GWLS-SVR模型的红枣树叶片叶绿素含量估算

Translated title of the contribution: Chlorophyll Content Estimation of Jujube Leaves Based on GWLS-SVR Model

Nigela Tuerxun, Sulei Naibi, Jian Gao, Jiang Long Shen, Jiang Hua Zheng, Dan Lin Yu

Research output: Contribution to journalArticlepeer-review

1 Scopus citations

Abstract

Chlorophyll Contentis an indicator of the photosynthetic capacity, growth and nutritional status of jujube trees. The distribution of chlorophyll content is different in jujube trees planted in different geographical locations under the influence of natural and human-made factors. The Hyperspectral reflectance of jujube leaves and the SPAD value of jujube leaves that representing chlorophyll content in Ruoqiang county were measured on the spots. To estimate the SPAD value of jujube leaves efficiently and losslessly, the global Moran's I of jujube SPAD value was calculated, The statistics was calculated based on the correlation between SPAD value and Hyperspectral bands to choose the most important characteristic bands. The GWLS-SVR(Geographically Weighted Least Squares-Support Vector Regression)model was used to predict the SPAD value and compared with multiple linear regression (MLR) and support vector regression (SVR) models, and explored the ability of the model to estimate the SPAD value of the jujube leaves. The results show that: (1) the First derivative of the spectrum can effectively remove the noise and highlight the spectral information, especially in the range of 492~510, 542~543, 642~652, 657~670 and 682~692 nm, and significantly improve the correlation of the spectrum with SPAD value. (2) statistics method can effectively select the feature bands of the sensitive range, thus improves the model estimation accuracy. The two variables with the highest importance of the original spectrum were 595 and 696 nm, and the feature band of the first derivative of the spectrum was 688 nm. Among them, the statistics of a single band were always lower than those of multiple band combinations of the same sensitive band interval, which may be caused by the strong collinearity between the adjacent bands. (3) There was significant spatial aggregation on the SPAD value of jujube leaves in Ruoqiang county, the global Moran's I was 0.125 8 (p< 0.1), which is suitable for the establishment of GWLS-SVR model that considers the spatial location. (4) By combining Bootstrap resampling and t-test, the GWLS-SVR model that combined with geographic location information was generally better than the support vector regression and multiple linear regression model, and the results were highly significant (p< 0.001). Among the models, the GWLS-SVR model based on the First derivative of the spectrum was the optimal estimation of SPAD value for jujube leaves (R2=0.975, MSE=1.082), which can provide a certain reference for the Hyperspectral quantitative inversion of the SPAD value of jujube and the rapid and non-destructive monitoring of jujube growth.

Translated title of the contributionChlorophyll Content Estimation of Jujube Leaves Based on GWLS-SVR Model
Original languageChinese (Simplified)
Pages (from-to)1730-1736
Number of pages7
JournalGuang Pu Xue Yu Guang Pu Fen Xi/Spectroscopy and Spectral Analysis
Volume41
Issue number6
DOIs
StatePublished - Jun 2021

Keywords

  • Bootstrap resampling
  • Chlorophyll content
  • GWLS-SVR model
  • Hyperspectral
  • Red jujube tree
  • SPAD value
  • T test

Fingerprint

Dive into the research topics of 'Chlorophyll Content Estimation of Jujube Leaves Based on GWLS-SVR Model'. Together they form a unique fingerprint.

Cite this