Researching | Dual-band Infrared Image Fusion Method based on VMamba

AEROSPACE SHANGHAI, Volume. 42, Issue 4, 180(2025)

Dual-band Infrared Image Fusion Method based on VMamba

Miao XIN¹, Yang RUAN², Yusong LI^3、*, and Shaoyi LI³

Author Affiliations

¹Unit 93145 of the People's Liberation Army，Shanghai201109，China

²Shanghai Academy of Spaceflight Technology，Shanghai201109，China

³School of Astronautics，Northwestern Polytechnical University，Xi΄an710072，，China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (11)

References (25)

Paper Information

Figures & Tables(11)

Fig. 1. Dual-band infrared image fusion network based on the VMamba

Download full size

View in Article

Fig. 2. Structure of the VMamba block

Download full size

View in Article

Fig. 3. Structure of the semantic feature extraction and fusion module

Download full size

View in Article

Fig. 4. Schematic diagram of the target and its neighborhood

Download full size

View in Article

Fig. 5. Qualitatively experimental results

Download full size

View in Article

Table 1. Tab.1Pseudo-code process of the S6 module

View in Article

Table 1. Tab.1Pseudo-code process of the S6 module

算法：S6

输入：图像特征序列

x \in R^{B \times L \times S}

，其中，

B

为批次大小，

L

为token长度，

S

为特征维度大小。

参数：

A 、 D

可训练的参数矩阵。

操作： $L_{L i n e a r} (\cdot)$ ，线性投影操作， $E_{E m b e d d i n g} (\cdot)$ ，块嵌入操作。

输出：图像特征序列 $y \in R^{B \times L \times S}$ ，

1）将 $x$ 通过不同的线性投影操作得到 $Δ, B, C$ ：

$Δ, B, C = L_{L i n e a r} (x), L_{L i n e a r} (x), L_{L i n e a r} (x)$ （7）

2）参数离散化:

$\bar{A} = Δ A$ （8）

$\bar{B} = {(Δ A)}^{- 1} (Δ A - I) \cdot Δ B$ （9）

式中：

Δ

为时间步长，用于控制关注或忽略当前输入的程度，类似于遗忘门。

3）计算下一状态 $h_{t}$ :

$h_{t} = \bar{A} h_{t - 1} + \bar{B} x_{t}$ （10）

4）计算当前计算单元的输出：

$y_{t} = C h_{t} + D x_{t}$ （11）

5）合并所有计算单元的输出：

$y = [y_{1}, y_{2}, \dots, y_{L}]$ （12）

6）返回

y \in R^{B \times L \times S}

Table 2. Tab.2Experimental environment
View table
View in Article
Table 2. Tab.2Experimental environment
配置版本
系统 Windows10/Ubuntu 18.04
CPU Intel®Core(TM) i9-10900X，3.70 Hz
显卡 GeForce GTX 3090
显存/GB 24
CUDA 12.1
CuDNN 8.9.5
Cudatoolkit 12.1
深度学习框架 PyTorch1.12.1

Table 3. Tab.3Quantitatively experimental results of the fusion algorithm

View in Article

Table 3. Tab.3Quantitatively experimental results of the fusion algorithm

类型	算法名称	SCR↑	BR $↓$
源图像	Long Wave	0.305 886	0.001 585
源图像	Mid Wave	0.640 464	0.004 022
传统算法	LP	0.605 097	0.004 096
	RP	0.618 828	0.004 045
	CVT	0.628 430	0.004 062
	MSVD	0.445 668	0.002 893
深度学习算法	DDcGAN	0.382 607	0.003 636
	SwinFusion	0.626 262	0.004 019
	RFN-Nest	0.415 463	0.004 893
	Ours	0.638 514	0.002 846

Table 4. Tab.4Computational efficiency results
View table
View in Article
Table 4. Tab.4Computational efficiency results
算法名称参数量/M↓ FLOPs/G↓ FPS↑
DDcGAN 2.040 9 23.415 9 5.414 5
SwinFusion 0.927 5 327.677 8 0.606 1
RFN-Nest 7.524 2 555.529 3 13.038 0
Ours 0.672 2 16.571 3 23.630 5

Table 5. Tab.5Results of the ablation experiments on the network structures
View table
View in Article
Table 5. Tab.5Results of the ablation experiments on the network structures
VMamba
特征提取
目标与杂波
语义特征提取
混合特征
融合模块
SCR↑ BR↓
√ 0.521 564 0.003 814
√ √ 0.588 006 0.003 228
√ √ √ 0.638 514 0.002 846

Table 6. Tab.6Results of the ablation experiments on the key parameters

View in Article

Table 6. Tab.6Results of the ablation experiments on the key parameters

目标语义特征权重	杂波语义特征权重	SCR↑	BR↓
$α = 0$	$β = 1$	0.508 648	0.002 380
$α = 1$	$β = 1$	0.543 706	0.003 048
$α = 5$	$β = 1$	0.586 835	0.003 688
$α = 5$	$β = 2$	0.559 235	0.003 275
$α = 12$	$β = 2$	0.596 277	0.002 955
$α = 20$	$β = 3$	0.572 877	0.003 626
$\tilde{α} = 12$	$\tilde{β} = 2$	0.638 514	0.002 846

Tools

Get Citation

Copy Citation Text

Miao XIN, Yang RUAN, Yusong LI, Shaoyi LI. Dual-band Infrared Image Fusion Method based on VMamba[J]. AEROSPACE SHANGHAI, 2025, 42(4): 180

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Speciality Discussion

Received: Oct. 17, 2024

Accepted: --

Published Online: Sep. 29, 2025

The Author Email:

DOI:10.19328/j.cnki.2096-8655.2025.04.019

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology