基于ResNet50与卷积稀疏表达的红外与可见光图像融合算法
THE INFRARED AND VISIBLE IMAGE FUSION ALGORITHM BASED ON RESNET50 AND CONVOLUTION SPARSE REPRESENTATION
邵大光 1邵现振 2刘鹏 2赵闯 2陶青川1
作者信息
- 1. 四川大学电子信息学院 四川成都 610000
- 2. 中国石油化工股份有限公司胜利油田分公司河口采油厂 山东东营 257000
- 折叠
摘要
提出一种基于ResNet50神经网络与卷积稀疏表达的红外与可见光图像融合算法.通过低通滤波将红外与可见光图像分解成基础层和细节层;运用卷积稀疏表达对基础层进行处理得到新的基础层,使用ResNet50神经网络对细节层进行特征提取,对得到的特征图进行L1正则化和最大选择策略得到最大权重层,经过权重分配得到新的细节层;对新的基础层和细节层进行重建,得到融合图像.该算法针对基础层和细节层提出了新的融合策略,并且能较好地保留细节信息和结构信息.实验结果表明,该算法在主观和客观指标证明上优于对比算法.
Abstract
This paper proposes an infrared and visible image fusion algorithm based on ResNet50 neural network and convolution sparse representation.The infrared and visible images were decomposed into the base layers and the detail layers through low-pass filtering.The convolution sparse representation was utilized to process the base layer to obtain the fused base layer,and the ResNet50 neural network was applied to extract the features of the detail layer.For the feature map,we performed L1-regularization and choose-max strategy to obtain the maximum weight layer.And the fused detail layer could be generated through weight distribution.We generated the new base layer and detail layer to obtain the fused image.We proposed new fusion strategies for the base layer and the detail layer,and the fused image retained more detail information and structural information.The experimental results show that our method is superior to the comparison algorithm in subjective and objective metrics.
关键词
图像融合/ResNet50/卷积稀疏表达/红外图像/可见光图像Key words
Image fusion/ResNet50/Convolution sparse representation/Infrared image/Visible image引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据