Quantification winter wheat LAI with HJ-1CCD image features over multiple growing seasons
Quantification winter wheat LAI with HJ-1CCD image features over multiple growing seasons
复制标题
使用 HJ-1CCD 多个生长季节的图像特征量化冬小麦 LAI
DOI：
10.1016/j.jag.2015.08.004
复制
发表时间：
2016-02-01
影响因子：
7.5
通讯作者：
Yang, Wenzhi
中科院分区：
文献类型：
--
作者：
Li, Xinchuan;Zhang, Youjing;Yang, Wenzhi
遥感影像被广泛用于景观叶面积指数（LAI）的连续绘制。本研究旨在探索中国HJ-1 A/B CCD影像中用于估算北京地区冬小麦LAI的理想影像特征。从这些影像中提取冬小麦生长四季的影像特征，包括5个植被指数（VIs）、主成分（PC）、流苏帽变换（TCT）和纹理参数。LAI与近红外反射率波段、5个VIs[归一化差异植被指数、增强植被指数（EVI）、修正非线性植被指数（MNLI）、土壤调整优化植被指数和比值植被指数]、第一主成分（PC1）和第二TCT成分（TCT2）呈显著相关。然而，这些图像特征与8种纹理测度结合，并不能显著提高冬小麦LAI的估计精度。为了确定具有最佳估计精度的少数理想特征，应用偏最小二乘回归（PLSR）和投影变量重要性（VIP）预测LAI值。基于VIP值选择4个遥感特征（TCT2、PC1、MNLI和EVI）。留一交叉验证结果表明，在整个生长季节，基于这4个特征的PLSR模型比基于10个特征的PLSR模型效果更好。本研究的结果表明，选择几个理想的图像特征就足以进行LAI估计。(C) 2015 Elsevier B.V.版权所有
Remote sensing images are widely used to map leaf area index (LAI) continuously over landscape. The objective of this study is to explore the ideal image features from Chinese HJ-1 A/B CCD images for estimating winter wheat LAI in Beijing. Image features were extracted from such images over four seasons of winter wheat growth, including five vegetation indices (VIs), principal components (PC), tasseled cap transformations (TCT) and texture parameters. The LAI was significantly correlated with the near-infrared reflectance band, five VIs [normalized difference vegetation index, enhanced vegetation index (EVI), modified nonlinear vegetation index (MNLI), optimization of soil-adjusted vegetation index, and ratio vegetation index], the first principal component (PC1) and the second TCT component (TCT2). However, these image features cannot significantly improve the estimation accuracy of winter wheat LAI in conjunction with eight texture measures. To determine the few ideal features with the best estimation accuracy, partial least squares regression (PLSR) and variable importance in projection (VIP) were applied to predict LAI values. Four remote sensing features (TCT2, PC1, MNLI and EVI) were chosen based on VIP values. The result of leave-one-out cross-validation demonstrated that the PLSR model based on these four features produced better result than the ten features' model, throughout the whole growing season. The results of this study suggest that selecting a few ideal image features is sufficient for LAI estimation. (C) 2015 Elsevier B.V. All rights reserved,