摘要
高光谱图像的波段范围广、光谱分辨率高,能为图像分类研究提供丰富的信息,但同时也给计算和存储带来了较大困难。论文提出一种基于SNMF聚类与类间可分性因子的方法来进行高光谱图像波段选择,以降低计算和存储开销。首先是数据预处理工作,将高光谱数据进行三维转二维表达,然后利用SNMF聚类算法得到波段的各个类簇,最后以各波段的类间可分性因子为指标在类簇内进行波段选择。实验采用波段子集的平均信息熵、平均相关系数和平均相对熵三类指标进行定量评价,并采用SVM分类器进行分类验证。
Hyperspectral image has a wide range of bands and high spectral resolution,which can provide rich information for image classification research,but it also brings great difficulties for calculation and storage.In this paper,a method based on SNMF clustering and inter-class separability factor is proposed to select representative subsets of hyperspectral images,which can reduce the computational complexity while ensuring classification accuracy.First is data preprocessing work,which transforms three-dimensional hyperspectral images into two-dimensional representations.Then SNMF clustering algorithm is used to get band cluster.Finally,the inter-class separability factor of each band is calculated according to the ground object type,and it is taken as a reference indicator to further select representative bands within the class cluster,so as to achieve the effective selection of bands.In the experiment,the average information entropy,average correlation coefficient and average relative entropy of band subset are used for quantitative evaluation,and SVM classifier is used for classification verification.
作者
赵玉英
任明武
ZHAO Yuying;REN Mingwu(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)
出处
《计算机与数字工程》
2021年第9期1884-1888,共5页
Computer & Digital Engineering
基金
国家自然科学基金重大仪器专项“熔焊与增材成形熔池多物理场在线测量仪”(编号:61727802)资助。
关键词
高光谱图像
波段选择
稀疏非负矩阵分解
类间可分性
hyperspectral image
band selection
sparse non-negative matrix factorization
inter-class separability
