Real-time super-resolution for infrared dynamic object video based on airborne platform

Deyan ZHU; Jiayi XU; Yongqi AO

doi:10.37188/OPE.20253305.0818

Optics and Precision Engineering, Volume. 33, Issue 5, 818(2025)

Real-time super-resolution for infrared dynamic object video based on airborne platform

Deyan ZHU^1,2、*, Jiayi XU^1,2, and Yongqi AO^1,2

Author Affiliations

¹College of Astronautics， Nanjing University of Aeronautics and Astronautics， Nanjing2006， China

²Key Laboratory of Space Photoelectric Detection and Sensing of Industry and Information Technology， Nanjing University of Aeronautics and Astronautics， Nanjing10016， China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(19)

Fig. 1. Architecture of recurrent residual neural network

Download full size

View in Article

Fig. 2. Interior view of residual block

Download full size

View in Article

Fig. 3. Equivalent scale scenes

Download full size

View in Article

Fig. 4. Loss curve of RNN-based model

Download full size

View in Article

Fig. 5. Super-resolution results of infrared small target under ground and air background dataset

Download full size

View in Article

Fig. 6. Super-resolution results of real infrared datasets

Download full size

View in Article

Fig. 7. Flowchart of angle resolution calibration

Download full size

View in Article

Fig. 8. Drawing board with point targets

Download full size

View in Article

Fig. 9. Super-resolution result of resolution test target image

Download full size

View in Article

Table 1. Details of infrared dynamic object video datasets

View table

View in Article

Table 1. Details of infrared dynamic object video datasets

Dataset	Sequence	Image size	Object	Frame number
Vimeo-90k	—	448×256	Multiple objects	641907
Infrared small target under ground and air background dataset	IR1	256×256	UAV	398
	IR2	256×256	UAV	598
	IR3	256×256	UAV	99
	IR4	256×256	UAV	398
Infrared UAV dataset	IR5	640×512	UAV and airplane	2000
Airplane model dataset	IR6	640×512	Airplane model	200
	IR7	640×512	Airplane model	200
	IR8	640×512	Airplane model	200

Table 2. Configuration of experimental environment
View table
View in Article
Table 2. Configuration of experimental environment
Name Information
CPU 13th Gen Intel（R） Core（TM） i7-13700F
GPU NVIDIA GeForce RTX 4060 Ti
OS Windows10
Framework Pytorch2.1.1+cuda12.1

Table 3. Details of models

View table

View in Article

Table 3. Details of models

Model	EDVR （M）	DUF （S）	BasicVSR	IART	ST-AVSR	Ours
Number of 2D/3D convolutional layers	57 （2D）	18 （3D）	125 （2D）	126（2D）	217（2D）	13 （2D）
Number of Channels in convolutional layers	64	64	64	64	64	32

Table 4. PSNR and SSIM of super-resolution results

View table

View in Article

Table 4. PSNR and SSIM of super-resolution results

Sequence	EDVR		DUF		BasicVSR		IART		ST-AVSR		Ours
Sequence	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
IR1	39.01	0.93	40.29	0.92	39.07	0.93	35.84	0.92	35.83	0.90	40.57	0.93
IR2	40.67	0.93	40.37	0.92	40.93	0.93	39.88	0.92	38.36	0.90	40.50	0.92
IR3	39.31	0.93	40.62	0.94	39.30	0.93	36.84	0.91	38.31	0.91	40.83	0.94
IR4	41.63	0.94	41.93	0.94	41.69	0.94	40.81	0.94	39.66	0.93	42.11	0.94

Table 5. Runtime of different methods on IR1-IR4 sequences
View table
View in Article
Table 5. Runtime of different methods on IR1-IR4 sequences
Method Runtime/s
EDVR 0.089
DUF 0.085
BasicVSR 0.024
IART 0.038
ST-AVSR 0.028
Ours 0.009

Table 6. PSNR and SSIM of super-resolution results of real infrared datasets

View table

View in Article

Table 6. PSNR and SSIM of super-resolution results of real infrared datasets

Sequence	EDVR		DUF		BasicVSR		IART		ST-AVSR		Ours
Sequence	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
IR5	43.14	0.98	52.25	0.99	43.22	0.98	36.74	0.95	27.77	0.75	49.35	0.99
IR6	47.58	0.99	50.16	0.99	47.68	0.99	44.41	0.98	43.90	0.97	49.04	0.98
IR7	44.79	0.99	54.33	0.99	45.02	0.99	38.30	0.98	33.27	0.91	50.42	0.99
IR8	52.42	0.99	56.26	0.99	52.48	0.99	47.41	0.99	46.33	0.99	56.35	0.99

Table 7. Runtime of different methods on IR5-IR8 sequences
View table
View in Article
Table 7. Runtime of different methods on IR5-IR8 sequences
Method Runtime/s
EDVR 0.371
DUF 0.417
BasicVSR 0.110
IART 0.181
ST-AVSR 0.135
Ours 0.022

Table 8. Results of magnification calibration

View table

View in Article

Table 8. Results of magnification calibration

Object distance/mm	A	B	Image distance/pixel	$β$
20	（261，493）	（389，494）	128	1/10.41
40	（196，416）	（452，416）	256	1/10.42
60	（131，343）	（513，344）	382	1/10.47
80	（64，271）	（579，272）	515	1/10.36

Table 9. Runtime of different models
View table
View in Article
Table 9. Runtime of different models
Model IR1~IR4 IR5~IR8
5L_32C 0.009 0.022
5L_64C 0.020 0.086
7L_32C 0.014 0.041

Table 10. PSNR and SSIM of super-resolution results of different models

View table

View in Article

Table 10. PSNR and SSIM of super-resolution results of different models

Model	PSNR/dB								SSIM
Model	IR1	IR2	IR3	IR4	IR5	IR6	IR7	IR8	IR1	IR2	IR3	IR4	IR5	IR6	IR7	IR8
5L_32C	40.57	40.50	40.83	42.11	49.35	49.04	50.42	56.35	0.93	0.92	0.94	0.94	0.99	0.98	0.99	0.99
5L_64C	40.80	40.63	41.13	42.25	46.77	48.03	45.95	56.26	0.94	0.92	0.95	0.94	0.99	0.98	0.99	0.99
7L_32C	40.75	40.54	40.98	42.18	49.75	49.16	50.72	56.38	0.94	0.92	0.95	0.94	0.99	0.99	0.99	0.99

Tools

Get Citation

Copy Citation Text

Deyan ZHU, Jiayi XU, Yongqi AO. Real-time super-resolution for infrared dynamic object video based on airborne platform[J]. Optics and Precision Engineering, 2025, 33(5): 818

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites