摘要:
在高光谱数据分类应用中,地物光谱特征分析是对地物进行分类和检索的基础性工作。选取禾本科斑竹、草本科蕨类、荨麻科冷水花、杉科杉木和棕榈科棕榈树等5种岷江上游亚高山森林植被进行实地光谱测量,建立高光谱相似性度量参量,如欧式距离( Euclidean distance , ED )、光谱角度( spectral angle mapper , SAM )、光谱信息散度(spectral information divergence ,SID)、SID和SAM混合SID(TAN)以及基于道格拉斯-普克算法的光谱降维距离(spectral distance based on Douglas -Peucker,SDDP)度量算法,定量分析对亚高山森林植被的识别能力。研究结果表明:5种亚高山森林植被光谱特征的差异主要表现在光谱曲线反射波峰和波谷位置; ED对冷水花的相对光谱识别概率最高;SID和SID( TAN)对斑竹与蕨类的识别概率最高;SDDP对杉木的识别概率最高;SAM,SDDP,ED, SID(TAN)和SID这5种光谱相似性测度算法对亚高山森林植被的相对光谱识别熵分别是1.51,1.59,1.61,2.16和2.18,说明光谱角度制图具有较高的识别能力;而道格拉斯-普克光谱检索算法是在提取光谱曲线特征向量的基础上进行相似性测度,其降低了光谱检索的时间频率,在保证相近识别能力的条件下,能够大大提高程序的检索效率,是一种快速有效的高光谱特征匹配和检索算子。
Abstract:
Spectral characteristics analysis is the basis of spectral feature classification and matching in hyperspectral image processing .In this paper , the authors selected five kinds of subalpine forest vegetation to measure their field spectra in the upper reaches of the Minjiang River , which include gramineae mottled bamboo , herbaceous fern ,pilea notate , arbor china fir and shrubs palm .Through constructing the high spectral similarity measure index, five measuring methods, i.e., Euclidean distance(ED), spectral angle mapper(SAM), spectral information divergence(SID), spectral information divergence -spectral angle mapper(SID(TAN))and spectral distance based on Douglas -Peucker ( SDDP ) , were used to analyze the relative capability for recognizing forest vegetation on the plateau .According to the results obtained , the spectral feature difference in the five kinds of forest vegetation mainly lies in peaks and troughs in the spectral curves; pilea notate has the highest relative spectral discriminatory probability in ED similarity measurement;mottled bamboo and fern have the highest relative spectral discriminatory probability in SID and SID ( TAN);China fir has the highest relative spectral discriminatory probability in SDDP.SAM, SDDP, ED, SID(TAN)and SID of the relative spectral discriminatory entropy are 1. 51, 1.59, 1.61, 2.16 and 2.18 respectively.The research results showed that the means reduced the amount of calculation for doing the similarity measurement which extracted the spectral feature vectors with the SFT , DPBSR and DABSR, DPSR.In order to ensure the condition of similar recognition capability , the means can greatly improve the retrieval efficiency of the program , and hence they are the fast and efficient hyperspectral feature matching and retrieval methods .
下载: