Remote Sensing Inversion Study of Relative Chlorophyll Content in Processing Tomato Based on Machine Learning

doi:10.13304/j.nykjdb.2024.0004

Abstract

Abstract:

Chlorophyll plays a key role in the process of plant photosynthesis， and the relative chlorophyll content （soil and plantan alyzer development，SPAD） is an important indicator to measure the growth status of crops. In order to construct an inversion model of SPAD value of processing tomato， this study used unmanned aerial vehicle（UAV） remote sensing technology to predict the SPAD value of canopy leaves at 4 key growth periods of processing tomato by machine learning method， and drew a visual mapping of SPAD value of processing tomato based on the optimal prediction model. The results showed that random forest （RF）， support vector machine （SVM） and back propagation（BP）neural network prediction model of SPAD values at different growth periods of processing tomato， the multispectral vegetation index as the independent variable， the RF model had the best prediction effect at the first flowering period， with determination coefficient（R² ）of 0.89 and root mean square error（RMSE） of 1.15， the SVM model was the best at the full flowering period， with R² of 0.87 and RMSE of 1.46， the SVM model was the best at the fruit setting period， with R² of 0.88 and RMSE of 1.25， and the BP neural network model was the best at the maturity period， with R² of 0.89 and RMSE of 1.07. In the full flowering period of crops， the flowering process would consume most of the nutrients inside the plant， resulting in different changes in chlorophyll content， and its prediction effect was relatively low， and the selection of appropriate models for modeling in different growth periods could achieve high-precision monitoring of chlorophyll content in processing tomato. The stability of the prediction results and verification results of SVM model in each growth period of processing tomato was better， with R² of 0.88， 0.87， 0.88， 0.85 and RMSE of 0.95， 1.46， 1.25，1.91， respectively， so the SPAD value of processing tomato was visualized and mapped at each growth stage based on the optimal SVM model to realize the dynamic monitoring of chlorophyll in processing tomato. Above results could be used to quickly and efficiently estimate the relative chlorophyll content of processing tomatoes， and provided decision-making information and technical support for the precision agriculture management of processing tomatoes.

Key words: UAV remote sensing, multispectral, inversion, SPAD value, dynamic monitoring, processing tomato

摘要：

叶绿素在植物光合作用过程中起关键作用，叶绿素相对含量（soil and plantan alyzer development，SPAD）是衡量农作物生长状况的重要指标。为构建加工番茄SPAD值的反演模型，基于无人机遥感技术，通过机器学习方法预测加工番茄4个关键生育期冠层叶片SPAD值，并基于最优预测模型绘制加工番茄SPAD值可视化制图。结果表明，在加工番茄不同生育时期SPAD值的随机森林（random forest，RF）、支持向量机（support vector machine，SVM）、BP（back propagation）神经网络3种预测模型中，以多光谱植被指数为自变量，在始花期RF模型预测效果最优，决定系数（R²）为0.89，均方根误差（root mean square error，RMSE）为1.15；在盛花期SVM模型预测效果最优，R²为0.87，RMSE为1.46；在坐果期SVM模型预测效果最优，R²为0.88，RMSE为1.25；在成熟期BP神经网络模型预测效果最优，R²为0.89，RMSE为1.07。在作物盛花期，开花过程会消耗植株内部的大部分营养，导致叶绿素含量发生不同变化，其预测效果相对较低，在不同生育期选择合适的模型进行建模，可以实现对加工番茄叶绿素含量较高精度的监测；SVM模型在加工番茄各生育期的预测结果和验证结果稳定性都较优，R² 分别为0.88、0.87、0.88、0.85，RMSE分别为0.95、1.46、1.25、1.91，因此基于最优SVM模型生成加工番茄每个生育期的SPAD值可视化制图，实现对加工番茄叶绿素的动态监测。研究结果可用于快速、高效地估算加工番茄叶绿素相对含量，为加工番茄精准农业管理提供决策信息和技术支持。

关键词: 无人机遥感, 多光谱, 反演, SPAD值, 动态监测, 加工番茄

CLC Number:

S127

Mingjun JIANG, Yanmin FAN, Hongqi WU, Hao ZHANG, Zhuo LIU, Dejun WANG. Remote Sensing Inversion Study of Relative Chlorophyll Content in Processing Tomato Based on Machine Learning[J]. Journal of Agricultural Science and Technology, 2025, 27(8): 89-99.

姜明君, 范燕敏, 武红旗, 张浩, 刘卓, 王德俊. 基于机器学习的加工番茄叶绿素相对含量遥感反演研究[J]. 中国农业科技导报, 2025, 27(8): 89-99.

Figures/Tables 8

Table 1 Formula for calculating the multispectral vegetation index

变量参数 Variable parameter	计算公式 Calculation formula	参考文献
归一化植被指数 Normalized differential vegetation index（NDVI）	$N D V I = (N I R - R) / (N I R + R)$	［16］
土壤调节植被指数 Soil-adjusted vegetation index（SAVI）	$S A V I = (N I R - R) / 1.5 (N I R + R + 0.5)$	［16］
优化土壤调节植被指数 Optimization soil-adjusted vegetation index（OSAVI）	$O S A V I = (N I R - R) / (N I R + R + 0.16)$	［17］
改良土壤调整植被指数 Modified soil-adjusted vegetation index（MSAVI）	$M S A V I = [2 N I R + 1 -$ $(2 N I R + 1) 2 - 8 (N I R - R)$ ］/2	［17］
绿光归一化植被指数 Green normalized difference vegetation index（GNDVI）	$G N D V I = (N I R - G) / (N I R + G)$	［17］
叶绿素植被指数 Chlorophyll vegetation index（CVI）	CVI=（NIR/G）（R/G）	［17］
差值植被指数 Difference vegetation index（DVI）	$D V I = N I R - R$	［17］
重归一化植被指数 Renormalized difference vegetation index（RDVI）	RDVI=（NIR $-$ R）/ $(N I R + R)$	［18］
大气阻抗植被指数 Atmospherically resistant vegetation index（ARVI）	$A R V I = N I R - (2 R - B) / N I R + (2 R - B)$	［18］
比值植被指数 Ratio vegetation index（RVI）	RVI=NIR/R	［18］

Table 1 Formula for calculating the multispectral vegetation index

变量参数 Variable parameter	计算公式 Calculation formula	参考文献
归一化植被指数 Normalized differential vegetation index（NDVI）	$N D V I = (N I R - R) / (N I R + R)$	［16］
土壤调节植被指数 Soil-adjusted vegetation index（SAVI）	$S A V I = (N I R - R) / 1.5 (N I R + R + 0.5)$	［16］
优化土壤调节植被指数 Optimization soil-adjusted vegetation index（OSAVI）	$O S A V I = (N I R - R) / (N I R + R + 0.16)$	［17］
改良土壤调整植被指数 Modified soil-adjusted vegetation index（MSAVI）	$M S A V I = [2 N I R + 1 -$ $(2 N I R + 1) 2 - 8 (N I R - R)$ ］/2	［17］
绿光归一化植被指数 Green normalized difference vegetation index（GNDVI）	$G N D V I = (N I R - G) / (N I R + G)$	［17］
叶绿素植被指数 Chlorophyll vegetation index（CVI）	CVI=（NIR/G）（R/G）	［17］
差值植被指数 Difference vegetation index（DVI）	$D V I = N I R - R$	［17］
重归一化植被指数 Renormalized difference vegetation index（RDVI）	RDVI=（NIR $-$ R）/ $(N I R + R)$	［18］
大气阻抗植被指数 Atmospherically resistant vegetation index（ARVI）	$A R V I = N I R - (2 R - B) / N I R + (2 R - B)$	［18］
比值植被指数 Ratio vegetation index（RVI）	RVI=NIR/R	［18］

Fig. 1 SPAD value box plot of processed tomato

Fig. 2 Correlation between spectral variables and SPAD content in processed tomatoesNote：SPAD—Soil and plantan alyzer development；DVI—Difference vegetation index；GNDVI—Green normalized difference vegetation index；MSAVI—Modified soil-adjusted vegetation index；NDVI—Normalized differential vegetation index；OSAVI—Optimization soil-adjusted vegetation index；RVI—Ratio vegetation index；RDVI—Renormalized difference vegetation index；SAVI—Soil-adjusted vegetation index；CVI—Chlorophyll vegetation index；ARVI—Atmospherically resistant vegetation index.

Table 2 Results for modeling and validating SPAD values for different estimation models

模型 Model	生育时期 Growth period	建模集Modeling			验证集Vertification
模型 Model	生育时期 Growth period	决定系数 R²	均方根误差 RMSE	平均相对误差 MRE/%	决定系数 R²	均方根误差 RMSE	平均相对误差 MRE/%
随机森林 RF	始花期 Anthesis period	0.89	1.15	2.11	0.90	1.01	2.00
	盛花期 Blooming period	0.86	1.33	2.35	0.80	1.80	3.45
	坐果期 Fruiting period	0.87	1.30	2.04	0.74	1.99	3.79
	成熟期 Maturation period	0.83	1.33	3.21	0.71	2.50	4.26
支持向量机 SVM	始花期 Anthesis period	0.88	0.95	1.27	0.79	1.41	1.20
	盛花期 Blooming period	0.87	1.46	2.08	0.82	1.83	1.81
	坐果期 Fruiting period	0.88	1.25	1.94	0.88	1.31	1.04
	成熟期 Maturation period	0.85	1.91	2.65	0.75	2.34	3.35
BP神经网络 BP neural network	始花期 Anthesis period	0.81	1.56	3.45	0.76	1.56	3.16
	盛花期 Blooming period	0.68	2.35	4.21	0.67	2.35	4.81
	坐果期 Fruiting period	0.69	2.43	4.78	0.65	2.43	4.97
	成熟期 Maturation period	0.89	1.07	2.53	0.80	1.25	2.17

Fig. 3 Verification of the inversion accuracy of RF model of SPAD values of processing tomatoes at different growth periods

Fig. 5 Verification of the inversion accuracy of BP of SPAD values of processing tomatoes at different growth periods

Fig. 6 Predicted SPAD values at different stages of processing tomatoesA： Anthesis period； B： Blooming period； C： Fruiting period； D： Maturation period

Fig. 4 Verification of the inversion accuracy of SVM of SPAD values of processing tomatoes at different growth periods

References 21

[1]	ZHANG H C, GE Y F, XIE X Y, et al.. High throughput analysis of leaf chlorophyll content in Sorghum using RGB, hyperspectral,and fluorescence imaging and sensor fusion [J/OL].Plant Methods,2022,18(1):60 [2024-12-19]. .
[2]	郭建鹏. 玉米叶倾角反演及其与叶绿素含量的关系研究[D]. 呼和浩特:内蒙古师范大学,2024.
	GUO J P. Retrieval of leaf inclination angle and its relationship with chlorophyll content in maize [D]. Hohhot: Inner Mongolia Normal University, 2024.
[3]	JIANG C Y， JOHKAN M， HOHJO M， et al.. A correlation analysis on chlorophyll content and SPAD value in tomato leaves ［J］. HortResearch, 2017, 71(71): 37-42.
[4]	PAJARES G. Overview and current status of remote sensing applications based on unmanned aerial vehicles (UAVs) [J].Photogramm. Eng. Remote. Sens., 2015, 81(4): 281-329.
[5]	HUANG Y B, REDDY K N, FLETCHER R S, et al.. UAV low-altitude remote sensing for precision weed management [J]. Weed Technol., 2018, 32(1): 2-6.
[6]	MAES W H, STEPPE K.Perspectives for remote sensing with unmanned aerial vehicles in precision agriculture [J]. Trends Plant Sci., 2019, 24(2): 152-164.
[7]	高亚利,王振锡,连玲,等.截形叶螨危害下枣叶片叶绿素含量高光谱估算模型[J].西北农业学报, 2020, 29(4): 613-621.
	GAO Y L, WANG Z X, LIAN L, et al.. Hyperspectral estimation model of jujube leaves chlorophyll contents under different damages of Tetranychus truncatus ehara [J]. Acta Agric. Bor-Occid. Sin., 2020, 29(4): 613-621.
[8]	ZHANG H, SHI P, CRUCIL G, et al.. Evaluating the capability of a UAV-borne spectrometer for soil organic carbon mapping in bare croplands [J]. Land Degrad. Dev., 2021, 32(15): 4375-4389.
[9]	毛智慧,邓磊,孙杰,等.无人机多光谱遥感在玉米冠层叶绿素预测中的应用研究[J].光谱学与光谱分析, 2018, 38(9):2923-2931.
	MAO Z H, DENG L, SUN J, et al.. Research on the application of UAV multispectral remote sensing in the maize chlorophyll prediction [J]. Spectrosc. Spectr. Anal., 2018, 38(9): 2923-2931.
[10]	刘小辉.基于无人机影像的小麦叶绿素含量及产量定量反演研究[D].合肥:安徽大学,2019.
	LIU X H. Inversion of wheat chlorophyll content and yield based on unmanned aerial vehicle images [D]. Hefei: Anhui University, 2019.
[11]	陈浩,冯浩,杨祯婷,等.基于无人机多光谱遥感的夏玉米冠层叶绿素含量估计[J].排灌机械工程学报, 2021, 39(6):622-629.
	CHEN H, FENG H, YANG Z T, et al..Estimation of chlorophyll content of summer maize canopy based on UAV multispectral remote sensing [J]. J. Drain. Irrig. Mach. Eng., 2021, 39(6): 622-629.
[12]	王丽爱,马昌,周旭东,等.基于随机森林回归算法的小麦叶片SPAD值遥感估算[J].农业机械学报, 2015, 46(1): 259-265.
	WANG L A, MA C, ZHOU X D, et al.. Estimation of wheat leaf SPAD value using RF algorithmic model and remote sensing data [J]. Trans. Chin. Soc. Agric. Mach., 2015, 46(1): 259-265.
[13]	HUANG Y Y, MA Q M, WU X M, et al.. Estimation of chlorophyll content in Brassica napus based on unmanned aerial vehicle images [J]. Oil Crop Sci., 2022, 7(3): 149-155.
[14]	JIANG J L, JOHANSEN K, STANSCHEWSKI C S, et al.. Phenotyping a diversity panel of quinoa using UAV-retrieved leaf area index, SPAD-based chlorophyll and a random forest approach [J]. Precis. Agric., 2022, 23(3): 961-983.
[15]	YANG S Q, HU L, WU H B, et al.. Integration of crop growth model and random forest for winter wheat yield estimation from UAV hyperspectral imagery [J]. IEEE J. Sel. Top. Appl. Earth Obs. Remote. Sens., 2021, 14: 6253-6269.
[16]	ZHANG Y, SU Z B, SHEN W Z, et al.. Remote monitoring of heading rice growing and nitrogen content based on UAV images [J].Int. J. Smart Home, 2016, 10(7):103-114.
[17]	彭晓伟,张爱军,杨晓楠,等.谷子叶绿素含量高光谱特征分析及其反演模型构建[J].干旱地区农业研究,2022,40(2):69-77.
	PENG X W, ZHANG A J, YANG X N, et al.. Hyperspectral characteristics and remote sensing inversion model of chlorophyll content of millet [J]. Agric. Res. Arid Areas, 2022, 40(2): 69-77.
[18]	田军仓,杨振峰,冯克鹏,等.基于无人机多光谱影像的番茄冠层SPAD预测研究[J].农业机械学报, 2020, 51(8): 178-188.
	TIAN J C, YANG Z F, FENG K P, et al.. Prediction of tomato canopy SPAD based on UAV multispectral image [J]. Trans.Chin. Soc. Agric. Mach., 2020, 51(8): 178-188.
[19]	岳云开,陈建福,赵亮,等.基于无人机多光谱遥感的苎麻叶绿素含量反演[J].山东农业科学, 2023, 55(7): 152-158.
	YUE Y K, CHEN J F, ZHAO L, et al.. Inversion of chlorophyll content in ramie based on UAV multispectral remote sensing [J].Shandong Agric. Sci., 2023, 55(7): 152-158.
[20]	FENG A J, ZHOU J F, VORIES E D, et al.. Yield estimation in cotton using UAV-based multi-sensor imagery [J]. Biosyst. Eng., 2020, 193: 101-114.
[21]	梁栋,杨勤英,黄文江,等.基于小波变换与支持向量机回归的冬小麦叶面积指数估算[J].红外与激光工程, 2015, 44(1):335-340.
	LIANG D, YANG Q Y, HUANG W J, et al.. Estimation of leaf area index based on wavelet transform and support vector machine regression in winter wheat [J]. Infrared Laser Eng.,2015, 44(1): 335-340.