《高光谱影像特征择选与提取.ppt》由会员分享,可在线阅读,更多相关《高光谱影像特征择选与提取.ppt(57页珍藏版)》请在金锄头文库上搜索。
1、遥感所 张艮中Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Pr
2、ofile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.高光谱影像特点高光谱影像特征选择与提取蚀变信息提取应用示例Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pt
3、y Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Pro
4、file 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created
5、with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.高光谱遥感是20世纪最后二十年,人类在对地观测方面取得的重大技术突破之一,也是当前及以后几十年内的遥感前沿技术。它利用成像光谱仪纳米级的光谱分辨率,获取许多非常窄且光谱连续的图像数据,实现地物
6、空间、辐射、光谱信息的同步获取;为每个像元提供数十至数百个窄波段的光谱信息,并生成一条完整而连续的光谱曲线。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Cr
7、eated with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.高光谱遥感大量的光谱波段数据为人们了解地物提供了丰富的信息,这对于后续进行的地物分类和目标识别是十分有益的然而波段的增多也必然导致信息的冗余和数据处理难度的增加。如何既能有效利用高光谱数据的最大信息,又能较快地处理高光谱数据成为光谱数据处理的研究热点和未来发展方向。其中压缩波段和光谱特征提取与选择的研究是两个重点。Evaluation only.Created with Aspose.Slides for
8、 .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.波段压缩
9、光谱特征Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profi
10、le 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.基于信息量原则(波段选择)基于类别可分性原则(波段选择)基于搜索方法(波段选择)基于数学变换Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-
11、2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.熵和联合熵最佳指数因子(optimal index factor,OIF)自动子空间划分自适应波段选择(adaptive band selection,ABS)波段指数(band index,BI)Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Prof
12、ile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.根据香农信息论原理,一幅8bit表示的图像X的熵为
13、:式中:X为输入图像,Pi为图像像素灰度值为i的概率。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides fo
14、r .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.同理,两个波段联合熵为:n个波段图像的联合熵为:Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004
15、-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.高光谱图像数据波段标准差大,则信息量丰富;而波段间的相关系数小,波段信息冗余度小。根据以上原理,Chavez等人与1982年提出了一种组合波段的选优方法即最佳指数因子。该方法根据下式给出N个波段组合中的最优指数大小:Evaluation only.Created with Aspose.Slides for .NET 3.5 C
16、lient Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.其中:Si为第i个波段的标准差
17、;Ri,j表示第i个波段与第j个波段之间的相关系数。选择的波段数目一般取3,即将所有可能的三个波段组合在一起。OIF越大,则相应组合波段图像的信息量就越大。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pt
18、y Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.在实际应用中,这种方法存在局限性。首先,它选择出来的最优波段未必是最优点;其次,OIF算法对于高光谱图像波段选择而言计算量过大。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Ev
19、aluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.该方法通过定义波段相关系数矩阵及其近邻可传递相关矢量,将高光谱数据空间划分为适合的数据子空间。这种划分方法有着充分的理论依据,反映了数据的局部特性。
20、自动子空间划分就是根据相关系数矩阵灰度图成块的特点,依据高光谱影像相邻波段相关系数的大小,将波段划分为适合的数据子空间。划分好子空间后,再进行波段选择。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty
21、Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.相关矩阵为:进一步地,我们将可传递的相关矢量定义为:我们对该矢量进行处理,从中提取局部相关的极小值。根据这些自动提取的极小值(设在波段开区间(1,N)内这样的极小值共有P-1个),我们将高光谱空间S划分为P个适合的数据子空间(它的维数是Lj(j=1,2,p-1,)Evaluation only.Created with Aspose.Slides for .NE
22、T 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluati
23、on only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.C
24、opyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Sl
25、ides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.该方法针对OIF方法在实际应用中存在的局限性,充分考虑了各波段的空间相关性和谱间相关性。其公式如下:其中:Si为第i个波段的标准差; 和 是第i波段与其前后两波段的相关系数或i波段与任意两个波段的相关系数;是第i幅图像指数的大小。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty
26、 Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.由于基于全局的波段选择算法,选择的波段往往是联虚地集中在某一个连续子空间中。而连续子空间往往相似性大,这就造成信息重复使用,影响后续的
27、处理效果。基于以上考虑,将高光谱数据分为K组,每组波段数分别为n1,n2,nk,定义波段指数为:Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created
28、with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd. 式中( 为第i波段的均方差, 为第i波段与所在组内其他波段相关系数的绝对值之和的平均值,为第i波段与所在组以外地其它波段之间的相关系数的绝对值之和)。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Crea
29、ted with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.均值间的标准距离离散度类间平均可分性B距离Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile
30、5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.“均值间的标准距离”d被定义为:式中,u1、u2分别为两类
31、对应的样本区域的光谱均值; 分别为两类对于的样本区域的方差。d反映两类在每一波段内地可分性大小。d越大,可分性越大。此法是一维特征空间中两类别间可分性的一种度量,它不适合进行多变量的研究。对于多维特征空间、多变量的可分性研究,可用离散度、B距离等方法。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Clie
32、nt Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.式中,Ui、Uj分别表示i、j类的亮度均值矢量, 分别为i、j类的协方差矩阵,trA表示矩阵A对角线元素之和。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.
33、0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.式中符号的意义同于“离散度”公式中的定义。对于任何一给定的地物类别,只
34、要算出这两个不同类别在所有可能的波段组合中的标准距离、离散度或B距离,并去最大者,便是区分这两个类别的最佳波段组合,即最优子集。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluati
35、on only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.上面几种方法是针对两个类别而言,也就是说它们都是类对间的可分型度量。对于对类别而言,一个常用的办法是计算平均可分型的平均值,并按平均值的大小排列所有被评价的子集顺序,从而选择最佳组合波段。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 A
36、spose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.分为最优(Optimal)搜索算法和次优搜索算法。目前实际高光谱图像波段选择都使用次优搜索算法。次优搜索算法是依照准
37、则函数选择一组性能较好的,但不一定是最好的特征集合。传统的方法有有序贯前向选择法(Sequential Forward Selection,SFS)和序贯后向选择法(Sequential Backward Selection ,SBS)Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profi
38、le 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.主要有主成份分析(Principal Component Analysis,PCA)、小波变换(Wavelet Transform,WT)、独立成分(Independent Component Analysis ,ICA、投影寻踪方法等Evaluation only.Create
39、d with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-
40、2011 Aspose Pty Ltd.从操作对象、算法原理、特征性质和应用方式等方面综合考虑,光谱体系包括光谱曲线特征、光谱变换特征和光谱相似度量特征三个层次,分别对应于全部波段的像元光谱曲线分析、部分波段的数据变换与组合、光谱相似性度量三种应用模式。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Cli
41、ent Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.光谱曲线特征光谱运算与变换特征光谱相似性度量特征Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose P
42、ty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.光谱曲线直接编码光谱吸收特征参数光谱吸收指数Evaluation only.Created with Aspose.Slides
43、for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.光
44、谱曲线特征的重要思想是强调曲线形状,直接编码是一种非常直观的方式,最常用的是二值编码(binary coding),即对每一像元,对各波段属性值与某一阀值比较,相应地赋予“0”或“1”的编码。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2
45、004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd. 其中Si为第i波段编码Xi为该波段原始属性值,T为阀值。通常阀值选整个光谱向量的平均值,也可以取向量中值或根据光谱曲线进行人工阀值选择。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004
46、-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd. 二值编码采用单阀值进行处理,划分取决范围较大,表达进度较低。一种改进算法是多值编码。四值编码基本方法
47、是:首先对整个像元光谱向量取平均值,得到阀值T0,将像元属性值分为 、 两个区间,确定两个区间的像元,在分别对两个区间的像元属性值取均值,得到两个新的阀值 、 ,最终形成四个区间间 、 、 、 ,分别用0、1、2、3表示,对每一像元向量,根据其各波段属性值所处区间分别赋以相应的编码,最后进行匹配比较。 Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspos
48、e.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd. 地物光谱曲线反映了地物的吸收和反射特征,对光谱吸收特征参数的提取将成为未来高光谱信息处理研究的主要方向。光谱吸收特征主要由以下特征参数表示,吸收波长波段位置(P)、反射值(R),深度(H)、宽度(W)、斜率(K
49、)、对称度(S)、面积(A)等。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5
50、Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation onl
51、y.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.吸收波段位置(P)是光谱最小值对应的波长,有时也可定义为光谱最大值对应的波长;吸收波段位置处的光谱值即反射值(R);波段深度(H)是由于矿物化学成分在某波长点上吸收光谱特征而比邻接波段有较低的反射率;宽度(W)是指波段深度一半处的宽度;斜率 。其中: 、 分别为吸收终点、吸收始点反射率值; 、 为相应的波长;吸收峰对称度S=A1/A(A1为吸收峰左半端的面积,A为吸收峰的整体面积)。Evaluat
52、ion only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.
53、Copyright 2004-2011 Aspose Pty Ltd.利用这些参数可以对地物进行识别和分类,如对同类地物光谱曲线特征求交得到识别地物的有效特征;对不同类地物光谱曲线特征求交得到区分不同类地物的有效特征,从而达到快速识别和实现地物分类的目的。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Cl
54、ient Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.不同地物光谱曲线,其吸收波峰波谷形状、位置、宽度、深度和对称度等属性也不同。每一个光谱吸收特征可以由光谱吸收谷点M及其两个肩部S1和S2组成,或由吸收峰值点及两个肩部组成。Evaluation only.Created with Aspose.Slides f
55、or .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Ev
56、aluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2
57、.0.0.Copyright 2004-2011 Aspose Pty Ltd.植被指数导数光谱Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created
58、with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.常规多光谱植被指数通常表达为近红外波段与可见光红波段的差值和比值的组合,常用的是比值植被指数(RVI)和标准化植被指数(NDVI)。对于高光谱遥感数据而言,NDVI可以被看作是一个梯级函数,来表达植被反射率在=0.7um处的突然递增。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004
59、-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.植被指数在高光谱应用中非常重要,能够描述植被的精细信息如叶面积指数LAI、植土比、植被组分等,以进行植
60、被指数与生物量预测,在有些分类如神经网络分类器中经常将其作为一个独立特征参加分类。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with As
61、pose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.导数光谱也称光谱微分技术(Spectral Derivative)。采用导数光谱技术可以消除光谱数据之间的系统误差、减弱大气辐射、散射和吸收对目标光谱的影响,以便提取可识别地物的光谱吸收参数(波长位置、深度、宽度和吸收光谱指数等);光谱一阶、二阶和高阶微分可以消除背景噪声,分辨重叠光谱。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.
62、2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.式中:X=X1,X2,,XN为光谱向量,N为波段数,=1, 2
63、,,N为波长集合。以此类推,可以计算出任意阶导数光谱。通过导数光谱运算可以发现待定地物某阶导数具有明显区别于其它地物的特征,从而用于地物识别。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.E
64、valuation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.植被指数与导数光谱实质是在光谱空间上进行特定运算以形成新的特征,按照这一思路,可以设计其它光谱运算特征,如波段求和、取均值以及其它更复杂的运算获取的特征。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty
65、Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.光谱角填图(spectral angle mapping)光谱信息散度 SID(Spectral information diverge
66、nce)Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profi
67、le 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.式中,N为波段数,A=( , , )和B=( , , )分别表示两个光谱向量,其中元素 、 表示像元在第i个波段上的反射率,a为光谱角度。在具体计算中并不需要求出实践角度,采用光谱角余弦作为判据即可。同类像元光谱角余弦较大,接近于1,而不同类像元的光谱角余弦则较少。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluat
68、ion only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.光谱角在衡量像元光谱相似性方面具有明显的优越性,光谱角填图SAM(spectral angle mapping)在高光谱遥感信息分类、聚类都得到了非常广泛的应
69、用。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile
70、 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.P、 Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created w
71、ith Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.此外,在高光谱遥感像元相似度量与特征衡量中,一些常规的指标如相关系数、高维空间距离等也可以作为量度指标。Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides fo
72、r .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.示例数据(Hyperion西藏驱龙)10nm光谱分辨率1-70波段覆盖356-1058nm的可 见光和近红外区域71-242波段覆盖852-2577nm 的短波红外波段像元大小 30m 图像大小 256x6460数据产品有Level0
73、(原始数据) 和Level1两种L1数据产品已经进行了辐射校正Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slide
74、s for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.回归偏度的表示Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose
75、Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.回归偏度分析Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Cl
76、ient Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.
77、Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.高光谱特征选择与特征提取研究 苏红军,杜培军高光谱遥感数据解译的最佳波段选择方法研究 刘建平,赵英时高光谱遥感数据最佳波段选择方法试验研究 刘建平,赵英时,孙淑玲高光谱遥感图像
78、特征选择和提取方法的比较_基于试验区Barrax的HyMap数据 陈桂红,唐伶俐,姜小光高光谱遥感数据特征提取算法与分类研究 苏红军,杜培军,盛业华基于自动子空间划分的高光谱数据特征提取 谷延锋,张晔投影寻踪方法与高光谱遥感图像数据特征提取的研究 刘卓,易东云成像光谱数据的光谱信息特点及最佳波段选择_以北京顺义区为例 姜小光,王长耀成象光谱图象光谱吸收鉴别模型与矿物填图研究 王晋年,郑兰芬,童庆禧高光谱数据的波段序列结构分析与应用研究 张远飞,吴德文,张艮中,朱谷昌,李红高光谱遥感数据光谱特征的提取与应用 杜培军,陈云浩,方涛,陈雍业基于高光谱图像的特征提取/选择及其应用的研究(硕士论文) 谈
79、晓晔高光谱遥感数据最佳波段选择方法研究(硕士论文)杨金红Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides f
80、or .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.谢谢大家!Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.Evaluation only.Created with Aspose.Slides for .NET 3.5 Client Profile 5.2.0.0.Copyright 2004-2011 Aspose Pty Ltd.