高光谱影像逆近邻密度峰值聚类的波段选择算法  

Band selection algorithm for reverse nearest neighbor density peak clustering of hyperspectral images

作　　者：孙根云[1,2] 李忍忍 张爱竹 安娜[3] 付航 潘兆杰 SUN Genyun;LI Renren;ZHANG Aizhu;AN Na;FU Hang;PAN Zhaojie(College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao 266580,China;Laboratory for Marine Mineral Resources,Qingdao National Laboratory for Marine Science and Technology,Qingdao 266071,China;China Aero Geophysical Survey and Remote Sensing Center for Natural Resources,Beijing 100083,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580 [2]海洋国家实验室海洋矿产资源评价与探测技术功能实验室,山东青岛266071 [3]中国自然资源航空物探遥感中心,北京100083

出　　处：《测绘学报》2024年第1期8-19,共12页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(42271347,41971292);科技部国家重点研发计划(2019YFE26700)。

摘　　要：密度峰值聚类波段选择算法利用局部密度描述波段的密度信息,然而现有的局部密度容易忽略波段分布的全局信息,不能有效描述波段的分布特征,导致波段子集分类精度有限。为解决上述问题,本文提出一种基于逆近邻的密度峰值聚类波段选择算法。首先,利用波段与其K近邻构建K近邻有向图,获取波段的逆近邻,以及波段之间的共享近邻和共享逆近邻;然后,利用共享近邻和共享逆近邻并集的个数作为波段之间的相似度,利用波段与其逆近邻的平均欧氏距离和相似度构造增强型局部密度;最后,将增强型局部密度、距离因子、信息熵三者的乘积作为权重值,根据权重值挑选波段子集。为提高试验效率和实用性,本文算法还提出一种自动获得K值的自适应K值方法。在3个高光谱标准数据集上的试验结果表明,本文算法得到的波段子集比其他先进算法挑选的波段有更好的分类性能,尤其是在波段数较少的情况下,而且计算效率较高。The density peak clustering band selection algorithm uses the local density to describe the density information of the band.However,the existing local density is easy to ignore the global information of the band distribution and can t effectively describe the distribution characteristics of the band,resulting in the limited classification accuracy of the band subset.In order to solve the above problems,this paper proposes a density peak clustering band selection algorithm based on reverse nearest neighbor.Firstly,the K-nearest neighbor directed graph is constructed by using the band and its K-nearest neighbor to obtain the reverse nearest neighbor of the band,as well as the shared nearest neighbor and shared reverse nearest neighbor between bands.Then,the union number of shared nearest neighbors and shared reverse nearest neighbors is used as the similarity between bands,and the enhanced local density is constructed by using the average Euclidean distance and similarity between bands and their reverse nearest neighbors.Finally,the product of enhanced local density,distance factor and information entropy is taken as the weight value,and the segment subset is selected according to the weight value.In order to improve the efficiency and practicability of the experiment,an adaptive K value method is also proposed in this paper.The experimental results on three hyperspectral standard data sets show that the band subset obtained by this algorithm has better classification performance than the band selected by other advanced algorithms,especially when the number of bands is small,and the calculation efficiency is high.

关 键 词：高光谱影像 波段选择 密度峰值聚类 逆近邻 局部密度 自适应K值 

分 类 号：P237[天文地球—摄影测量与遥感]