基于卷积神经网络的多波段融合图像彩色化方法  

Multiband Image Fusion Results Colorization Based on Convolution Neural Network

作　　者：韩泽 蔺素珍[1] 黄福升 赵竞超 刘震[1] HAN Ze;LIN Suzhen;HUANG Fusheng;ZHAO Jingchao;LIU Zhen(School of Data Science, North University of China, Tai Yuan 030051, China)

机构地区：[1]中北大学大数据学院,山西太原030051

出　　处：《测试技术学报》2018年第3期201-206,共6页Journal of Test and Measurement Technology

基　　金：山西省应用基础研究资助项目(201701D121062);中北大学第十四届研究生科技立项资助项目(20171443)

摘　　要：为解决多波段融合图像多在灰度空间不利于人眼观察的问题,提出了基于深度卷积神经网络(Deep Convolution Neural Network,DCNN)的多波段融合图像彩色化方法.首先将利用颜色迁移和伪彩色融合制作的部分彩色融合结果添加到网络训练库中;其次将训练图像转换到YUV颜色空间;再其次构建卷积神经网络,在其输入端输入Y通道图像,以UV通道图像为目标训练网络,使其能根据输入自动生成UV通道;最后将灰度融合结果作为Y通道输入到训练好的网络,将输出的UV通道与输入再转换到RGB空间,即可得到彩色化的融合结果.实验结果表明DCNN能对灰度融合结果自动彩色化,方法简单易用,彩色更便于人眼观察.In order to improve the observation effect of human eyes on multi-band fusion image in gray space,a multi band fusion image colorization method based on deep convolution neural network（DCNN）is proposed.Firstly,the color fusion results which are made of color migration and pseudo-color fusion are added into the training set;secondly,the training images are converted to YUVcolor space;then the convolution neural network is constructed,and the Y channel image is input at the input end,the UV channel image is taken as the target training network,so that the UVchannel can be generated automatically according to the input;finally,the gray level fusion results are input to the trained network as Y channel,and the output UVchannel and input are converted to RGB space to obtain the color fusion result.The experimental results show that DCNN can automatically color the gray fusion results.The method is simple and easy,and the color is more convenient to be observed.

关 键 词：图像融合 深度学习 彩色化 卷积神经网络 多波段探测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]