A Dual Branch Edge Convolution Fusion Network for Infrared and Visible Images

Hongde ZHANG; Xin FENG; Jieming YANG; Guohang QIU

doi:10.3788/gzxb20245308.0810004

Acta Photonica Sinica, Volume. 53, Issue 8, 0810004(2024)

A Dual Branch Edge Convolution Fusion Network for Infrared and Visible Images

Hongde ZHANG... Xin FENG*, Jieming YANG and Guohang QIU |Show fewer author(s)

School of Mechanical Engineering， Chongqing Key Laboratory of Green Design and Manufacturing of Intelligent Equipment， Chongqing Technology and Business University， Chongqing 400067， China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Infrared and visible image fusion based on dual-branch convolution network

Download full size

View in Article

Fig. 2. Edge convolution block

Download full size

View in Article

Fig. 3. Convolutional block attention module

Download full size

View in Article

Fig. 4. Results of infrared and visible image fusion with different modules

Download full size

View in Article

Fig. 5. Results of infrared and visible image fusion with different methods

Download full size

View in Article

Fig. 6. Line chart of evaluation indexes with different methods

Download full size

View in Article

Fig. 7. Radar chart of evaluation indexes with different methods

Download full size

View in Article

Table 1. Mean of normalized evaluation indexes with different modules

View table

View in Article

Table 1. Mean of normalized evaluation indexes with different modules

Methods	MSE	SF	CC	PSNR	$Q_{A B}^{F}$	MS-SSIM
Baseline	0.068 9	0.153 4	0.517 0	0.062 3	0.763 1	0.000 0
Without ECB	0.594 5	0.204 5	0.721 0	0.558 2	0.247 0	0.709 7
Without CBAM	0.415 8	0.877 4	0.333 1	0.393 4	0.465 6	0.895 4
DBECFuse	1.000 0	0.352 0	0.800 5	1.000 0	0.613 5	0.980 3

Table 2. Mean of normalized evaluation indexes with different params

View table

View in Article

Table 2. Mean of normalized evaluation indexes with different params

Params	MSE	SF	CC	PSNR	$Q_{A B}^{F}$	MS-SSIM
$α = 0.5, μ = 0.1$	0.489 5	0.082 1	0.416 2	0.479 3	0.774 6	0.215 7
$α = 0.5, μ = 1.0$	0.507 9	0.108 2	0.570 2	0.495 5	0.595 8	0.301 4
$α = 0.5, μ = 10$	0.419 2	0.085 8	0.550 3	0.403 6	0.997 1	0.126 2
$α = 1.0, μ = 0.1$	0.520 4	0.078 8	0.563 5	0.506 8	0.534 4	0.384 8
$α = 1.0, μ = 1.0$	0.522 9	0.089 6	0.638 2	0.507 6	0.516 3	0.416 8
$α = 1.0, μ = 10$	0.420 5	0.067 6	0.604 1	0.402 9	0.841 6	0.136 8
$α = 5.0, μ = 0.1$	0.364 5	1.000 0	0.328 6	0.349 9	0.154 1	0.920 3
$α = 5.0, μ = 1.0$	0.959 9	0.082 7	0.942 5	0.951 9	0.010 0	0.880 4
$α = 5.0, μ = 10$	0.717 6	0.104 0	0.836 0	0.694 4	0.306 3	0.749 4

Table 3. Mean of normalized evaluation indexes with different methods

View table

View in Article

Table 3. Mean of normalized evaluation indexes with different methods

Methods	MSE	SF	CC	PSNR	$Q_{A B}^{F}$	MS-SSIM
MST-SR	0.676 1	0.775 9	0.702 6	0.633 7	0.050 1	0.710 1
GTF	0.307 6	0.196 6	0.172 9	0.252 2	0.687 9	0.400 3
NestFuse	0.527 9	0.530 6	0.679 0	0.447 4	0.099 2	0.783 0
Res2NetFuse	0.571 9	0.363 5	0.653 9	0.491 2	0.186 5	0.651 3
SwinFuse	0.233 5	0.559 0	0.441 7	0.204 8	0.676 3	0.237 1
SeAFusion	0.239 5	0.820 5	0.746 9	0.182 1	0.063 7	0.708 5
PIAFusion	0.379 8	0.611 0	0.672 3	0.301 3	0.131 5	0.727 2
FECFusion	0.452 6	0.706 8	0.663 7	0.374 3	0.248 4	0.751 8
DBECFuse	0.904 6	0.373 7	0.997 2	0.877 3	0.494 5	1.000 0

Table 4. Params and FLOPs with different methods

View table

View in Article

Table 4. Params and FLOPs with different methods

Methods	Params	FLOPs
NestFuse	2.732 8 M	152.422 7 G
Res2NetFuse	89.617 K	18.549 3 G
SwinFuse	336.193 K	67.439 4 G
SeAFusion	166.657 0 K	21.758 0 G
PIAFusion	1.176 1 M	154.023 2 G
FECFusion	144.453 K	18.772 1 G
DBECFuse	33.317 K	8.907 1 G

Tools

Get Citation

Copy Citation Text

Hongde ZHANG, Xin FENG, Jieming YANG, Guohang QIU. A Dual Branch Edge Convolution Fusion Network for Infrared and Visible Images[J]. Acta Photonica Sinica, 2024, 53(8): 0810004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites