基于二叉树锚点的高光谱快速聚类算法  

Hyperspectral Fast Clustering Algorithm Based on Binary Tree Anchor Points

作　　者：许裕雄 杨晓君 蔡湧达 杜晓颜 张鑫 Xu Yuxiong;Yang Xiaojun;Cai Yongda;Du Xiaoyan;Zhang Xin(College of Information Engineering,Guangdong University of Technology,Guangzhou,Guangdong 510006,China;Chinese People's Liberation Army 96630 Troops,Beijing 102206,China)

机构地区：[1]广东工业大学信息工程学院,广东广州510006 [2]中国人民解放军96630部队,北京102206

出　　处：《激光与光电子学进展》2021年第2期211-218,共8页Laser & Optoelectronics Progress

基　　金：科技部重大专项(2018YFB1802100)。

摘　　要：高光谱图像聚类问题一直是图像处理领域的研究热点。谱聚类算法是最流行的聚类算法之一,但其计算复杂度较大,难以处理大规模的高光谱图像数据。由于二叉树能够较快地选取锚点,因此基于二叉树锚点图,充分利用高光谱图像的光谱和空间特性,可保证聚类性能并降低计算复杂度。然而,该聚类算法一般采用有核的聚类方法,因此不可避免地引入了参数调节。在二叉树锚点选取的基础上,提出了一种基于二叉树锚点的高光谱快速聚类算法,该算法创新性地将二叉树锚点选取和无核聚类方法应用于高光谱图像中。首先,利用二叉树从高光谱数据中选取一些具有代表性的锚点;紧接着构造基于锚点的无核相似图,有效避免了通过人为调节热核参数来构造相似图;然后进行谱聚类分析获得聚类结果;最后,将该算法应用到高光谱图像聚类中。该算法不仅提高了聚类速度,还减少了原有热核参数调节。实验结果表明,与传统的聚类算法相比,所提算法能够在较短的时间内获得更佳的聚类精度。Hyperspectral image clustering has always been a hot topic in the field of image processing.Spectral clustering algorithm,as one of the most popular clustering algorithms,is widely used in hyperspectral image clustering.However,due to the large computational complexity of the spectral clustering algorithm,it is difficult to process large-scale hyperspectral image data.Because the binary tree can select anchor points very fast,the spectral and spatial characteristics of a hyperspectral image are fully utilized to ensure the clustering performance and reduce the computational complexity based on the binary tree anchor graph.However,the clustering algorithm generally adopts the kernel clustering method,therefore it is inevitable to introduce parameter adjustment.Thus,based on the selection of anchor points in the binary tree,we proposes a hyperspectral fast clustering algorithm based on the binary tree anchor graph.This algorithm innovatively applies the method of binary tree anchor selection and coreless clustering to the hyperspectral images.First,the binary tree is used to select some representative anchor points from the hyperspectral data.Second,a coreless similarity map is constructed based on these anchor points,which effectively avoids the artificial adjustment of the thermonuclear parameters to construct the similarity map.Third,the spectral clustering analysis is performed to obtain the clustering results.Finally,this algorithm is used for hyperspectral image clustering.This algorithm not only improves the clustering speed,but also reduces the necessity of original thermonuclear parameter adjustment.The experimental results show that the proposed algorithm can obtain better clustering accuracy in a shorter time compared with the traditional clustering algorithm.

关 键 词：图像处理 高光谱图像 谱聚类 二叉树 锚点 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]