基于分类分块的全色多光谱图像比值变换融合方法  

Panchromatic and Multispectral Image Fusion by Ratio Enhancement Based on Classification and Block

作　　者：魏晴 刘晋铭 徐其志 WEI Qing;LIU Jinming;XU Qizhi(Beijing University of Chemical Technology,Beijing 100029,China;National Defense Engineering Research Institute of Academy of Military Science,Beijing 100036,China)

机构地区：[1]北京化工大学,北京100029 [2]军事科学院国防工程研究院,北京100036

出　　处：《航天返回与遥感》2021年第5期108-117,共10页Spacecraft Recovery & Remote Sensing

基　　金：国家基金面上项目(61972021,61672076)。

摘　　要：低分辨率全色图像合成是全色与多光谱图像融合的关键环节,决定了融合图像的光谱色彩和空间细节品质。由于光学卫星成像传感器的非线性光谱响应以及不同地物的光谱响应差异,导致直接利用多光谱图像各波段合成的低分辨率全色图像存在灰度失真,以致融合图像产生光谱色彩失真和空间细节失真。文章首先利用一种迭代求解的聚类分析算法k均值聚类算法(k-means算法)对全色和多光谱图像所有像素进行分类;然后根据各类像素的方差将其划分为不同大小的子块,利用带约束最小二乘法计算多光谱图像每个子块中各个像素的加权值;最后利用加权值合成低分辨率全色图像,并通过比值变换方法生成高保真融合图像。文章利用QuickBird和WorldView3卫星的全色与多光谱图像进行实验验证,与PNN、PanNet、FusionCNN等方法相比,该方法有效提高了融合图像的光谱色彩和融合细节的保真度。The synthesis of low-resolution panchromatic image is a key step in panchromatic and multispectral image fusion,it determines the spectral and spatial quality of the fused imagery.Due to the nonlinear spectral response of optical satellite sensor and the differences in spectral response of different ground objects,the low-resolution panchromatic image synthesized by each band of the multispectral image has gray-scale distortion which causes the spectral and spatial distortion of fused imagery.The proposed method firstly used an iterative solution clustering analysis algorithm,namely k-means algorithm to classify all pixels of panchromatic and multispectral images;then divide them into sub-blocks with different sizes according to the variance of each class of pixels,after that,each pixel′s weights in each sub-block of the multispectral image are calculated by constrained least square method;finally,the weights are used to synthesize the low-resolution panchromatic image,and the high-Fidelity fused imagery is generated by ratio enhancement.This paper uses the panchromatic and multispectral images of QuickBird and WorldView3 satellite for experimental verification.Compared with PNN,PanNet and FusionCNN,the proposed method effectively improves the fidelity of the fused imagery’s spectral color and fusion detail.

关 键 词：图像融合 比值变换 像素分类 带约束最小二乘法 遥感图像 

分 类 号：V445[航空宇航科学与技术—飞行器设计]