一种基于谱域插值的强鲁棒性特征光谱构建算法  

A Robust Characteristic Spectrum Construction Algorithm Based on Spectral Domain Interpolation

作　　者：栗旭升 汪大明[1,2,3] 汪翡翠 仝云霄 曹思琦 LI Xu-sheng;WANG Da-ming;WANG Fei-cui;TONG Yun-xiao;CAO Si-qi(Tianjin Center,China Geological Survey,Tianjin 300170,China;North China Center for Geoscience Innovation,China Geological Survey,Tianjin 300170,China;Tianjin Key Laboratory of Coast Geological Processes and Environmental Safety,Tianjin 300170,China)

机构地区：[1]中国地质调查局天津地质调查中心,天津300170 [2]中国地质调查局华北地质科技创新中心,天津300170 [3]天津市海岸带地质过程与环境安全重点实验室,天津300170

出　　处：《光谱学与光谱分析》2025年第4期1168-1174,共7页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2022YFB3902000);国家自然科学基金面上项目(42272346)资助。

摘　　要：传统的特征光谱构建算法中,常用若干条地物光谱的算术平均值来表征地物特征光谱。但均值的表征能力易受地物内部差异程度影响,强化极值信息,弱化部分特征信息。针对上述问题,基于地理学近相似定律,参考空间插值的思想,提出一种谱域插值的特征光谱提取算法。首先,基于多条地物光谱,计算各波长上地物最大和最小反射率,即地物的谱域。然后,以单一地物光谱为中心,谱域为范围,进行归一化反距离插值,获得多个单特征谱域空间。最后,将多个谱域空间进行相加得到地物累计谱域空间,逐波长求取累计谱域空间中的最大值作为该波长地物反射率构成地物特征光谱。为验证谱域插值特征提取算法在特征光谱形态和幅值构建上的有效性和优越性,以航空高光谱遥感影像和ASD实测树种光谱为数据源,分别求取均值特征光谱(MCS)和谱域插值特征光谱(ICS)。基于两种特征光谱分别进行航空高光谱数据的光谱角制图(SAM)和ASD实测数据的特征参量提取、重要性评价和线性判别分析(LDA),以探究ICS在整体形状上的表征能力和细节特征上的再现能力,验证其有效性与优越性。实验结果显示,相较于MCS,ICS在表征特征光谱形态的SAM中,总体精度提升4.24%;在表征特征光谱细节的幅值特征参量重要性评价和LDA中,幅值参量重要性得分平均提高0.35,各树种判别精度提升2.51%,总体精度提升2.5%。研究表明,ICS无论是在光谱特征整体形态的表征,还是对细节特征的再现上,都要优于传统MCS。可以用于改进分类场景中目标地物的特征光谱提取流程,提高类间可分离性;优化反演场景中特征参量的构建,提升光谱的表征能力。In the traditional characteristic spectrum extraction algorithm,the arithmetic mean value of spectra is often used to indicate the characteristic spectrum.However,by strengthening the extreme value information and weakening some characteristic information,the indication capability of the mean value is easily affected by the degree of internal differences between objects,Based on the first theorem of geography and the idea of spatial interpolation,a characteristic spectrum extraction algorithm of spectral domain interpolation is proposed.First,the spectral domain of the objects,maximum and minimum reflectances of the objects at each wavelength,are calculated on several object spectra.To obtain single-feature spectral domain spaces,normalized inverse distance interpolation is performed at the center of a single object spectrum with the range of spectral domain.Finally,as multiple spectral domains are added,the cumulative spectral domain space of ground objects is obtained,and the maximum value in the cumulative spectral domain space,which is calculated by wavelength,is taken as the reflectivity,forming the characteristic spectrum of ground objects.To verify the validity and superiority of the spectral domain interpolation characteristic extraction algorithm s performance on the construction of characteristic spectral shape and amplitude,tree species spectra measured from aerial hyperspectral remote sensing images and ASD are used as data sources to calculate the mean characteristic spectrum(MCS)and spectral domain interpolation characteristic spectrum(ICS).To explore the ICS s ability to characterize the overall shape and reproduce detail features,spectral angle mapping(SAM)of aerial hyperspectral data,feature parameter extraction importance evaluation,and linear discriminant analysis(LDA)of ASD-measured data were performed.The experimental results show that ICS improves the overall accuracy by 4.24%in the SAM when indicating characteristic spectral morphology compared with MCS when it comes to the amplitude featur

关 键 词：高光谱 特征光谱 谱域插值 特征提取 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]