虫害胁迫下毛竹叶绿素含量高光谱估算方法

引用本文

李凯, 陈芸芝, 许章华, 黄旭影, 胡新宇, 汪小钦. 虫害胁迫下毛竹叶绿素含量高光谱估算方法[J]. 光谱学与光谱分析, 2020,40(8): 2578-2583.
LI Kai, CHEN Yun-zhi, XU Zhang-hua, HUANG Xu-ying, HU Xin-yu, WANG Xiao-qin. Hyperspectral Estimation Method of Chlorophyll Content in MOSO Bamboo under Pests Stress[J]. Spectroscopy and Spectral Analysis, 2020,40(8): 2578-2583.
Doi:10.3964/j.issn.1000-0593(2020)08-2578-06 复制到剪切板

Permissions

《光谱学与光谱分析》期刊社所有

虫害胁迫下毛竹叶绿素含量高光谱估算方法

李凯^1,², 陈芸芝^1,^2,^*, 许章华^1,^2,³, 黄旭影⁴, 胡新宇³, 汪小钦^1,²

1. 福州大学卫星空间信息技术综合应用国家地方联合工程研究中心, 福建福州 350116

2. 空间数据挖掘与信息共享教育部重点实验室, 福建福州 350116

3. 福州大学环境与资源学院, 福建福州 350116

4. 南京大学国际地球系统科学研究所, 江苏南京 210093

*通讯联系人 e-mail: chenyunzhi@fzu.edu.cn

作者简介: 李凯, 1992年生, 福州大学数字中国研究院(福建)硕士研究生 e-mail: 709331379@qq.com

收稿日期: 2019-07-17 接受日期: 2019-11-30

基金: 国家重点研发计划课题(2017YFB0504203), 中央引导地方科技发展专项(2017L3012), 中国博士后基金面上项目(2018M630728), 3S技术与资源优化利用福建省高校重点实验室开放课题(fafugeo201901)资助

摘要

叶绿素作为参与植被光合作用最重要的色素, 是监测毛竹虫害的一项重要指标。通过对不同光谱数据集进行波长筛选, 建立虫害胁迫下竹叶叶绿素含量的高光谱估算模型, 为利用高光谱遥感监测毛竹虫害提供理论依据。试验在福建省毛竹生产基地顺昌县进行, 使用ASD FieldSpec 3光谱仪采集不同虫害程度竹叶光谱102条, 并利用SPAD-502叶绿素计测定相应叶片叶绿素含量。通过对比不同虫害程度竹叶的光谱特征, 探测利用高光谱数据估算叶绿素含量的机理。对竹叶原始光谱(OS)进行包络线去除(CR)、一阶导数(FD)、包络线去除一阶导数(CR-FD)变换, 分析不同光谱数据与叶绿素含量的相关性, 并利用连续投影算法(SPA)分别提取4种光谱的特征波长。采用基于x-y距离结合的样本划分法(SPXY)和随机法对4种光谱数据集进行划分, 结合多元逐步回归(MSR)建立竹叶叶绿素含量估算模型, 分析光谱变换及样本划分对估算叶绿素含量的影响。结果表明, 不同虫害程度竹叶光谱反射率差异明显, 主要表现为可见光波段范围内的“绿峰”和“红谷”的逐渐消失, “红边”斜率减小, 近红外波长反射率降低。通过光谱变换可有效提升光谱与叶绿素含量的相关性, 其中CR-FD光谱与叶绿素含量在724 nm处的相关系数最大。经连续投影算法提取的不同光谱数据集的特征波长集中分布在绿光、红光、 “红边”位置, 多个被选择波长位于与叶绿素含量相关性较高的波长区(600~750 nm)。基于SPXY样本划分法建立的MSR模型相比于随机样本划分法能显著提升叶绿素含量的估算精度, 其中 R²和RPD平均提高0.1和0.5, RMSE平均降低0.7。以CR-FD光谱特征波长结合SPXY样本划分法建立的多元逐步回归模型对竹叶叶绿素含量的估算精度最高, R², RMSE和RPD分别为0.835, 2.604和2.364, 可对虫害胁迫下毛竹叶片叶绿素含量进行准确的估算。

关键词: 叶绿素含量; 高光谱; 光谱变换; 波长筛选; SPXY样本划分

中图分类号:TP79 文献标志码:A

Hyperspectral Estimation Method of Chlorophyll Content in MOSO Bamboo under Pests Stress

LI Kai^1,², CHEN Yun-zhi^1,^2,^*, XU Zhang-hua^1,^2,³, HUANG Xu-ying⁴, HU Xin-yu³, WANG Xiao-qin^1,²

1. National & Local Joint Engineering Research Center of Satellite Geospatial Information Technology, Fuzhou University, Fuzhou 350116, China

2. Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education, Fuzhou 350116, China

3. College of Environment and Resources, Fuzhou University, Fuzhou 350116, China

4. International Institute for Earth System Science, Nanjing University, Nanjing 210093, China

*Corresponding author

Abstract

As the most important pigment involved in photosynthesis of plant, the chlorophyll is an important indicator for monitoring bamboo pests. This study aims to establish the hyperspectral estimation model for the chlorophyll content of bamboo leaves under pests stress by wavelength screening of different spectral data sets, and provide a theoretical basis for monitoring the pests of bamboo by hyperspectral remote sensing. The test was carried out in Shunchang County, the bamboo production base in Fujian Province. The ASD FieldSpec 3 spectrometer was used to collect 102 bamboo leaves spectra of different pest levels, and the chlorophyll content of the corresponding leaves was determined by SPAD-502 chlorophyll meter. By comparing the spectral characteristics of bamboo leaves with different pest levels, the mechanism of estimating chlorophyll content with hyperspectral data was explored. The original spectrum (OS) of the bamboo leaves was subjected to continuum removal (CR), first derivative (FD), and continuum removal-first derivative (CR-FD), and the correlation between different spectral data and chlorophyll content was analyzed. The characteristic wavelengths of the four spectra were extracted by the successive projection algorithm (SPA). Four spectral datasets were divided by sample set partitioning based on joint x-y distances method (SPXY) and random method. Combined with multiple stepwise regression (MSR), the chlorophyll content estimation model of bamboo leaves was established, and the effects of spectral transformation and sample partitioning on estimating chlorophyll content were analyzed. The results showed that there were significant differences in the spectral reflectance of bamboo leaves with different pest levels. The main manifestations were the gradual disappearance of the “green peak” and “red valley” in the visible light range, the “red edge” was levelled and the near-infrared wavelength reflectance was reduced. The spectral transformation could effectively improve the correlation between the spectrum and chlorophyll content, and the correlation coefficient between the CR-FD spectrum and chlorophyll content at 724 nm was the largest. The characteristic wavelengths of different spectral data sets extracted by the successive projection algorithm were concentrated in the green band, red band, and “red edge”, and the multiple selected wavelengths were located in bands (600~750 nm) that highly correlated with chlorophyll content. The MSR model based on SPXY sample partitioning method could significantly improve the estimation accuracy of chlorophyll content compared with the random sample partitioning method, in which R² and RPD increased by 0.1 and 0.5, and RMSE decreased by 0.7 on average. The multiple stepwise regression model established by CR-FD spectrum characteristic wavelengths combined with SPXY sample partitioning method had the highest accuracy for estimating chlorophyll content of bamboo leaves, and the R², RMSE, RPD were 0.835, 2.604 and 2.364 respectively, which could accurately estimate the chlorophyll content of bamboo leaves under pests stress.

Keyword: Chlorophyll content; Hyperspectral; Spectral transformation; Characteristic wavelengths; SPXY algorithm

文章图片

引言

毛竹作为我国竹林产业的重要组成部分, 在我国林业的经济发展中起到关键作用。刚竹毒蛾是毛竹的主要虫害, 严重威胁毛竹的生长, 制约了竹林产业的可持续发展^[1]。叶绿素是绿色植被吸收光能的物质, 作为参与植被光合作用最重要的色素, 直接影响到绿色植被对光能的利用^[2]。当植被受虫害胁迫而“ 失绿” 时, 常伴随着内部叶绿素含量的变化, 并在光谱特性上发生变化^[3]。因此, 植被叶绿素含量的状态已成为监测植被虫害的一项重要指标。

高光谱数据能够提供丰富的光谱信息, 可以更加精确地检测植被色素光谱特征, 估计植被的生化参数进而诊断植被的健康状态。实际应用中, 高光谱数据在提供丰富信息的同时, 也存在信息冗余量大的问题, 导致统计参数的估计偏差增大^[3]。特征波长的提取可有效地对高光谱数据进行降维, 从而达到优化模型的效果。通过对高光谱数据进行光谱变换能够增强特征波长信息, 进而提高模型的估算精度^[4]。国内外学者通过利用不同光谱的特征光谱位置、多波段的光谱指数、特征光谱参数等光谱信息开展了大量的植被生化参数研究。如王雪等^[5]利用竞争性自适应重加权采样算法进行波长筛选, 并结合SPXY样本划分法建立玉米含水率的偏最小二乘估算模型。宫兆宁等^[6]系统分析了典型湿地植被的多项光谱特征参数, 确定了“ 红边” 位置、 565和735 nm归一化光谱指数为估算叶绿素含量的最佳光谱特征参数。刘文雅等^[7]利用相关系数法筛选出与马尾松叶绿素含量高度相关的植被指数, 并对马尾松冠层叶绿素含量进行估算。毛博慧等^[8]采取遗传算法进行波长筛选, 利用优选波长组合植被指数, 建立基于敏感植被指数的冬小麦叶绿素含量估算模型。

为提高竹叶叶绿素含量估算精度, 通过对不同虫害程度的毛竹叶片光谱和叶绿素含量进行测定, 采用连续投影算法(successive projections algorithm, SPA)对光谱数据进行压缩, 提取特征波长, 对比不同样本集划分方法对叶绿素含量建模精度的影响, 建立虫害胁迫下竹叶叶绿素含量的多元逐步回归(multiple stepwise regression, MSR)模型, 以期为今后毛竹虫害监测提供理论依据。

1 实验部分

试验于2019年4月在福建省顺昌县内毛竹林进行; 该县位于福建省西北部, 季风气候明显, 全年雨量充沛, 森林资源丰富, 是福建省的毛竹生产基地县。根据刚竹毒蛾虫害发生的特点, 随机选取6个不同样方, 参照《植物病理学》的虫害划分标准, 将毛竹样本分为健康、轻度虫害、中度虫害及重度虫害4个等级^[9], 对采集的不同虫害等级毛竹叶片进行光谱及叶绿素含量的现场测定。

1.1 数据采集

使用美国ASD FieldSpec 3便携式光谱仪进行光谱数据测量, 光谱仪波长范围为350~2 500 nm, 其中350~1 000 nm波长的采样间隔为1.4 nm, 1 000~2 500 nm波长的采用间隔为2 nm, 共2 151个波段。选择不同虫害程度的叶片进行测定, 测定时间为10:00— 15:00, 探头距叶片15 cm处垂直向下测量, 共采集102个叶片样本, 每个样本测得10条光谱, 取平均值作为该样本的光谱, 每组数据采样前均进行标准白板校正。为保证数据的精度, 每个样本光谱测量后采用日本SPAD-502叶绿素计测定叶片相对叶绿素含量(soil and plant analyzer development, SPAD)。

1.2 光谱数据处理

利用ASD FieldSpec 3光谱仪配套软件ViewSpecPro对采集的光谱数据进行处理, 计算叶片光谱反射率, 保留400~900 nm范围内的光谱数据, 并通过Savitzky-Golay滤波算法对光谱数据进行平滑去噪处理。对原始光谱(original spectrum, OS)进行一阶导数(first derivative, FD)处理, 可消除与波长无关的漂移^[10]。包络线去除(continuum removal, CR)是将原始光谱曲线与包络线曲线进行比值运算后得到, 该方法能够增强各生化参量敏感波段的光谱吸收特征^[11], 并计算包络线去除后光谱的一阶导数(continuum removal-first derivative, CR-FD)。

1.3 连续投影算法

连续投影算法是一种使矢量空间共线性最小化的前向变量选择方法, 通过对光谱向量进行投影操作, 能够找寻到最具代表性的变量组, 获取具有最小共线性的波长子集。该算法的步骤如下^[12]:

(1)在光谱矩阵中任选一条光谱列向量作为起始向量;

(2)分别计算剩余列向量在起始向量的正交平面上的投影向量;

(3)挑选出最大投影作为下一次投影的起始向量, 直到挑选变量个数达到所需个数;

(4)将提取的所有波长组合进行多元线性回归, 最小的RMSE所对应的组合即为最优波长组合。

1.4 SPXY算法

样本数据集的划分会在一定程度上影响模型估算的精度, SPXY(sample set partitioning based on joint x-y distance)法^[13]是一种同时考虑样本光谱以及生化参量欧氏距离的样本集选择方法, 使选择的样本集更具有代表性, 从而改善所建模型的估算能力。计算公式如式(1)和式(2)

$d_{x} (p, q) = \sqrt[]{\overset{J}{\sum_{j = 1}} [x_{p} (j) - x_{q} {(j)]}^{2}}$ (1)

$d_{xy} (p, q) = \frac{d_{x} (p, q)}{\max_{p, q \in [1, N]} d_{x} (p, q)} + \frac{d_{y} (p, q)}{\max_{p, q \in [1, N]} d_{y} (p, q)}$ (2)

式中, J为光谱中的波长数, x_p(j)和x_q(j)分别为样本p和q的光谱值, d_x(p, q)和d_y(p, q)分别表示样本p和q的x变量和y变量的欧式距离。该算法的步骤如下:

(1)计算样本间的两两距离, 选择距离最大的两个样本作为起始样本;

(2)分别计算剩余样本与已选样本间的距离, 对于每个剩余样本而言, 选择与起始样本间的最短距离;

(3)在所选择的最短距离中选择最长距离所对应的样本, 作为第三个样本, 以此逐步选择, 直至所选样本数达到目标个数。

1.5 模型精度检验

采用决定系数(R²), 均方根误差(RMSE)和相对分析误差(RPD)3个指标对叶绿素含量估算模型进行精度评价。其中, R²越大模型估算效果越好; RMSE越小模型估算效果越好; 当RPD> 3时, 表明模型的估算效果极好, 当RPD为3~2.25时, 表示模型的估算效果较好, 当RPD为2.25~1.75时, 模型的估算效果一般, 当RPD< 1.75时, 模型的估算效果较差^[14]。各指标计算公式分别如式(3)— 式(5)

$R^{2} = \frac{SSR}{SST} = \frac{\overset{n}{\sum_{i = 1}} ({\dot{y}}_{i} {- \overset{̅}{y})}^{2}}{\overset{n}{\sum_{i = 1}} (y_{i} {- \overset{̅}{y})}^{2}}$ (3)

$RMSE = \sqrt[]{\frac{1}{n} \overset{n}{\sum_{i = 1}} (y_{i} - {\dot{y}}_{i})^{2}}$ (4)

$RPD = \frac{SD}{RMSE}$ (5)

式中, $\overset{̅}{y}$ 为实测值y_i的平均值, ${\dot{y}}_{i}$ 为样本i的估算值, SD代表样本实测值的标准差。

2 结果与讨论

2.1 虫害毛竹叶片光谱特征

图1为不同虫害程度竹叶光谱曲线, 可以看出, 健康竹叶在520~570 nm波长范围因叶绿素吸收较少而形成了一个绿色波段的小反射峰, 称之为“ 绿峰” 。在640~700 nm波长范围因叶绿素的强吸收形成了一个红色波段吸收谷, 称之为“ 红谷” 。由于叶片的多孔薄壁细胞结构特性对近红外光强烈的反射, 形成光谱上的最高峰。当毛竹遭受虫害侵蚀后光谱发生明显变化, 主要表现为可见光波段范围内的“ 绿峰” 和“ 红谷” 的逐渐消失, “ 红边” 斜率减小, 近红外波长反射率降低, 随着虫害程度加重, 这一现象更加明显。导致这一现象发生的主要原因是毛竹遭受虫害侵蚀后, 竹叶叶绿素含量相对减少, 细胞结构遭到破坏。因此, 根据光谱间的差异, 进行竹叶叶绿素含量的估算是可行的。

	Figure Option View Download New Window
	图1 不同虫害程度的竹叶光谱曲线Fig.1 Spectra of bamboo leaves under pests stress

2.2 叶绿素含量与光谱数据之间的相关性分析

利用Pearson相关分析法计算不同光谱与竹叶叶绿素含量的相关性, 如图2所示。从图2可以看出, 原始光谱(original spectrum, OS)在可见光波段与叶绿素含量呈负相关, 在波长644 nm处相关性最高, 相关系数为-0.759, 在近红外波长基本不相关; 包络线去除(continuum removal, CR)与叶绿素含量的相关性曲线与原始光谱相似, 相关系数有明显提升, 其中在波长702 nm处相关性最高, 相关系数为-0.852; 一阶导数(first derivative, FD)与叶绿素含量的相关系数呈现正负起伏变化, 在波长629 nm处相关性最高, 相关系数为-0.776; 包络线去除一阶导数(continuum removal-first derivative, CR-FD)相关性曲线与一阶导数相似, 在波长724 nm处的相关系数高达0.862。可见, 竹叶叶绿素含量与不同光谱相关性较高的波长均处于600~750 nm, 通过光谱变换可以有效提升与叶绿素含量的相关性。

	Figure Option View Download New Window
	图2 竹叶叶绿素含量与光谱的相关系数Fig.2 Correlation coefficient between chlorophyll content and spectra of bamboo leaves

2.3 特征波长提取

使用连续投影算法对不同光谱数据集进行特征波长选择, 结果如表1所示。 OS所选波长个数为7个, 集中分布在“ 绿峰” 、 “ 红谷” 、 “ 红边” 位置; CR有5个波长被选择, 所选波长主要分布在“ 红边” 位置; FD有4个波长被选择, 所选波长位置较为分散; CR-FD有5个波长被选择, 所选波长大多分布在“ 红边” 位置。从各数据集波长选择结果可以看出, 不同光谱数据集选择的波长位置及个数有所差异, 但所选波长位置基本上集中在绿光、红光、 “ 红边” 3个区域, 各光谱数据集都有多个与叶绿素含量相关性较高的波长(600~750 nm)被选择。经包络线去除后的近红外波长光谱值较为接近, 且与可见光波长的光谱值差异变大, 从而导致在近红外范围有1个波长被选择。由于算法本身的原因导致易在波长范围边缘处选取波长^[15]。

表1 SPA波长选择结果 Table 1 SPA wavelengths selection result

2.4 模型的建立和预测

共采集不同虫害程度竹叶样本102个, 利用SPA算法对各光谱数据集特征波长进行提取后, 分别利用SPXY法和随机法进行样本集的划分, 建模集与验证集的划分比例为7:3, 然后建立基于样本划分及特征波长的多元逐步回归模型, 并计算每种模型在验证集中的叶绿素含量估算精度, 结果如表2所示。从表2可以看出, 利用SPXY法对样本进行划分后的模型对叶绿素含量的估算精度明显优于基于随机法样本划分的模型估算精度。基于SPXY样本划分法的OS, FD, CR以及CR-FD模型的R²均高于0.8, 其中OS和CR-FD模型R²高达0.840和0.835; OS和CR-FD模型的RMSE分别为2.925和2.604; CR和CR-FD模型的RPD值分别达到2.294和2.364, 表明这2种模型具有较好的预测能力。图3为基于SPXY样本划分法的竹叶叶绿素含量预测结果, CR和CR-FD模型的叶绿素含量预测效果较好。综合考虑, 选CR-FD模型为竹叶叶绿素含量的最优估算模型。表3为各模型的回归方程式。

表2 竹叶叶绿素含量估算模型参数 Table 2 Estimation model parameters of chlorophyll content in bamboo leaves

	Figure Option View Download New Window
	图3 叶绿素含量预测值与实测值比较Fig.3 Comparison of predicted and measured chlorophyll contents

表3 竹叶叶绿素含量多元逐步回归模型 Table 3 Multiple Stepwise Regression model of chlorophyll content in bamboo leaves

3 结论

(1)虫害胁迫下竹叶光谱特征发生明显变化, 其可见光波段范围内的“ 绿峰” 和“ 红谷” 逐渐消失, “ 红边” 斜率减小, 近红外波长反射率降低。

(2)对刚竹毒蛾虫害胁迫下竹叶原始光谱进行光谱变换可以有效提升与叶绿素含量的相关性, 可见光区域是估算竹叶叶绿素含量的敏感区域, 包络线去除一阶导数光谱在724 nm处与竹叶叶绿素含量的相关系数最高, 达到0.826。

(3)采用连续投影算法提取的特征波长主要分布于绿光、红光、 “ 红边” 位置, 且每种光谱数据集都有多个与叶绿素含量相关性较高的波长(600~750 nm)被选择。

(4)基于SPXY样本划分法对建模样本进行选择, 建立的叶绿素含量估算模型的R², RMSE, RPD相比于随机样本划分法都有明显的提升, 其中R²和RPD平均提高0.1和0.5, RMSE平均降低0.7。

(5)多元逐步回归可对连续投影算法提取的特征波长进一步压缩, 结合连续投影算法与SPXY样本划分法建立的包络线去除和包络线去除一阶导数多元逐步回归模型可有效地对竹叶叶绿素含量进行估算, 其中包络线去除一阶导数模型的估算精度最高, 为刚竹毒蛾虫害胁迫下竹叶叶绿素含量的最优估算模型。

参考文献

文献列表

[1]	HUANG Xu-ying, XU Zhang-hua, LIN Lu, et al(黄旭影, 许章华, 林璐, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2018, 38(12): 183. [本文引用:1]
[2]	FENG Peng, SUN Li, SEHN Xiao-hui, et al(冯鹏, 孙力, 申晓慧, 等). Chinese Agricultural Science Bulletin(中国农学通报), 2018, 34(23): 122. [本文引用:1]
[3]	TONG Qing-xi, ZHANG Bing, ZHENG Lan-fen(童庆禧, 张兵, 郑兰芬). Hyperspectral Remote Sensing: Principles, Techniques and Applications(高光谱遥感: 原理、技术与应用). Beijing: Higher Education Press(北京: 高等教育出版社), 2006. [本文引用:2]
[4]	ZHANG Dong-hui, ZHAO Ying-jun, QIN Kai, et al(张东辉, 赵英俊, 秦凯, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2018, 34(20): 149. [本文引用:1]
[5]	WANG Xue, MA Tie-min, YANG Tao, et al(王雪, 马铁民, 杨涛, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2018, 34(13): 203. [本文引用:1]
[6]	GONG Zhao-ning, ZHAO Ya-li, ZHAO Wen-ji, et al(宫兆宁, 赵雅莉, 赵文吉, 等). Acta Ecologica Sinica(生态学报), 2014, 34(20): 5736. [本文引用:1]
[7]	LIU Wen-ya, PAN Jie(刘文雅, 潘洁). Chinese Journal of Applied Ecology(应用生态学报), 2017, 28(4): 1128. [本文引用:1]
[8]	MAO Bo-hui, LI Min-zan, SUN Hong, et al(毛博慧, 李民赞, 孙红, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2017, 33(Z1). [本文引用:1]
[9]	Agrios G N. Plant Pathology(植物病理学). Translated by SHEN Chong-yao(沈崇尧, 译). Beijing: China Agricultural University Press(北京: 中国农业大学出版社), 2009. [本文引用:1]
[10]	DING Hai-ning, CHEN Yu, CHEN Yun-zhi(丁海宁, 陈玉, 陈芸芝). Remote Sensing Technology and Application(遥感技术与应用), 2019, 34(2): 53. [本文引用:1]
[11]	GUO Chao-fan, GUO Xiao-yu(郭超凡, 郭逍宇). Acta Ecologica Sinica(生态学报), 2016, 36(20): 6538. [本文引用:1]
[12]	Soares S, Gomes A A. Trends in Analytical Chemistry, 2013, 42(42): 84. [本文引用:1]
[13]	Galvāo R K H, Araujo M C U, José G E, et al. Talanta, 2005, 67(4): 736. [本文引用:1]
[14]	DOU Zhi-guo, CUI Li-juan, WU Gao-jie, et al(窦志国, 崔丽娟, 武高洁, 等). Chinese Journal of Ecology(生态学杂志), 2018, 37(10): 298. [本文引用:1]
[15]	LIU Ming-bo, TANG Yan-lin, LI Xiao-li, et al(刘明博, 唐延林, 李晓利, 等). Infrared and Laser Engineering(红外与激光工程), 2014, 43(4): 1265. [本文引用:1]

2018

0.0

... 刚竹毒蛾是毛竹的主要虫害, 严重威胁毛竹的生长, 制约了竹林产业的可持续发展^[1] ...

2018

0.0

... 叶绿素是绿色植被吸收光能的物质, 作为参与植被光合作用最重要的色素, 直接影响到绿色植被对光能的利用^[2] ...

2006

0.0

... 时, 常伴随着内部叶绿素含量的变化, 并在光谱特性上发生变化^[3] ...

... 实际应用中, 高光谱数据在提供丰富信息的同时, 也存在信息冗余量大的问题, 导致统计参数的估计偏差增大^[3] ...

2018

0.0

... 通过对高光谱数据进行光谱变换能够增强特征波长信息, 进而提高模型的估算精度^[4] ...

2018

0.0

... 如王雪等^[5]利用竞争性自适应重加权采样算法进行波长筛选, 并结合SPXY样本划分法建立玉米含水率的偏最小二乘估算模型 ...

2014

0.0

GONG

Zhao-ning

, ZHAO

Ya-li

, ZHAO

Wen-ji

, et al(宫兆宁, 赵雅莉, 赵文吉, 等). Acta Ecologica Sinica(生态学报), 2014, 34(20): 5736.

Chlorophyll can be an indicator in photosynthesis capacity and vegetation developmental stages,which is also one of important indicators to monitor health status of wetland vegetation growth. Hyperspectral remote sensing technology can provide a simple, effective and non-destructive data acquisition, which can offer processing method for quantifying diagnosis plant chlorophyll content as well. This study used the Fieldspec 3 spectrometer and a plant probe leaf clip spectral detector to guarantee the spectrum are detected in the same area of the leaf, it is also eliminating the background reflectance, spectral fluctuations caused by bending of the blade surface and the impact caused by leaf internal variability. This study determined the typical wetland plants leaf hyperspectral reflectance data at Wild Duck Lake, and at the same time the corresponding leaf chlorophyll content was measured using a spectrophotometer indoor. The relationship between chlorophyll content and the Trilateral parameters, as well as the ratio of spectral index model (SR) and normalized difference spectral index (ND) were established respectively using linear regression model., then 3-Fold Cross Validation(3K-CV) was used to test the accuracy of the estimation model. The results showed that most of the "trilateral" parameters were significantly correlated with plant leaf chlorophyll content; the maximum correlation coefficient reached 0.867. The correlation coefficient between ratio (SR) and normalized (ND) and chlorophyll content were high in general. Suitable band combinations were 550-700 nm,700-1400nm, 550-700 nm and 1600-1900 nm. The best indices with highest correlation with chlorophyll content were SR (calculated from bands 565 nm and 740 nm) and ND (calculated from bands 565 nm and 735 nm). And then by choosing the best correlation spectrum characteristic parameters based on the Trilateral parameters and ND model index, a plant chlorophyll estimation model was constructed. Among them,a chlorophyll content estimation model established by Red edge position (WP_r) of spectral characteristic parameters and ND (565nm, 735nm) spectral index achieved better test results, and R 2 both reached above 0.8, the estimation model were y =0.113 x -78.74, y =5.5762 x +4.4828. Using 3K-CV method for testing and validation, the prediction accuracies of both plant leaf chlorophyll content estimation models were 93.9% and 90.7%, respectively. The quantitative analysis of hyperspectral remote sensing technology shows a strong advantage in detecting vegetation weak spectral differences and provides an important theoretical basis and technical support for the practical application in the diagnosis of plant chlorophyll content.

叶绿素是光合作用能力和植被发育阶段的指示器，是监测湿地植被生长健康状况的重要指标之一；高光谱遥感技术可以为植物叶绿素含量的定量化诊断提供简便有效、非破坏性的数据采集和处理方法。为保证被探测叶片面积相同，消除背景反射、叶片表面弯曲造成的光谱波动及叶片内部变异造成的影响，研究采用Field Spec 3光谱仪加载手持叶夹式叶片光谱探测器，测定野鸭湖湿地典型植物的叶片高光谱反射率数据，同时通过分光光度计室内测定相应叶片的叶绿素含量。采用相关性及单变量线性拟合分析技术，建立二者的关系模型，包括叶绿素含量与"三边"参数的相关模型以及比值光谱指数（SR）模型和归一化差值光谱指数（ND）模型，并采用交叉检验中的3K-CV方法对估算模型进行模型精度检验。结果表明：植物叶片叶绿素含量与"三边"参数大多都呈极显著相关，相关系数最大达到0.867；计算光谱反射率组成的比值（SR）和归一化（ND）光谱指数与叶绿素含量的决定系数，总体相关性比较高，较好的波段组合均为550-700nm与700-1400nm以及550-700nm与1600-1900nm，与叶绿素含量相关性最好的指数分别是SR（565nm，740nm）和ND（565nm，735nm）。并通过选取相关性最佳的光谱特征参数，分别基于"三边"参数和ND模型指数构建了植物叶片叶绿素含量的估算模型。其中，基于红边位置（WP_r）光谱特征参数和ND（565nm，735nm）光谱指数建立的叶绿素含量估算模型，取得了较好的测试效果，检验拟合方程的决定系数（ R 2 ）都达到0.8以上，估算模型分别为 y =0.113 x -78.74， y =5.5762 x +4.4828。通过3K-CV方法进行测试和检验，植物叶绿素含量估算模型均取得了较为理想的预测精度，预测精度的分别为93.9%及90.7%。高光谱遥感技术对植被进行微弱光谱差异的定量分析，在植被遥感研究与应用中表现出强大优势，为植物叶绿素含量诊断中的实际应用提供了重要的理论依据和技术支持。

... 宫兆宁等^[6]系统分析了典型湿地植被的多项光谱特征参数, 确定了#cod#x0201c ...

2017

0.0

... 刘文雅等^[7]利用相关系数法筛选出与马尾松叶绿素含量高度相关的植被指数, 并对马尾松冠层叶绿素含量进行估算 ...

2017

0.0

... 毛博慧等^[8]采取遗传算法进行波长筛选, 利用优选波长组合植被指数, 建立基于敏感植被指数的冬小麦叶绿素含量估算模型 ...

2009

0.0

... 根据刚竹毒蛾虫害发生的特点, 随机选取6个不同样方, 参照《植物病理学》的虫害划分标准, 将毛竹样本分为健康、轻度虫害、中度虫害及重度虫害4个等级^[9], 对采集的不同虫害等级毛竹叶片进行光谱及叶绿素含量的现场测定 ...

2019

0.0

... 对原始光谱(original spectrum, OS)进行一阶导数(first derivative, FD)处理, 可消除与波长无关的漂移^[10] ...

2016

0.0

GUO

Chao-fan

, GUO

Xiao-yu

(郭超凡, 郭逍宇). Acta Ecologica Sinica(生态学报), 2016, 36(20): 6538.

Increasing amounts of recycled water are being used in urban wetlands; as such, monitoring the growth of wetland plants over large areas is of great significance to assessing the restoration and reconstruction of wetlands created by recycled water. At present, remote-sensing technology is considered an important method for monitoring the growth of plants on a large scale. In this study, typical wetland plants ( Phragmites australis and Typha angustifolia ) growing in the South Park Wetland of Olympic Park were selected as research subjects. Spectral reflectance was determined at a domain ranging from 400 to 760 nm to avoid the influence of leaf water on the established model. Chlorophyll content was obtained from data sources. Statistical analysis, including correlation and stepwise regression analysis, was conducted to establish chlorophyll content inversion models with different derivative transformation spectrums at the leaf level for:(1) band depth (BD), (2) continuum-removed derivative reflectance (CRDR), (3) band depth ratio (BDR), and (4) normalized band depth index (NBDI). We found that 550 nm to 750 nm, particularly 700 nm to 750 nm (red edge range), was the key range to estimate biochemical parameters. Single removal cross-validation results indicated that optimal models of chlorophyll content inversion in reeds, cattails, and combined samples were BD, CRDR, and NBDI, respectively. Corresponding R 2 values were 0.87, 0.83, and 0.81, and the respective RMSE values were 0.16, 0.15, and 0.33, respectively. Kruskal-Wallis non-parametric tests and multi-way ANOVAs were performed to elucidate the influence of relevant factors individually and in combination with one another on the regression results of biochemical parameters of plant water. Results showed that vegetation type (reed, cattail) and data type (single or mixed species) greatly influenced the inversion model. In contrast, the spectral derivative transformation(BD, CRDR, BDR, and CRDR)and the interaction between spectral derivative transformation and data types did not significantly affect the inversion model. In this study, an estimation model of wetland plant biochemical parameters was established and functions of related factors in the estimation model were analyzed. Our results could be used as a scientific basis for non-destructive monitoring of growth in wetland plants. This study also provided a reference for the use of recycled water in restoration and management.

研究采用芦苇和香蒲叶片光谱及实测叶绿素含量数据，选取波段谱带范围为可见光波段400-760nm（为了避免近红外波段受叶片水分含量的影响，降低构建模型的稳定性），利用相关分析与逐步回归分析的统计学分析方法，建立叶面尺度下不同包络线去除衍生转换光谱：BD（band depth）、CRDR（continuum-removed derivative reflectance）、BDR（band depth ratio）、NBDI（normalized band depth index）与叶绿素含量估算模型。通过对入选波段的统计表明在550-750nm，特别是700-750nm（红边）波段范围内产生了较多的有效波段，是今后进行生物参量反演的重点波段范围。舍一交叉验证结果表明芦苇、香蒲和混合样本绿素含量估测的最佳模型分别为BD、CRDR和NBDI模型，其交叉验证决定系数依次为0.87、0.83和0.81，交叉验证均方根误差RMSE依次为0.16、0.15和0.33。并在此基础上利用独立样本非参数检验和多因子方差分析，探讨相关因素对于叶绿素含量估算模型精度的影响。结果表明物种差异、数据类型差异对于叶绿素回归模型的影响较大，而光谱类型差异及光谱数据与数据类型交互作用对于回归模型精度的影响较小。

... 包络线去除(continuum removal, CR)是将原始光谱曲线与包络线曲线进行比值运算后得到, 该方法能够增强各生化参量敏感波段的光谱吸收特征^[11], 并计算包络线去除后光谱的一阶导数(continuum removal-first derivative, CR-FD) ...

2013

0.0

... 该算法的步骤如下^[12]: ...

2005

0.0

... 4 SPXY算法样本数据集的划分会在一定程度上影响模型估算的精度, SPXY(sample set partitioning based on joint x-y distance)法^[13]是一种同时考虑样本光谱以及生化参量欧氏距离的样本集选择方法, 使选择的样本集更具有代表性, 从而改善所建模型的估算能力 ...

2018

0.0

... 75时, 模型的估算效果较差^[14] ...

2014

0.0

LIU

Ming-bo

, TANG

Yan-lin

, LI

Xiao-li

, et al(刘明博, 唐延林, 李晓利, 等). Infrared and Laser Engineering(红外与激光工程), 2014, 43(4): 1265.

5 segments moving average, baseline correction, area normalization, and multiplicative scatter correction (MSC) was used to preprocess Visible-NIR reflective spectrum of rice leaf. Successive projection algorithm (SPA) was used in the selecting of effective wavelengths. Multiple linear regression (MLR) models were built based on spectral indexes of RVI, NDVI and effective wavelengths selected by SPA. Principal components regression (PCR) models and Partial least squares regression (PLS) models were built based on all wavelengths in the spectrum. Nitrogen contents of rice leaves were predicted by these models. From comparison, It was found that the predictive validity of models based on SPA effective wavelengths were obviously better than models based on spectral indexes of RVI and NDVI, and slightly worse than PCR and PLS models based on all wavelengths in the spectrum. Models based on MSC preprocessed spectrum and SPA effective wavelengths has the predictive validity of r =0.7943, RMSE =0.4558. It is feasible to use successive projections algorithm in spectral monitoring of rice leaves nitrogen contents.

College of Sciences, Guizhou University, Guiyang 550025, China

使用5段移动平滑法、基线校正、光谱面积归一化、多元散射校正方法对水稻叶片可见-近红外光谱进行预处理，使用连续投影算法（SPA）进行有效波长的选取。分别基于光谱指数RVI、NDVI建立多元线性回归（MLR）模型，基于SPA有效波长建立MLR模型，基于全部波长建立主成分回归（PCR）及偏最小二乘法（PLS）回归模型。利用模型预测水稻叶片氮含量，对比发现基于SPA有效波长建立的模型的预测效果明显好于基于光谱指数RVI及NDVI建立的模型，略差于基于全部波长建立的PCR及PLS模型。基于MSC预处理光谱及SPA有效波长建立的模型预测集预测结果 r =0.7943， RMSE =0.4558。在水稻叶片氮含量光谱监测中使用连续投影算法进行有效波长的选取是可行的。

... 由于算法本身的原因导致易在波长范围边缘处选取波长^[15] ...