非线性变换和信息相邻相关的高光谱自适应波段选择

发布时间：2018-07-15 08:26

【摘要】：通过非线性函数变换改进后的谱间Pearson相关分析可同时获取高光谱影像光谱间的综合相关系数(r_(cl))、相关类型和统计显著性水平;研究发现,非线性是高光谱影像的谱间相关性的主要类型。基于相关系数的波段相邻相关系数(r_(ac))在自适应波段选择算法(ABS)中是为了表达波段的独立性,然而发现ABS算法中rac并不能有效表达波段独立性。鉴于此,提出了一种信息相邻相关系数(r_(iac))和基于此指数改进的自适应波段选择算法(MABS)。使用公共数据和实验室采集数据,对ABS、基于线性相关系数(r_l)的MABS(r_l)和基于r_(cl)的MABS(r_(cl))等三种算法进行实验。结果表明:在波谱范围和算法有效性及精度方面,MABS均优于ABS;MABS较好地兼顾了大信息量和强独立性原则,其波段选择结果的光谱范围明显大于ABS;MABS(r_(cl))的光谱范围略大于MABS(r_l);三种算法的总体分类精度(OA)和Kappa系数的大小顺序均为:MABS(r_(cl))MABS(r_l)ABS。
[Abstract]:The comprehensive correlation coefficient (r _ (cl), correlation type and statistical significance level) between spectra of hyperspectral images can be obtained simultaneously by the improved Pearson correlation analysis of nonlinear function transformation. Nonlinearity is the main type of spectral correlation of hyperspectral images. The wavelength adjacent correlation coefficient (r _ (ac) based on the correlation coefficient is used to express the band independence in the adaptive band selection algorithm. However, it is found that the rac can not express the band independence effectively. In view of this, an information adjacent correlation coefficient (r _ (iac) and an improved adaptive band selection algorithm (Mabs) based on this index are proposed. By using common data and laboratory data, three algorithms, Abs, MABS based on linear correlation coefficient (rashl) and MABS (r _ (cl) based on r _ (cl), are tested. The results show that MABS is better than ABSMABS in spectrum range, validity and accuracy of algorithm, and the principle of large amount of information and strong independence is well taken into account. The spectral range of band selection is obviously larger than that of (cl) (r _ (cl), and the order of total classification accuracy (OA) and Kappa coefficient of the three algorithms is in the order of: MABS (r _ (cl) MABS (rl) ABSs).
【作者单位】：首都师范大学三维信息获取与应用教育部重点实验室;首都师范大学空间信息技术教育部工程研究中心;
【基金】：国家自然科学基金(41571369) 教育部博士点基金(20131108110005) 北京市长城学者(20150323)
【分类号】：TP751

【参考文献】