基于多尺度空间注意力互补的红外与可见光图像融合

打开文本图片集

中图分类号：TP391. 4 文献标识码：Adoi：10. 37188/OPE. 20253307. 1152 CSTR：32169. 14. OPE. 20253307. 1152

Infrared and visible image fusion based on multi-scale spatial attention complementary

ZHANG Yongxing 1，2，3 ，LIAN Bowen 1，2，3 ，GU Naiting 4* ，LI Fangzhao 4 ，LI Yang 1，2*

（1. National Key Laboratory of Adaptive Optics，Chengdu 610209，China； 2. Institute on Optics and Electronics Technology，Chinese Academy of Sciences， Chengdu 610209，China； 3. University of Chinese Academy of Sciences，Beijing 100049，China； 4. College of Frontier Interdisciplinary Sciences，National University of Defense Technology， Changsha ，China） * Corresponding author，E-mail：gnt7328@163. com；liyang@ioe. ac. cn

Abstract：Current infrared and visible image fusion methods tend to introduce excessive redundant infrared information，impairing the ability to balance complex scene details and resulting in suboptimal fusion out⁃ comes. To address these limitations，a novel fusion approach based on multi-scale spatial attention com ⁃ plementarity is proposed. This method employs a dual-branch convolutional network to separately extract features from infrared and visible images，followed by difference-based complementary processing. Multiscale spatial attention mechanisms are then applied to the feature maps，culminating in regression-based su⁃ perposition to achieve balanced fusion of complementary features. Experimental evaluations demonstrate that， compared to mainstream methods such as Densefuse and PIAFusion， the proposed approach achieves improvements of 4.1% and 4.3% in mutual information（MI），and 5.0% and 2.3% in visual in⁃ formation fidelity（VIF），respectively. These results indicate enhanced retention of target features and ef⁃ fective suppression of redundant information within complex scenes. The method exhibits strong feature balancing capabilities and holds significant potential for applications in target detection and recognition un⁃ der challenging environmental conditions.

Key words：image fusion；infrared and visible image；double branch convolutional network；complemen⁃ tary difference；multi-scale spatial attention；regression overlay

1 引 言

随着光电探测水平的提升，对目标检测与识别的需求不断增强。（剩余26635字）