paste images; (b) PB method; (c) MPB method; (d) PR-GAN method; (e) GP-GAN method; (f) ED_org method; (g) ED_sup method; (h) ED_usup method

Fig. 13. P-R curves of each detector model in different datasets. (a) P-R curves of H_B in D_S when IoU threshold is 0.50; (b) P-R curves of H_B in D_S when IoU threshold is 0.75; (c) P-R curves of H_O in D_B when IoU threshold is 0.50; (d) P-R curves of H_O in D_B when IoU threshold is 0.75