基于最优子集准则的高光谱图像波段选择算法研究

发布时间：2018-05-05 09:17

本文选题：高光谱图像 + 波段选择　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：高光谱遥感丰富的光谱信息带来的海量数据大大增加了数据处理的时空复杂度并影响处理结果,有效的波段选择方法可以在大大提高高光谱图像处理速度的同时改善处理效果。因此,研究高光谱图像波段选择技术具有重要意义。本文主要研究基于最优子集准则的高光谱波段选择算法框架及具体算法过程,论文主要工作如下:(1)综述了高光谱图像波段选择的背景及意义、并分析总结国内外研究现状,确定本文研究的主要内容和预期成果,最后说明本文组织结构。(2)对本文涉及到的相关基础知识进行系统介绍,包括粒子群优化算法的原理及参数设置、常见的聚类算法,为后面提出的几种波段选择算法奠定了理论基础。(3)介绍高光谱图像波段选择的最优子集准则和常见的搜索策略,并引出本文用到的准则和搜索策略。然后构建基于最优子集准则直接寻优的波段选择算法框架,以此框架为基础提出基于MNBS准则粒子群直接寻优的波段选择算法,该算法利用连续前向选择(SFS)搜索一组较好的波段子集,以此来初始化粒子群算法的初始粒子位置列表,跟随机初始化相比,降低了算法本身的初始敏感问题带来的影响;构建并简单介绍基于聚类结合最优子集寻优的波段选择算法框架,基于该框架提出一种使用谱聚类对波段进行聚类,然后结合MNBS准则粒子群直接寻优的波段选择算法;最后设计基于关键波段选择和最优子集寻优的波段选择算法框架,提出了基于光谱角距离度量的聚类性能有效性指标,并结合视觉评估集群趋势确定聚类类别,然后根据关键波段提取结合MNBS准则粒子群寻优的波段选择算法。(4)对本文验证实验用到的RX异常检测和两种分类方法(KNN、SVM)做了简单介绍,通过异常检测和分类实验,对提出的几种波段选择算法进行性能比较和分析。(5)总结本文主要做的工作,并对基于最优子集的波段选择的发展方向进行展望。
[Abstract]:The massive data brought by hyperspectral remote sensing information greatly increases the time and space complexity of the data processing and affects the processing results. The effective band selection method can improve the processing speed of hyperspectral image processing and improve the processing effect. Therefore, it is of great significance to study the band selection technique of hyperspectral images. This paper mainly studies the framework and specific algorithm process of hyperspectral band selection algorithm based on optimal subset criteria. The main work of this paper is as follows: (1) the background and significance of spectral band selection in hyperspectral images are reviewed, and the research status at home and abroad is summarized, the main content and expected results of this study are determined, and the structure of this paper is explained at the end of the paper (2) The basic knowledge involved in this paper is systematically introduced, including the principle and parameter setting of particle swarm optimization algorithm. The common clustering algorithm lays a theoretical foundation for several band selection algorithms proposed later. (3) introduce the optimal subset criterion and common search strategy of spectral band selection for hyperspectral images, and lead to the use of this paper. In this framework, a band selection algorithm based on MNBS criterion particle swarm optimization is proposed. The algorithm uses continuous forward selection (SFS) to search a set of better set of wavelet subsets to initialize the initial particle swarm optimization algorithm. The initial particle location list, compared with random initialization, reduces the impact of the initial sensitivity of the algorithm itself. The framework of band selection algorithm based on clustering and optimal subset optimization is introduced. Based on the framework, a spectral clustering is used to cluster the bands, and then the MNBS criterion particle swarm optimization is used to direct the optimization. The band selection algorithm based on the key band selection and the optimal subset optimization is designed. The clustering performance effectiveness index based on the spectral angular distance measurement is proposed. The clustering classification is determined by the vision evaluation cluster trend, and then the band selection based on the MNBS criterion particle swarm optimization is extracted according to the key wave segment. (4) a brief introduction is made to the RX anomaly detection and two classification methods (KNN, SVM) used in this experiment. The performance comparison and analysis of several band selection algorithms are carried out by abnormity detection and classification experiments. (5) the main work of this paper is summarized, and the development direction of band selection based on the optimal subset is carried out. Look ahead.

【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】