基于高光谱的油菜叶片SPAD值估测模型比较

王克晓; 周蕊; 李波; 欧毅; 黄祥; 虞豹

doi:10.19303/j.issn.1008-0384.2021.11.003

基于高光谱的油菜叶片SPAD值估测模型比较

doi: 10.19303/j.issn.1008-0384.2021.11.003

重庆市农业科学院农业科技信息研究所，重庆　401329

基金项目: 重庆市技术创新与应用发展专项（cstc2019jscx-msxmX0355、cstc2020jscx-lyjsAX0012）；重庆市农业科学院农业发展资金项目（NKY-2020AB009、NKY-2021AC012）

详细信息

作者简介:
王克晓（1986−），男，硕士，助理研究员，主要从事农业遥感相关研究(E-mail：447215670@qq.com)

通讯作者:
周蕊(1980−)，女，硕士，副研究员，主要从事农业信息化研究（E-mail：12087836@qq.com）

中图分类号: S 565.4; S 127
计量
- 文章访问数: 617
- HTML全文浏览量: 167
- PDF下载量: 29
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-14
- 修回日期: 2021-10-11
- 网络出版日期: 2021-12-30
- 刊出日期: 2021-11-28

Comparison of Estimation Models for Hyperspectral-based Rape Leaf SPAD

Institute of Agricultural Science and Technology Information, Chongqing Academy of Agricultural Sciences, Chongqing　401329, China

摘要

摘要: 目的比较基于高光谱参数的油菜叶片SPAD值估算模型效果。方法在分析光谱反射特征和光谱参数与SPAD值相关性的基础上，利用光谱特征参数优选并构建了偏最小二乘回归（PLSR）、传统反向传播神经网络（BPNN）、支持向量回归（SVR）和深度学习神经网络（DNN）等模型对叶片样本叶绿素SPAD值进行估测。结果 ①叶片原始光谱与叶片SPAD值在425～495 nm的蓝波、665～680 nm的红波区域呈现微弱正相关，与红边波段均呈现负相关，并在510～650 nm的绿、黄波段和690～735 nm的红边波段显著负相关；②与叶片SPAD值显著线性相关的S_Db与S_Dy、CARI与MCARI、CI与NDVI₇₀₅等三组光谱特征的组内参数具有一定的可替代性，而且有助于提高SPAD模型预测精度；③基于高光谱参数的深度学习DNN模型决定系数R²为0.93，RPD为3.92，具有较高的预测能力，SVR模型次之，PLSR和BPNN模型效果一般。结论油菜叶片光谱参数之间存在不同程度的相关性，基于机器学习的非线性估计模型具有较高的稳定性和预测能力，深度学习算法在油菜叶片叶绿素SPAD值估测方面具有更好的估测能力。
- 高光谱 /
- 叶绿素 /
- 机器学习 /
- 油菜
Abstract: Objective In order to compare the estimation model effect of SPAD of rape leaves based on hyperspectral parameters. Method Models of partial least squares regression (PLSR), back propagation neural network (BPNN), support vector regression (SVR), and deep neural network (DNN) based on the spectral parameters selected from the correlation analysis between the spectral reflectance parameters and SPAD data were constructed and compared for the estimation of chlorophyll SPAD of rape leaves. Result The SPADs and the original spectra in the blue wave of 425-495 nm and red wave of 665-680 nm of the leaves had a weak positive correlation. However, significantly inverse correlations between the SPADs and the green-yellow band of 510-650 nm and between that and the red edge band of 690-735nm were observed. The negative correlation coefficient between S_Db and S_Dy was as high as −0.98, while the positive correlation coefficient between CARI and MCARI, CI and NDVI₇₀₅ 0.99. The 3 sets of S_Db and S_Dy, CARI and MCARI, and CI and NDVI₇₀₅ had significant linear correlations with the leaf SPAD. They could be somewhat interchangeable rendering them potential for accuracy improvement. The DNN model had an R² of 0.93 and an RPD of 3.92 indicating a high predictability of the two models. They were followed by SVR, while PLSR and BPNN models being similar. Conclusion There were different degrees of correlation between the SPADs and the spectral parameters of rape leaves. The non-linear prediction model based on machine learning showed higher stability and predictability than the others, and the deep learning algorithm more effective in estimating SPAD of rape leaves.
- hyperspectral /
- chlorophyll /
- machine learning /
- rape

HTML全文

图 1 叶片光谱与SPAD值相关性

Figure 1. Correlation between spectra and SPAD of leaves

下载: 全尺寸图片幻灯片

图 2 光谱参数及与SPAD值的相关性

注：×表示在0.01的检验水平上显著不相关。

Figure 2. Correlation between spectral parameters and SPAD

Note: × means significantly irrelevant at p<0.01.

下载: 全尺寸图片幻灯片

图 3 叶片SPAD值PLSR模型

Figure 3. PLSR model of leaf SPAD

下载: 全尺寸图片幻灯片

图 4 叶片SPAD值DNN回归模型

Figure 4. DNN model on rape leaf SPAD

下载: 全尺寸图片幻灯片

表 1 光谱参数及其定义或计算公式

Table 1. Spectral parameters and definitions or calculation formula

类型 Type	特征变量 Characteristic variable	定义或算法 Definition or algorithm
位置特征 Location feature	D_b（蓝边幅值） Blue edge amplitude	490～530 nm一阶微分最大值 Maximal 1^st-order differential 490-530 nm
	D_y（黄边幅值） Yellow edge amplitude	560～640 nm一阶微分最大值 Maximal 1^st-order differential 560-640 nm
	D_r（红边幅值） Red edge amplitude	680～760 nm一阶微分最大值 Maximal 1^st-order differential 680-760 nm
	R_g（绿峰反射率） Green peak reflectance	510～560 nm内光谱反射率最大值 Maximal reflectance 510-560 nm
	R_r（红谷反射率） Red valley reflectance	640～680 nm内光谱反射率最小值 Minimal reflectance 640-680 nm
面积特征 Area feature	S_Db（蓝边面积） Blue edge area	波长490～530 nm一阶导数光谱积分 1^st-derivative integral at 490-530nm
	S_Dy（黄边面积） Yellow edge area	波长560～640 nm一阶导数光谱积分 1^st-derivative integral at 560-640nm
	S_Dr（红边面积） Red edge area	波长680～760 nm一阶导数光谱积分 1^st-derivative integral at 680-760nm
植被指数 Vegetation index	CARI（叶绿素吸收比） Chlorophyll absorption reflectance index	（R₇₀₀−R₆₇₀）−0.2×（R₇₀₀−R₅₅₀）
	MCARI（改进叶绿素吸收比） Modified chlorophyll absorption reflectance index	[（R₇₀₀−R₆₇₀）−0.2×（R₇₀₀−R₅₅₀）]×（R₇₀₀/R₆₇₀）
	CI（红边叶绿素指数） Red edge chlorophyll index	R₇₅₀/R₇₀₅−1
	NDVI₇₀₅（红边归一化植被指数） Red edge normalized difference vegetation index	（R₇₅₀−R₇₀₅）/（R₇₅₀+R₇₀₅）
	NPCI（归一化总色素叶绿素指数） normalized total pigment to chlorophyll index	（R₆₈₀−R₄₃₀）/（R₆₈₀+R₄₃₀）

下载: 导出CSV

表 2 深度神经网络算法参数

Table 2. Parameters applied for deep neural network algorithm

参数 Parameter	取值 Value	参数 Parameter	取值 Value
隐层层数（Hidden）	2	节点（Nodes）	c(5，5)
激活函数（Activation）	Rectifier	停止测度（Stopping metric）	方均误差（MSE）
停止容差（Stopping tolerance）	0.02	周期（Epochs）	500

下载: 导出CSV

表 3 不同模型精度比较

Table 3. Accuracy of models

模型 Models	建模精度 Modeling accuracy			测试精度 Test accuracy
模型 Models	R²	RMSE	RPD	R²	RMSE
PLSR	0.66（0.61）	1.76（1.86）	1.69（1.61）	0.63（0.58）	1.77（1.96）
BPNN	0.72（0.69）	1.58（1.68）	1.89（1.78）	0.70（0.67）	2.02（2.22）
SVR	0.80（0.91）	1.39（1.02）	2.14（2.95）	0.69（0.74）	1.57（1.53）
DNN	0.93（0.92）	0.77（0.85）	3.92（3.51）	0.78（0.80）	1.78（1.76）
注：括号内数据为变量优选前相应模型的指标精度。 Note: Numbers in brackets are factors of models before variable optimization.

下载: 导出CSV

参考文献(25)

[1]	李哲, 张飞, 陈丽华, 等. 光谱指数的植物叶片叶绿素含量估算模型 [J]. 光谱学与光谱分析, 2018, 38(5):1533−1539. LI Z, ZHANG F, CHEN L H, et al. Research on spectrum variance of vegetation leaves and estimation model for leaf chlorophyll content based on the spectral index [J]. Spectroscopy and Spectral Analysis, 2018, 38(5): 1533−1539.(in Chinese)
[2]	黄祥, 周蕊, 王茜, 等. 遥感定量反演农作物叶绿素的现状与发展 [J]. 安徽农业科学, 2018, 46(32):192−194, 202. doi: 10.3969/j.issn.0517-6611.2018.32.056 HUANG X, ZHOU R, WANG Q, et al. The status and development of quantitative retrieval of crop chlorophyll by remote sensing [J]. Journal of Anhui Agricultural Sciences, 2018, 46(32): 192−194, 202.(in Chinese) doi: 10.3969/j.issn.0517-6611.2018.32.056
[3]	丁希斌, 刘飞, 张初, 等. 基于高光谱成像技术的油菜叶片SPAD值检测 [J]. 光谱学与光谱分析, 2015, 35(2):486−491. doi: 10.3964/j.issn.1000-0593(2015)02-0486-06 DING X B, LIU F, ZHANG C, et al. Prediction of SPAD value in oilseed rape leaves using hyperspectral imaging technique [J]. Spectroscopy and Spectral Analysis, 2015, 35(2): 486−491.(in Chinese) doi: 10.3964/j.issn.1000-0593(2015)02-0486-06
[4]	殷紫, 常庆瑞, 刘淼, 等. 基于光谱指数的不同生育期油菜叶片SPAD估测 [J]. 西北农林科技大学学报(自然科学版), 2017, 45(5):66−72. YIN Z, CHANG Q R, LIU M, et al. Estimation of rape leaf SPAD in different periods based on spectral indices [J]. Journal of Northwest A & F University (Natural Science Edition), 2017, 45(5): 66−72.(in Chinese)
[5]	张锐, 廖桂平, 王访, 等. 基于冠层高光谱的油菜角果期红边参数及叶片SPAD值反演模型 [J]. 江苏农业科学, 2019, 47(20):255−259. ZHANG R, LIAO G P, WANG F, et al. Red edge parameters and SPAD inversion model of rapeseed based on canopy hyperspectral data [J]. Jiangsu Agricultural Sciences, 2019, 47(20): 255−259.(in Chinese)
[6]	鲍义东, 陈秋实, 陈果. 高光谱技术在农业遥感中的应用 [J]. 电子技术与软件工程, 2020(8):170−171. BAO Y D, CHEN Q S, CHEN G. Application of hyperspectral technique in agricultural remote sensing [J]. Electronic Technology & Software Engineering, 2020(8): 170−171.(in Chinese)
[7]	何勇, 彭继宇, 刘飞, 等. 基于光谱和成像技术的作物养分生理信息快速检测研究进展 [J]. 农业工程学报, 2015, 31(3):174−189. doi: 10.3969/j.issn.1002-6819.2015.03.024 HE Y, PENG J Y, LIU F, et al. Critical review of fast detection of crop nutrient and physiological information with spectral and imaging technology [J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31(3): 174−189.(in Chinese) doi: 10.3969/j.issn.1002-6819.2015.03.024
[8]	HORLER D N H, DOCKRAY M, BARBER J. The red edge of plant leaf reflectance [J]. International Journal of Remote Sensing, 1983, 4(2): 273−288. doi: 10.1080/01431168308948546
[9]	DAUGHTRY C S T, WALTHALL C L, KIM M S, et al. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance [J]. Remote Sensing of Environment, 2000, 74(2): 229−239. doi: 10.1016/S0034-4257(00)00113-9
[10]	BROGE N H, MORTENSEN J V. Deriving green crop area index and canopy chlorophyll density of winter wheat from spectral reflectance data [J]. Remote Sensing of Environment, 2002, 81(1): 45−57. doi: 10.1016/S0034-4257(01)00332-7
[11]	姚付启, 张振华, 杨润亚, 等. 基于主成分分析和BP神经网络的法国梧桐叶绿素含量高光谱反演研究 [J]. 测绘科学, 2010, 35(1):109−112. YAO F Q, ZHANG Z H, YANG R Y, et al. Research on Platanus orientalis L. chlorophyll concentration estimation with hyperspectral data based on BP-artificial neural network and principal component analysis [J]. Science of Surveying and Mapping, 2010, 35(1): 109−112.(in Chinese)
[12]	孙明馨, 刘琪, 王帅, 等. 基于高光谱的低温胁迫下冬小麦SPAD估算 [J]. 福建农林大学学报(自然科学版), 2020, 49(6):728−733. SUN M X, LIU Q, WANG S, et al. SPAD estimation of winter wheat under low temperature stress based on hyper-spectrum [J]. Journal of Fujian Agriculture and Forestry University (Natural Science Edition), 2020, 49(6): 728−733.(in Chinese)
[13]	赵琨, 王珺珂, 王楚锋, 等. 基于高光谱成像技术的油菜SPAD值空间分布预测及最佳测量叶位 [J]. 华中农业大学学报, 2018, 37(4):78−84. ZHAO K, WANG J K, WANG C F, et al. Hyperspectral imaging technology based prediction of spatial distribution of SPAD value of rapeseed and optimal measurement of leaf position [J]. Journal of Huazhong Agricultural University, 2018, 37(4): 78−84.(in Chinese)
[14]	由明明, 常庆瑞, 田明璐, 等. 基于随机森林回归的油菜叶片SPAD值遥感估算 [J]. 干旱地区农业研究, 2019, 37(1):74−81. doi: 10.7606/j.issn.1000-7601.2019.01.10 YOU M M, CHANG Q R, TIAN M L, et al. Estimation of rapeseed leaf SPAD value based on random forest regression [J]. Agricultural Research in the Arid Areas, 2019, 37(1): 74−81.(in Chinese) doi: 10.7606/j.issn.1000-7601.2019.01.10
[15]	崔小涛, 常庆瑞, 屈春燕, 等. 基于高光谱和MLSR-GA-BP神经网络模型油菜叶片SPAD值遥感估算 [J]. 东北农业大学学报, 2020, 51(8):74−84. doi: 10.3969/j.issn.1005-9369.2020.08.010 CUI X T, CHANG Q R, QU C Y, et al. Remote sensing estimation of SPAD value for rape leaf based on hyperspectral and MLSR-GA-BP neural network model [J]. Journal of Northeast Agricultural University, 2020, 51(8): 74−84.(in Chinese) doi: 10.3969/j.issn.1005-9369.2020.08.010
[16]	李媛媛, 常庆瑞, 刘秀英, 等. 基于高光谱和BP神经网络的玉米叶片SPAD值遥感估算 [J]. 农业工程学报, 2016, 32(16):135−142. doi: 10.11975/j.issn.1002-6819.2016.16.019 LI Y Y, CHANG Q R, LIU X Y, et al. Estimation of maize leaf SPAD value based on hyperspectrum and BP neural network [J]. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(16): 135−142.(in Chinese) doi: 10.11975/j.issn.1002-6819.2016.16.019
[17]	康丽, 高睿, 孔庆明, 等. 水稻叶片SPAD值高光谱成像估测 [J]. 东北农业大学学报, 2020, 51(10):89−96. doi: 10.3969/j.issn.1005-9369.2020.10.011 KANG L, GAO R, KONG Q M, et al. Estimation of SPAD value of rice leaves based on hyperspectral image [J]. Journal of Northeast Agricultural University, 2020, 51(10): 89−96.(in Chinese) doi: 10.3969/j.issn.1005-9369.2020.10.011
[18]	刘恬琳, 朱西存, 白雪源, 等. 土壤有机质含量高光谱估测模型构建及精度对比 [J]. 智慧农业(中英文), 2020, 2(3):129−138. LIU T L, ZHU X C, BAI X Y, et al. Hyperspectral estimation model construction and accuracy comparison of soil organic matter content [J]. Smart Agriculture, 2020, 2(3): 129−138.(in Chinese)
[19]	于雷, 洪永胜, 耿雷, 等. 基于偏最小二乘回归的土壤有机质含量高光谱估算 [J]. 农业工程学报, 2015, 31(14):103−109. doi: 10.11975/j.issn.1002-6819.2015.14.015 YU L, HONG Y S, GENG L, et al. Hyperspectral estimation of soil organic matter content based on partial least squares regression [J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31(14): 103−109.(in Chinese) doi: 10.11975/j.issn.1002-6819.2015.14.015
[20]	张燕平, 张铃. 机器学习理论与算法[M]. 北京: 科学出版社, 2012: 27.
[21]	ERIN L D, NAVDEEP G, SPENCER A, et al (2021). H2O: R Interface for the 'H2O' scalable machine learning platform. R package version 3.32.1.3. https://CRAN.R-project.org/package=h2o.
[22]	程显毅, 施佺. 深度学习与R语言[M]. 北京: 机械工业出版社, 2017: 164 − 165.
[23]	刘宁, 邢子正, 乔浪, 等. 基于模型集群的马铃薯叶绿素检测光谱变量筛选讨论 [J]. 光谱学与光谱分析, 2020, 40(7):2259−2266. LIU N, XING Z Z, QIAO L, et al. Discussion on spectral variables selection of potato chlorophyll using model population analysis [J]. Spectroscopy and Spectral Analysis, 2020, 40(7): 2259−2266.(in Chinese)
[24]	徐逸, 董轩妍, 王俊杰. 4种机器学习模型反演太湖叶绿素a浓度的比较 [J]. 水生态学杂志, 2019, 40(4):48−57. XU Y, DONG X Y, WANG J J. Use of remote multispectral imaging to monitor chlorophyll-a in Taihu lake: A comparison of four machine learning models [J]. Journal of Hydroecology, 2019, 40(4): 48−57.(in Chinese)
[25]	董哲, 杨武德, 张美俊, 等. 基于高光谱遥感的玉米叶片SPAD值估算模型研究 [J]. 作物杂志, 2019(3):126−131. DONG Z, YANG W D, ZHANG M J, et al. Estimation models of maize leaf SPAD value based on hyperspectral remote sensing [J]. Crops, 2019(3): 126−131.(in Chinese)