中山大学学报(自然科学版) ›› 2020, Vol. 59 ›› Issue (4): 56-63.doi: 10.13471/j.cnki.acta.snus.2019.09.20.2019D037

• • 上一篇    下一篇

博斯腾湖西岸湖滨绿洲土壤含盐量高光谱估算 

赵慧, 李新国, 靳万贵, 麦麦提吐尔逊·艾则孜, 牛芳鹏   

  1. 新疆师范大学地理科学与旅游学院/新疆干旱区湖泊环境与资源实验室,新疆 乌鲁木齐 830054
  • 收稿日期:2019-09-20 出版日期:2020-07-25 发布日期:2020-07-25
  • 通讯作者: 李新国(1971年生),男;研究方向:干旱区土壤资源变化及其遥感应用;E-mail:onlinelxg@sina.com
  • 作者简介:赵慧(1994年生),女;研究方向:干旱区土壤资源变化及其遥感应用;E-mail:ziyuhe528@163.com

Hyperspectral estimation of soil salt content in lake oasis on the west bank of Bosten lake 

ZHAO Hui, LI Xinguo, JIN Wangui, Mamattursun·Eziz, NIU Fangpeng   

  1. College of Geographic Sciences and TourismXinjiang Normal University / Xinjiang Laboratory of Lake Environment and Resources in Arid ZoneUrumqi 300548China
  • Received:2019-09-20 Online:2020-07-25 Published:2020-07-25

摘要: 以博斯腾湖西岸湖滨绿洲土壤盐分及其对应的高光谱数据为研究对象,对原始光谱反射率(R) 进行均方根(R )、对数(lgR)、倒数(1/R)、对数倒数(1/lgR) 变换,引入分数阶微分对变换的光谱反射率进行0~2内的微分预处理,通过显著性检验优选特征波段,并利用特征波段进行偏最小二乘回归的建模和验证。结果表明:1) 通过显著性检验较多的特征波段是:1/R 为691个波段,比通过原始波段多226个波段;且随着阶数的增加,在各阶数中通过显著性检验的波段数量呈先增加后减少的趋势;通过数量较多的特征波段是1/R 的0. 4阶和0. 6阶,分别为145、150;2) 原始光谱反射率的相关性极大值绝对值是:2阶为0. 53,其他4种数学变换的相关性极大值绝对值比R高0. 11~0. 16,特征波段主要集中在600~1 000 nm和2 020~2 330 nm;3) 在原始光谱反射率及其变换值的各分数阶微分利用偏最小二乘回归建模中,1/lgR 的0. 2阶建立的模型为最佳模型,R2C=0. 78,RMSEC=1. 56,R2V=0. 63,RMSEV=1. 44。

关键词: 分数阶微分, 土壤含盐量, 光谱变换, 偏最小二乘回归, 高光谱估算

Abstract: Taking the soil salinity and corresponding hyperspectral data of the lakeside oasis on the west bank of the Bosten Lake as the research objectthe root mean square Rlogarithmic lgRrecipro⁃cal 1/Rreciprocal 1/lgR were used to transform the original spectral reflectance. The fractional differentiation was introduced to perform differential preprocessing within 0-2 on the transformed spectral reflectance. The characteristic band through significance test was used to model and verify partial least squares regression. The results show that: 1)The characteristic bands that passed the significance test were mostly 1/R of 691 bandswhich was 226 more bands than the original band R. With increasing orders,the number of bands that passed the significance test in each order showed a trend of increasing first and then decreasing. The characteristic bands that passed more were 0.4 and 0.6 orders of 1/Rwhich were 145 and 150respectively. 2)The absolute value of the correlation maximum value of the original spectrum R was 2nd order 0.53. The absolute value of the correlation maximum value of the other four mathematical transformations was 0.11-0.16higher than R.The characteristic bands were mainly concentrated at 600-1 000 nm and 2 020-2 330 nm. 3)The fractional differential of the original spectral Rthe root mean square R,the logarithm lgRthe reciprocal 1/Rand the logarithm reciprocal 1/lgR was modeled by partial least squares regression. The model established at 0.2th order of 1/lgR was the best onewith R2C=0.78RMSEC=1.56R2V=0.63RMSEV=1.44

Key words: fractional differential,  , soil salinity,  , spectral transformation,  , partial least squares regression methods,  , hyperspectral estimation

中图分类号: