基于卷积神经网络的红外与可见光图像融合方法
Infrared and visible image fusion method based on convolutional neural network
李景景 1杜梅 1孙滨1
作者信息
- 1. 郑州工业应用技术学院信息工程学院,郑州 451100
- 折叠
摘要
针对当前红外与可见光图像融合方法存在融合效果差、效率低等问题,为了获得更优的红外与可见光图像融合结果,提出了基于卷积神经网络的红外与可见光图像融合方法.首先收集待融合红外与可见光图像,采用Retinex算法对图像进行增强操作,提升图像亮度和细节信息,然后采用卷积神经网络提取图像融合特征,并设计红外与可见光图像融合规则,根据规则得到图像融合结果,最后应用多个数据集进行红外与可见光图像融合性能测试实验,结果表明,卷积神经网络的图像融合后整体视觉效果好,重要细节信息丰富,熵和平均梯度的值超过了 6,融合时间低于1 s,整体性能优于其它红外与可见光图像对比融合方法.
Abstract
Aiming at the problems of the current infrared and visible image fusion methods such as poor fusion effect and low efficiency,in order to obtain better infrared and visible image fusion results,an infrared and visible im-age fusion method based on convolution neural network is proposed.First,collect the infrared and visible images to be fused,use the Retinex algorithm to enhance the image brightness and detail information,then use the convolution neu-ral network to extract the image fusion features,and design the infrared and visible image fusion rules,and get the im-age fusion results according to the rules.Finally,conduct the infrared and visible image fusion performance test on multiple data sets,and the results show that the image fusion of convolutional neural network has good overall visual effect,rich details,more than 6 values of entropy and average gradient,and the fusion time is less than 1 s.The over-all performance is better than its infrared and visible image contrast fusion method.
关键词
红外成像系统/可见光成像系统/卷积神经网络/融合规则/均梯度值Key words
infrared imaging system/visible light imaging system/convolution neural network/fusion rules/av-erage gradient value引用本文复制引用
基金项目
河南省科技厅科技攻关支持项目(222102210159)
教育部产学合作协同育人项目(202102258003)
出版年
2024
NETL
NSTL
万方数据