Cross-Scale Pooling Embedding Image Fusion Algorithm with Long- and Short-Distance Attention Collaboration

Xicheng Sun; Fu Lü; Yongan Feng

doi:10.3788/LOP242159

Laser & Optoelectronics Progress, Volume. 62, Issue 12, 1237003(2025)

Cross-Scale Pooling Embedding Image Fusion Algorithm with Long- and Short-Distance Attention Collaboration

Xicheng Sun^1、*, Fu Lü^1,2, and Yongan Feng³

Author Affiliations

¹School of Software, Liaoning Technical University, Huludao 125105, Liaoning , China

²Basic Teaching Department, Liaoning Technical University, Huludao 125105, Liaoning , China

³Informatization and Network Management Center, School of Software, Liaoning Technical University, Huludao 125105, Liaoning , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Network architecture of proposed algorithm

Download full size

Fig. 2. Structure of MDCM module

Download full size

Fig. 3. Structure of CA module

Download full size

Fig. 4. Structure of LSACM module

Download full size

Fig. 5. Comparative experimental results on the TNO dataset by different algorithms

Download full size

Fig. 6. Comparative experimental results on the Roadscene dataset by different algorithms

Download full size

Fig. 7. Qualitative comparison of ablation experiments on the TNO dataset

Download full size

Table 1. CPEL parameters
View table
Table 1. CPEL parameters
Stage Input dimension Kernel Stride Output dimension
M1 64 8 4 16
M2 128 4 4 16
M3 256 4 2 32
M4 512 2 1 64

Table 2. Comparative experimental results on the TNO and Roadscene datasets by different algorithms

View table

Table 2. Comparative experimental results on the TNO and Roadscene datasets by different algorithms

Algorithm	TNO dataset						Roadscene dataset
Algorithm	EN	SD	VIF	SF	AG	SCD	EN	SD	VIF	SF	AG	SCD
DenseFuse	6.4572	8.3479	0.6722	0.0326	2.5764	1.5794	7.0279	9.8652	0.7628	0.0368	3.0981	1.6528
DIDFuse	6.9832	10.0915	0.8172	0.0468	5.2389	1.8024	7.2380	10.3480	0.8504	0.0428	4.0921	1.6923
RFN	7.0429	9.9320	0.7743	0.0239	2.7643	1.5820	7.3420	9.5490	0.8044	0.0279	2.4590	1.7732
SwinFusion	6.8753	9.5628	0.7763	0.0562	4.2981	1.5290	6.9342	10.2390	0.8135	0.0423	3.0918	1.6583
U2Fusion	6.4380	9.3489	0.6270	0.0347	2.5881	1.3780	7.0234	10.1246	0.7392	0.0423	4.2370	1.4593
LRRNet	6.8583	9.3091	0.7123	0.0478	3.5692	1.5890	6.9023	9.3780	0.8318	0.0428	3.9820	1.5638
CoCoNet	7.7552	10.153	0.6532	0.0632	5.6893	1.7378	7.6322	10.3567	0.8543	0.0717	5.3480	1.8023
Proposed	7.8328	9.9738	0.8073	0.0611	6.0170	1.8027	7.7281	10.4027	0.8530	0.0627	5.5273	1.8172

Table 3. Quantitative comparison of ablation experiments on the TNO and Roadscene datasets

View table

Table 3. Quantitative comparison of ablation experiments on the TNO and Roadscene datasets

Algorithm	TNO dataset				Roadscene dataset
Algorithm	EN	SD	VIF	SCD	EN	SD	VIF	SCD
w/o LSACM	6.4721	7.3482	0.6573	1.6483	6.6728	9.2416	0.7351	1.7263
w/o CA	7.2368	7.4522	0.6758	1.6783	7.4728	10.0273	0.8126	1.7826
w/o CPEL	6.3621	7.6721	0.7134	1.6834	6.8356	9.3726	0.7635	1.7362
w/o MDCM	7.1242	8.3472	0.7284	1.7365	7.3627	10.1637	0.8015	1.8126
Propsoed	7.8283	9.9637	0.7936	1.8028	7.8328	10.4136	0.8541	1.8162

Table 4. Comparative experimental results on the running efficiency of eight algorithms

View table

Table 4. Comparative experimental results on the running efficiency of eight algorithms

Algorithm	Params /10⁶	FLOPs /10⁹	Time /s
DenseFuse	0.925	497.06	0.124
DIDFuse	0.261	18.71	0.055
RFN	10.936	676.09	0.239
U2Fusion	0.659	366.34	0.123
SwinFusion	0.974	471.04	1.345
LRRNet	0.492	134.79	0.079
CoCoNet	9.130	115.37	0.052
Proposed	4.5382	146.83	0.069

Table 5. Comparison of object detection performance on the M³FD dataset

View table

Table 5. Comparison of object detection performance on the M³FD dataset

Image	AP						mAP@0.5
Image	Bus	Car	Lamp	Motor	People	Truck	mAP@0.5
Infrared	78.23	87.39	70.12	61.24	78.33	65.27	74.43
Visible	78.29	90.39	87.39	68.35	70.14	70.28	77.47
DenseFuse	94.32	88.17	90.27	68.83	65.83	71.93	79.89
DIDFuse	79.63	92.49	84.37	68.39	79.36	68.63	78.81
RFN	78.17	91.83	84.38	72.91	78.39	68.73	79.07
SwinFusion	94.39	87.34	89.74	67.83	65.32	74.02	79.77
U2Fusion	95.15	88.84	88.94	68.52	65.38	72.63	79.91
LRRNet	77.81	86.39	83.68	67.73	62.38	69.63	74.60
CoCoNet	94.23	88.50	90.63	71.63	65.01	73.81	80.64
Proposed	94.52	88.68	90.53	71.83	71.58	74.63	81.96

Tools

Get Citation

Copy Citation Text

Xicheng Sun, Fu Lü, Yongan Feng. Cross-Scale Pooling Embedding Image Fusion Algorithm with Long- and Short-Distance Attention Collaboration[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1237003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Digital Image Processing

Received: Oct. 22, 2024

Accepted: Dec. 12, 2024

Published Online: Jun. 9, 2025

The Author Email: Xicheng Sun (3079194134@qq.com)

DOI:10.3788/LOP242159

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. CPEL parameters

Table 1. CPEL parameters

Table 2. Comparative experimental results on the TNO and Roadscene datasets by different algorithms

Table 2. Comparative experimental results on the TNO and Roadscene datasets by different algorithms

Table 3. Quantitative comparison of ablation experiments on the TNO and Roadscene datasets

Table 3. Quantitative comparison of ablation experiments on the TNO and Roadscene datasets

Table 4. Comparative experimental results on the running efficiency of eight algorithms

Table 4. Comparative experimental results on the running efficiency of eight algorithms

Table 5. Comparison of object detection performance on the M3FD dataset

Table 5. Comparison of object detection performance on the M3FD dataset

Table 5. Comparison of object detection performance on the M³FD dataset

Table 5. Comparison of object detection performance on the M³FD dataset