基于RF-VR的紫丁香叶片叶绿素含量高光谱反演

doi:10.3969/j.issn.1004-1524.2021.11.19

浙江农业学报 ›› 2021, Vol. 33 ›› Issue (11): 2164-2173.DOI: 10.3969/j.issn.1004-1524.2021.11.19

基于RF-VR的紫丁香叶片叶绿素含量高光谱反演

肖志云(), 王伊凝

内蒙古工业大学电力学院,内蒙古机电控制重点实验室,内蒙古呼和浩特 010051

收稿日期:2020-08-25 出版日期:2021-11-25 发布日期:2021-11-26
作者简介:肖志云(1974—),男,湖南嘉禾人,教授,博士,主要从事机器视觉在农业中的应用研究。E-mail: xiaozhiyun@imut.edu.cn
基金资助:
国家自然科学基金(61661042)

Hyperspectral retrieval for chlorophyll contents of Syringa oblata leaves based on RF-VR

XIAO Zhiyun(), WANG Yining

College of Electricity Power, Inner Mongolia Key Laboratory of Mechatronic Control, Inner Mongolia University of Technology, Huhhot 010051, China

Received:2020-08-25 Online:2021-11-25 Published:2021-11-26

摘要/Abstract

摘要：

利用高光谱技术精确估测植物叶片叶绿素含量,对植物生长趋势和营养状况的监测和管理具有重要意义。本文以紫丁香为研究对象,针对高光谱所含波段数量大、波段间相关性强导致数据中冗余信息增多的现象,通过卷积平滑和二阶微分(SG-SD)处理光谱数据,应用随机蛙跳(RF)算法筛选特征波段,最后结合偏最小二乘(PLSR)和投票回归器(VR)建立了植物叶片叶绿素含量反演模型,并与全波段光谱法和5种经典变量提取方法进行了比较。结果显示,相比于原始光谱数据,SG-SD是一种有效的提高建模精度的光谱预处理方法;相比于全波段光谱和经典变量提取方法,RF算法筛选出的敏感波段建模效果最佳;相比于PLSR模型,VR模型的预测精度和预测稳定性能更优。本文对原始光谱数据进行SG-SD预处理后,对经RF算法筛选出的特征波段建立VR模型,变量数由全波段数204个减少为35个,建模集决定系数0.944 2,验证集决定系数0.951 4,最后利用RF-VR模型结合伪彩图技术得到紫丁香叶片叶绿素分布反演图,为紫丁香叶片养分分布提供更直观的信息表达。结果表明,该方法可为紫丁香叶片营养含量诊断和长势监测提供技术支持。

关键词: 紫丁香, 叶绿素含量, 高光谱, 光谱预处理, 随机蛙跳算法, 投票回归器

Abstract:

Accurate estimation of the chlorophyll content of plant leaves by using hyperspectral technology is of great significance to monitor and manage plant growth trends and nutritional status. Taking Syringa oblata as the research object, aiming at the phenomenon that the large number of bands and strong correlation between bands lead to the increase of redundant information in the data, the spectral data were processed by convolution smoothing and second-order differentiation (SG-SD), the random leapfrog (RF) algorithm was used to screen the characteristic bands, and finally combined with partial least squares (PLSR) and voting regression (VR). The inversion model of chlorophyll content in plant leaves was established and compared with full band spectroscopy and five classical variable extraction methods. The results showed that compared with the original spectral data, SG-SD was an effective spectral pretreatment method to improve modeling accuracy; compared with full-band spectrum and 5 classical variable selection methods, the sensitive bands selected by the RF algorithm had the best modeling accuracy; compared with PLSR model, the prediction accuracy and stability of the VR model were better. In the present paper, after the SG-SD pretreatment of the original spectral data, a VR model was established for the sensitive bands selected by the RF algorithm, the variable number reduced from 204 to 35, the determination coefficients of modeling set and validation set were 0.944 2 and 0.951 4 respectively. Finally, using RF-VR model and pseudo color map technology, the inversion map of chlorophyll distribution of Syringa oblata leaves was obtained, which provided more intuitive information expression for nutrient distribution of Syringa oblata leaves. It was concluded that this method could provide technical support for the diagnosis and growth monitoring of the nutrient content of Syringa oblata leaves.

Key words: Syringa oblata, chlorophyll content, hyperspectral image, spectral data processing, random frog, vote regressor

中图分类号:

TP391.41

肖志云, 王伊凝. 基于RF-VR的紫丁香叶片叶绿素含量高光谱反演[J]. 浙江农业学报, 2021, 33(11): 2164-2173.

XIAO Zhiyun, WANG Yining. Hyperspectral retrieval for chlorophyll contents of Syringa oblata leaves based on RF-VR[J]. Acta Agriculturae Zhejiangensis, 2021, 33(11): 2164-2173.

图/表 11

图1 高光谱成像系统 1,样品;2、3,卤素电源;4,可控载物台;5,高光谱相机;6,数据传输线;7,计算机;8,三脚架。

Fig.1 Hyperspectral imaging monitoring system 1, Leaf sample; 2, 3, Light source; 4, Storage platform; 5, The Specim IQ hyperspectral camera; 6, Transmission data line; 7, Computer; 8, Tripod.

图2 投票回归器算法流程图

Fig.2 Voting regression algorithm flow chart

图3 紫丁香叶片样本原始光谱曲线Rraw(A)和预处理后光谱曲线RSG-SD(B)

Fig.4 original spectral curve Rraw (A) and pretreated spectral curve RSG-SD (B) of syringa oblata leaves

表1 样本SPAD值统计与划分

Table 1 Statistics and division of samples SPAD value

样本集 Samples set	样本数 Samples number	最大值 Maximum	最小值 Minimum	平均值 Mean	标准差 Standard derivation
总样本Overall	200	44.3	18.3	31.55	7.085 2
建模集Modeling set	160	44.3	18.3	30.09	7.167 7
验证集Validation set	40	43.6	19.7	28.87	6.770 7

图4 CA(A)、CARS(B)、UVE(C)和MWPLS(D)算法筛选变量分布图

Fig.4 Distribution map of variables selected by CA(A), CARS(B), UVE(C) and MWPLS(D) method

图5 RF运行结果每个波段被选择概率(A)和波长筛选结果(B)

Fig.5 Results of of RF, Probability of each wavelength selected (A) and wavelength selected results (B)

表2 不同变量筛选方法PLSR建模精度

Table 2 Accuracies of PLSR modeling with different variable selection methods

光谱 Spectrum	筛选方法 Selection method	波段数量 Bands number	主成分数 Principal component	建模集 $R C 2$	验证集 $R V 2$
R_raw	FULL	204	10	0.846 6	0.896 9
R_SG-SD		204	7	0.865 2	0.913 8
R_raw	CA	31	8	0.857 9	0.893 6
R_SG-SD		31	18	0.885 7	0.911 0
R_raw	RF	49	10	0.899 7	0.921 1
R_SG-SD		35	10	0.944 2	0.951 4
R_raw	CARS	48	14	0.857 7	0.924 3
R_SG-SD		29	9	0.910 6	0.928 8
R_raw	UVE	25	15	0.854 2	0.872 7
R_SG-SD		40	7	0.906 4	0.927 8
R_raw	MWPLS	190	8	0.856 9	0.897 0
R_SG-SD		190	5	0.905 2	0.927 7

表2 不同变量筛选方法PLSR建模精度

Table 2 Accuracies of PLSR modeling with different variable selection methods

光谱 Spectrum	筛选方法 Selection method	波段数量 Bands number	主成分数 Principal component	建模集 $R C 2$	验证集 $R V 2$
R_raw	FULL	204	10	0.846 6	0.896 9
R_SG-SD		204	7	0.865 2	0.913 8
R_raw	CA	31	8	0.857 9	0.893 6
R_SG-SD		31	18	0.885 7	0.911 0
R_raw	RF	49	10	0.899 7	0.921 1
R_SG-SD		35	10	0.944 2	0.951 4
R_raw	CARS	48	14	0.857 7	0.924 3
R_SG-SD		29	9	0.910 6	0.928 8
R_raw	UVE	25	15	0.854 2	0.872 7
R_SG-SD		40	7	0.906 4	0.927 8
R_raw	MWPLS	190	8	0.856 9	0.897 0
R_SG-SD		190	5	0.905 2	0.927 7

图6 RF-PLSR模型预测值和实测值散点图

Fig.6 Scatter diagram of predicted and measured values for the RF-PLSR model

表3 不同变量筛选方法VR建模精度

Table 3 Accuracies of VR modeling with different variable selection methods

预处理 Spectrum	筛选方法 Selection method	波段数量 Bands number	建模集 $R C 2$	验证集 $R V 2$
R_raw	FULL	204	0.919 3	0.952 5
R_SG-SD		204	0.949 7	0.953 4
R_raw	CA	31	0.853 1	0.899 8
R_SG-SD		31	0.949 3	0.942 0
R_raw	RF	49	0.919 3	0.965 8
R_SG-SD		35	0.952 9	0.964 3
R_raw	CARS	48	0.866 4	0.957 8
R_SG-SD		29	0.940 5	0.954 6
R_raw	UVE	25	0.881 3	0.947 7
R_SG-SD		40	0.944 8	0.960 3
R_raw	MWPLS	190	0.926 1	0.953 2
R_SG-SD		190	0.945 9	0.948 6

表3 不同变量筛选方法VR建模精度

Table 3 Accuracies of VR modeling with different variable selection methods

预处理 Spectrum	筛选方法 Selection method	波段数量 Bands number	建模集 $R C 2$	验证集 $R V 2$
R_raw	FULL	204	0.919 3	0.952 5
R_SG-SD		204	0.949 7	0.953 4
R_raw	CA	31	0.853 1	0.899 8
R_SG-SD		31	0.949 3	0.942 0
R_raw	RF	49	0.919 3	0.965 8
R_SG-SD		35	0.952 9	0.964 3
R_raw	CARS	48	0.866 4	0.957 8
R_SG-SD		29	0.940 5	0.954 6
R_raw	UVE	25	0.881 3	0.947 7
R_SG-SD		40	0.944 8	0.960 3
R_raw	MWPLS	190	0.926 1	0.953 2
R_SG-SD		190	0.945 9	0.948 6

图7 RF-VR模型预测值和实测值散点图

Fig.7 Scatter plot of predicted and measured values for the RF-VR model

图8 丁香叶片叶绿素分布图

Fig.8 Distribution of chlorophyll in Syringa oblata leaves

参考文献 27

[1]	XUE L H, YANG L Z. Deriving leaf chlorophyll content of green-leafy vegetables from hyperspectral reflectance[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2009, 64(1):97-106. DOI URL
[2]	LIMANTARA L, DETTLING M, INDRAWATI R, et al. Analysis on the chlorophyll content of commercial green leafy vegetables[J]. Procedia Chemistry, 2015, 14:225-231. DOI URL
[3]	董晶晶, 王力, 牛铮. 植被冠层水平叶绿素含量的高光谱估测[J]. 光谱学与光谱分析, 2009, 29(11):3003-3006.
	DONG J J, WANG L, NIU Z. Estimation of canopy chlorophyll content using hyperspectral data[J]. Spectroscopy and Spectral Analysis, 2009, 29(11):3003-3006.(in Chinese with English abstract)
[4]	关锦毅, 郝再彬, 张达, 等. 叶绿素提取与检测及生物学功效的研究进展[J]. 东北农业大学学报, 2009, 40(12):130-134.
	GUAN J Y, HAO Z B, ZHANG D, et al. A review on the extraction, detection and biological function of chlorophyll[J]. Journal of Northeast Agricultural University, 2009, 40(12):130-134.(in Chinese with English abstract)
[5]	张兵. 高光谱图像处理与信息提取前沿[J]. 遥感学报, 2016, 20(5):1062-1090.
	ZHANG B. Advancement of hyperspectral image processing and information extraction[J]. Journal of Remote Sensing, 2016, 20(5):1062-1090.(in Chinese with English abstract)
[6]	杨峰, 范亚民, 李建龙, 等. 高光谱数据估测稻麦叶面积指数和叶绿素密度[J]. 农业工程学报, 2010, 26(2):237-243.
	YANG F, FAN Y M, LI J L, et al. Estimating LAI and CCD of rice and wheat using hyperspectral remote sensing data[J]. Transactions of the Chinese Society of Agricultural Engineering, 2010, 26(2):237-243.(in Chinese with English abstract)
[7]	唐延林, 黄敬峰, 王秀珍, 等. 玉米叶片高光谱特征及与叶绿素、类胡萝卜素相关性的研究[J]. 玉米科学, 2008, 16(2):71-76.
	TANG Y L, HUANG J F, WANG X Z, et al. Study on hyper spectral characteristics of corn leaves and their correlation to chrolophyll and carotenoid[J]. Journal of Maize Sciences, 2008, 16(2):71-76.(in Chinese with English abstract)
[8]	王强, 易秋香, 包安明, 等. 基于高光谱反射率的棉花冠层叶绿素密度估算[J]. 农业工程学报, 2012, 28(15):125-132.
	WANG Q, YI Q X, BAO A M, et al. Estimating chlorophyll density of cotton canopy by hyperspectral reflectance[J]. Transactions of the Chinese Society of Agricultural Engineering, 2012, 28(15):125-132.(in Chinese with English abstract)
[9]	朱亚星, 于雷, 洪永胜, 等. 土壤有机质高光谱特征与波长变量优选方法[J]. 中国农业科学, 2017, 50(22):4325-4337.
	ZHU Y X, YU L, HONG Y S, et al. Hyperspectral features and wavelength variables selection methods of soil organic matter[J]. Scientia Agricultura Sinica, 2017, 50(22):4325-4337.(in Chinese with English abstract)
[10]	LI Y K, JING J. A consensus PLS method based on diverse wavelength variables models for analysis of near-infrared spectra[J]. Chemometrics and Intelligent Laboratory Systems, 2014, 130:45-49. DOI URL
[11]	赵艳茹, 余克强, 李晓丽, 等. 基于高光谱成像的南瓜叶片叶绿素分布可视化研究[J]. 光谱学与光谱分析, 2014, 34(5):1378-1382.
	ZHAO Y R, YU K Q, LI X L, et al. Study on SPAD visualization of pumpkin leaves based on hyperspectral imaging technology[J]. Spectroscopy and Spectral Analysis, 2014, 34(5):1378-1382.(in Chinese with English abstract)
[12]	邵园园, 王永贤, 玄冠涛, 等. 基于高光谱成像的肥城桃品质可视化分析与成熟度检测[J]. 农业机械学报, 2020, 51(8):344-350.
	SHAO Y Y, WANG Y X, XUAN G T, et al. Visual detection of SSC and firmness and maturity prediction for Feicheng peach by using hyperspectral imaging[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(8):344-350.(in Chinese with English abstract)
[13]	龙燕, 连雅茹, 马敏娟, 等. 基于高光谱技术和改进型区间随机蛙跳算法的番茄硬度检测[J]. 农业工程学报, 2019, 35(13):270-276.
	LONG Y, LIAN Y R, MA M J, et al. Detection of tomato hardness based on hyperspectral technology and modified interval random frog algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(13):270-276.(in Chinese with English abstract)
[14]	孙红, 刘宁, 吴莉, 等. 高光谱成像的马铃薯叶片含水率分布可视化[J]. 光谱学与光谱分析, 2019, 39(3):910-916.
	SUN H, LIU N, WU L, et al. Visualization of water content distribution in potato leaves based on hyperspectral image[J]. Spectroscopy and Spectral Analysis, 2019, 39(3):910-916.(in Chinese with English abstract)
[15]	孙红, 郑涛, 刘宁, 等. 高光谱图像检测马铃薯植株叶绿素含量垂直分布[J]. 农业工程学报, 2018, 34(1):149-156.
	SUN H, ZHENG T, LIU N, et al. Vertical distribution of chlorophyll in potato plants based on hyperspectral imaging[J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34(1):149-156.(in Chinese with English abstract)
[16]	GATIUS F, MIRALBÉS C, DAVID C, et al. Comparison of CCA and PLS to explore and model NIR data[J]. Chemometrics and Intelligent Laboratory Systems, 2017, 164:76-82. DOI URL
[17]	KAWAMURA K, TSUJIMOTO Y, RABENARIVO M, et al. Vis-NIR spectroscopy and PLS regression with waveband selection for estimating the total C and N of paddy soils in Madagascar[J]. Remote Sensing, 2017, 9(10):1081. DOI URL
[18]	GENISHEVA Z, QUINTELAS C, MESQUITA D P, et al. New PLS analysis approach to wine volatile compounds characterization by near infrared spectroscopy (NIR)[J]. Food Chemistry, 2018, 246:172-178. DOI URL
[19]	陈凯, 朱钰. 机器学习及其相关算法综述[J]. 统计与信息论坛, 2007, 22(5):105-112.
	CHEN K, ZHU Y. A summary of machine learning and related algorithms[J]. Statistics & Information Forum, 2007, 22(5):105-112.(in Chinese with English abstract)
[20]	金秀, 朱先志, 李绍稳, 等. 基于梯度提升树的土壤速效磷高光谱回归预测方法[J]. 激光与光电子学进展, 2019, 56(13):141-150.
	JIN X, ZHU X Z, LI S W, et al. Predicting soil available phosphorus by hyperspectral regression method based on gradient boosting decision tree[J]. Laser & Optoelectronics Progress, 2019, 56(13):141-150.(in Chinese with English abstract)
[21]	刘茂成, 李志洪. 植物叶绿素光谱测定仪的原理与设计[J]. 中国农机化学报, 2017, 38(1):74-79.
	LIU M C, LI Z H. Principle and design of chlorophyll spectrometry meter for plants[J]. Journal of Chinese Agricultural Mechanization, 2017, 38(1):74-79.(in Chinese with English abstract)
[22]	第五鹏瑶, 卞希慧, 王姿方, 等. 光谱预处理方法选择研究[J]. 光谱学与光谱分析, 2019, 39(9):2800-2806.
	DIWU P Y, BIAN X H, WANG Z F, et al. Study on the selection of spectral preprocessing methods[J]. Spectroscopy and Spectral Analysis, 2019, 39(9):2800-2806.(in Chinese with English abstract)
[23]	田安红, 熊黑钢, 赵俊三, 等. 分数阶微分对盐渍土野外光谱预处理精度提升的机理分析[J]. 光谱学与光谱分析, 2019, 39(8):2495-2500.
	TIAN A H, XIONG H G, ZHAO J S, et al. Mechanism improvement for pretreatment accuracy of field spectra of saline soil using fractional differential algorithm[J]. Spectroscopy and Spectral Analysis, 2019, 39(8):2495-2500.(in Chinese with English abstract)
[24]	陈立旦, 赵艳茹. 可见-近红外光谱联合随机蛙跳算法检测生物柴油含水量[J]. 农业工程学报, 2014, 30(8):168-173.
	CHEN L D, ZHAO Y R. Measurement of water content in biodiesel using visible and near infrared spectroscopy combined with Random-Frog algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering, 2014, 30(8):168-173.(in Chinese with English abstract)
[25]	YUN Y H, LI H D, WOOD L R E, et al. An efficient method of wavelength interval selection based on random frog for multivariate spectral calibration[J]. Spectrochimica Acta. Part A, Molecular and Biomolecular Spectroscopy, 2013, 111:31-36. DOI URL
[26]	葛翔宇, 丁建丽, 王敬哲, 等. 基于竞争适应重加权采样算法耦合机器学习的土壤含水量估算[J]. 光学学报, 2018, 38(10):393-400.
	GE X Y, DING J L, WANG J Z, et al. Estimation of soil moisture content based on competitive adaptive reweighted sampling algorithm coupled with machine learning[J]. Acta Optica Sinica, 2018, 38(10):393-400.(in Chinese with English abstract)
[27]	李盛芳, 贾敏智, 董大明. 随机森林算法的水果糖分近红外光谱测量[J]. 光谱学与光谱分析, 2018, 38(6):1766-1771.
	LI S F, JIA M Z, DONG D M. Fast measurement of sugar in fruits using near infrared spectroscopy combined with random forest algorithm[J]. Spectroscopy and Spectral Analysis, 2018, 38(6):1766-1771.(in Chinese with English abstract)

基于RF-VR的紫丁香叶片叶绿素含量高光谱反演

Hyperspectral retrieval for chlorophyll contents of Syringa oblata leaves based on RF-VR

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 27

相关文章 15

编辑推荐

Metrics

本文评价

[1]	王佳, 慕瑞瑞, 杨乔乔, 刘伟, 张月荷, 康建宏. 滴灌水肥一体化下施钾量对宁夏春玉米叶绿素荧光特性与产量的影响[J]. 浙江农业学报, 2021, 33(8): 1347-1357.
[2]	钱家炜, 刘晓青, 张静静, 周卫红, 李建龙. 张家港市农田土壤重金属含量高光谱遥感监测模型构建[J]. 浙江农业学报, 2020, 32(8): 1437-1445.
[3]	陆锡昆, 罗亚辉, 蒋蘋, 胡文武. 基于高光谱的油茶籽含水量检测方法[J]. 浙江农业学报, 2020, 32(7): 1302-1310.
[4]	禹文杰, 王彩霞, 乔芦, 王松磊, 贺晓光. 基于高光谱成像技术的泾源黄牛肉色度PLSR预测模型构建[J]. 浙江农业学报, 2020, 32(3): 527-533.
[5]	王念一, 于丰华, 许童羽, 杜文, 郭忠辉, 张国圣. 基于机器学习的粳稻叶片叶绿素含量高光谱反演建模[J]. 浙江农业学报, 2020, 32(2): 359-366.
[6]	宋春雨, 甘淑, 袁希平, 胡琳, 李雁, 闫馨方. 滇中典型乔木植被高光谱特征[J]. 浙江农业学报, 2020, 32(11): 1978-1986.
[7]	方向, 金秀, 朱娟娟, 李绍稳. 基于可见-近红外光谱预处理建模的土壤速效氮含量预测[J]. 浙江农业学报, 2019, 31(9): 1523-1530.
[8]	孟庆龙, 张艳, 尚静. 基于高光谱成像的猕猴桃表面疤痕无损识别[J]. 浙江农业学报, 2019, 31(8): 1372-1378.
[9]	方芳, 何序晨, 张志豪, 张勤, 关亚静, 胡晋, 胡伟民. 玉米自交系苗期对高温胁迫的响应机制及其抗逆性[J]. 浙江农业学报, 2019, 31(7): 1045-1056.
[10]	王美玲, 焦琳琳, 王晓红, 吴兵, 肖星星. 曹妃甸湿地典型植被光谱特征差异性分析[J]. 浙江农业学报, 2019, 31(6): 963-969.
[11]	王彦翔, 张艳, 杨成娅, 孟庆龙, 尚静. 基于深度学习的农作物病害图像识别技术进展[J]. 浙江农业学报, 2019, 31(4): 669-676.
[12]	孟幼青, 翁海勇, 岑海燕, 李红叶, 何勇. 潜伏期柑橘黄龙病宿主糖代谢及近红外光谱特征[J]. 浙江农业学报, 2019, 31(3): 428-435.
[13]	金秀, 卢杰, 傅运之, 王帅, 许高健, 李绍稳. 基于深度卷积神经网络的小麦赤霉病高光谱病症点分类方法[J]. 浙江农业学报, 2019, 31(2): 315-325.
[14]	杨红云, 周琼, 杨珺, 孙玉婷, 路艳, 殷华. 基于高光谱的水稻叶片氮素营养诊断研究[J]. 浙江农业学报, 2019, 31(10): 1575-1582.
[15]	余昌乐, 许童羽, 王洋, 于丰华. 双子叶植物叶片类胡萝卜素含量高光谱反演估算[J]. 浙江农业学报, 2018, 30(3): 393-398.