基于高光谱遥感的玉米叶片SPAD值估算模型研究

doi:10.16035/j.issn.1001-7283.2019.03.020

摘要/Abstract

摘要：

灌浆期玉米叶片叶绿素含量对玉米光合作用及产量形成具有重要作用。为通过高光谱特征准确、高效估测玉米叶片叶绿素含量,以SPAD值表征叶绿素相对含量,构建了基于光谱特征参数的传统回归模型、基于全谱和光谱特征参数的PLSR模型和BP神经网络模型,并进行了比较分析。结果表明：基于全谱构建的PLSR模型SPAD值拟合效果最好（R ²=0.910,RMSE=2.071）,而基于光谱特征参数所建立的PLSR模型拟合效果可达到与全谱PLSR模型相近的水平。但后者的实测值与预测值拟合效果（R ²=0.867,RMSE=2.581,RPD=2.628）优于全谱PLSR模型,且建模时间短,模型复杂程度降低。BP神经网络模型相较于两种PLSR模型预测效果略差,但优于传统回归模型。综合来看,基于光谱特征参数建立的PLSR模型估测效果最好。

关键词: 高光谱, 玉米叶片, SPAD, 偏最小二乘, BP神经网络

Abstract:

The leaf chlorophyll content at the filling stage plays an important role in photosynthesis and yield formation of maize. To estimate the maize leaf chlorophyll content accurately and efficiently, hyperspectral remote sensing was used by taking SPAD value as the relative chlorophyll content. Conventional regression models based on spectral characteristic parameters, PLSR models based on full spectrum and spectral characteristic parameters, and BP neural network model were constructed and compared. The results showed that SPAD value of PLSR model based on full spectrum had the best fitting effect (R ²=0.910, RMSE=2.071). The fitting effect of PLSR model based on spectral characteristic parameters was close to that of PLSR model based on full spectrum. But the measured and predicted values of the latter fitting effect (R ²=0.867, RMSE=2.581, RPD=2.628) was better than that of full spectrum PLSR model, and the PLSR model based on spectral characteristic parameters had short modeling time and tow complexity. The prediction effect of the BP neural network model was worse than the two PLSR models, but better than the conventional regression models based on spectral characteristic parameters. In brief, the estimation effect of PLSR model based on spectral parameters was the best.

Key words: Hyperspectral, Maize leaf, SPAD, Partial least square regression, BP neural network

董哲,杨武德,张美俊,朱洪芬,王超. 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 Zhe,Yang Wude,Zhang Meijun,Zhu Hongfen,Wang Chao. Estimation Models of Maize Leaf SPAD Value Based on Hyperspectral Remote Sensing[J]. Crops, 2019(3): 126–131

图/表 7

表1

表2

表3

表4

表5

表6

图1

参考文献 18

[1]	Curran P J, Windham W R, Gholz H L . Exploring the relationship between reflectance red edge and chlorophyll concentration in slash pine leaves. Tree Physiology, 1995,15(3):203-206. doi: 10.1093/treephys/7.1-2-3-4.33 pmid: 14965977
[2]	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. 光谱学与光谱分析, 2018,38(5):1533-1539.
[3]	王晓星 . 夏玉米冠层光谱特征及其生理生态参量的高光谱估算模型. 杨凌:西北农林科技大学, 2015.
[4]	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. 农业工程学报, 2016,32(16):135-142.
[5]	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. 光谱学与光谱分析, 2015,35(2):486-491.
[6]	Thenkabail P S, Enclona E A, Ashton M S , et al. Accuracy assessments of hyperspectral waveband performance for vegetation analysis applications. Remote Sensing of Environment, 2004,91(3):354-376. doi: 10.1016/j.rse.2004.03.013
[7]	Thenkabail P S, Smith R B, Pauw E D . Evaluation of narrowband and broadband vegetation indices for determining optimal hyperspectral wavebands for agricultural crop characterization. Photogrammetric Engineering & Remote Sensing, 2002,68(6):607-621.
[8]	Chan C W, Paelinckx D . Evaluation of Random Forest and Adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery. Remote Sensing of Environment, 2008,112(6):2999-3011. doi: 10.1016/j.rse.2008.02.011
[9]	Gitelson A A, Zur Y, Chivkunova O B , et al. Assessing carotenoid content in plant leaves with reflectance spectroscopy. Photochemistry & Photobiology, 2010,75(3):272-281.
[10]	刘海启, 李召良 . 高光谱植被遥感. 北京: 中国农业科学技术出版社, 2015: 85-86.
[11]	Gitelson A, Kaufman Y J, Stark R , et al. Novel algorithms for remote estimation of vegetation fraction. Remote Sensing of Environment, 2002,80(1):76-87. doi: 10.1016/S0034-4257(01)00289-9
[12]	Penuelas J, Baret F, Filella I . Semiempirical indexes to assess carotenoids chlorophyll-a ratio from leaf spectral reflectance. Photosynthetica, 1995,31(2):221-230.
[13]	Horler D N H, Barber J, Darch J P , et al. Approaches to detection of geochemical stress in vegetation. Advances in Space Research, 1983,3(2):175-179.
[14]	姚付启, 张振华, 杨润亚 , 等. 基于红边参数的植被叶绿素含量高光谱估算模型. 农业工程学报, 2009,25(S2):123-129.
[15]	马赵扬, 杜昌文, 周健民 . 土壤碳酸钙中红外光声光谱特征及其应用. 光谱学与光谱分析, 2012,32(5):1255-1258.
[16]	陈志强, 王磊, 白由路 , 等. 整个生育期玉米叶片SPAD高光谱预测模型研究. 光谱学与光谱分析, 2013,33(10):2838-2842.
[17]	宋开山, 张柏, 李方 , 等. 高光谱反射率与大豆叶面积及地上鲜生物量的相关分析. 农业工程学报, 2005,21(1):36-40.
[18]	汤旭光, 宋开山, 刘殿伟 , 等. 基于可见/近红外反射光谱的大豆叶绿素含量估算方法比较. 光谱学与光谱分析, 2011,31(2):371-374.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

光谱特征参数Spectral characteristic parameter	定义Definition	来源Source
NDVI	NDVI=(R₇₅₀-R₇₀₅)/(R₇₅₀+R₇₀₅)	Gitelson等^[11]
GNDVI	GNDVI=(R₇₅₀-R₅₅₀)/(R₇₅₀+R₅₅₀)	Gitelson等^[11]
RVI	RVI=R₇₆₀/R₇₁₀	Penuelas等^[12]
λ_r	红光范围内一阶导数光谱最大值对应的波长	Horler等^[13]
k_λr	红光范围内一阶导数光谱的峰度	姚付启等^[14]
s_λr	红光范围内一阶导数光谱的偏度	姚付启等^[14]

参数Parameter		相关系数 Correlation coefficient
植被指数Vegetation index	NDVI	0.843^**
	GNDVI	0.811^**
	RVI	0.893^**
红边参数The trilateral parameter	λ_r	0.935^**
	k_λr	0.872^**
	s_λr	0.914^**

光谱特征参数 Spectral characteristic parameter	最优拟合模型 The optimal fitting equation	建模集Modeling set		验证集Validation set
光谱特征参数 Spectral characteristic parameter	最优拟合模型 The optimal fitting equation	R²	RMSE	R²	RMSE	RPD
NDVI	y=12.696e^2.6521x	0.728	3.686	0.636	4.072	1.500
GNDVI	y=10.909e^2.7102x	0.672	4.020	0.574	4.366	1.340
RVI	y=20.758x^0.9129	0.812	3.095	0.716	3.715	1.791
λ_r	y=3E-104x^36.805	0.880	3.774	0.854	3.327	2.005
k_λr	y=640.3e^1.7036x	0.784	3.432	0.714	3.657	1.739
s_λr	y=87.296x²+54.674x+46.509	0.847	2.697	0.804	3.028	2.137

模型Model	建模集 Modeling set		验证集 Validation set
模型Model	R²	RMSE	R²	RMSE	RPD
PLSR（全谱）	0.910	2.071	0.862	2.600	2.545
PLSR（光谱特征参数）	0.900	2.200	0.867	2.581	2.628

建模集Modeling set			验证集Validation set
R²	RMSE		R²	RMSE	RPD
0.904	2.165		0.822	2.861	2.305