Fruit Growth Modelling Based on Multi-Methods - A Case Study of Apple in Zhaotong, Yunnan
SUN Qing,1, ZHAO YanXia,1, CHENG JinXin2, ZENG TingYu3, ZHANG Yi11Chinese Academy of Meteorological Sciences, Beijing 100081 2Yunnan Climate Center, Kunming 650000 3Agricultural Meteorological Experimental Station of Zhaotong, Zhaotong 657000, Yunnan
Abstract 【Objective】 Meteorological factors are closely related to fruit diameter during growth process, but this relationship between them tends to be non-linear and non-stationary, thus making it hard to monitor the fruit and trunk diameter continuously. Comparing the simulation capabilities of various growth models for fruit diameter could provide scientific support for fruit growth monitoring and predicting, timely irrigation and fertilization, and the regulation of growth environment. 【Method】 Taking apples in Zhaotong, Yunan Province as an example, this study first analyzed the characteristics of diameter change during apple growth in 2019 and 2020 and its relationship with environmental and climate factors. Subsequently, a deep learning method of Long Short-Term Memory (LSTM) model was adopted to simulate and predict the fruit diameter by integrating these factors, which was evaluated with the multi-linear regression (MLR) model and machine learning methods including Decision Tree (DT) and Random Forests (RF) using three sampling methods. 【Result】 The apple diameter had obvious diurnal cycle characteristics, which shrunk in the daytime and expanded in the nighttime. The maximum diameter was in the morning, while the minimum diameter was near the sunset. The growth rate of apple diameter was higher in the early growth period than near mature. The hourly and daily mean apple diameters were moderately or highly-positive correlated with soil temperature and soil moisture, while there was a highly-negative correlation with UVI. The daily mean increase (FMDG), daily increase (FDG), and maximum daily shrinkage (MDFS) of apple diameter had a weak negative correlation with 60 cm soil temperature as well as 20 and 40 cm soil moisture (-0.5≤R<-0.3). The simulation accuracy of the LSTM model was significantly higher than that of MLR, DT and RF model. The correlation coefficient (R) of LSTM model increased (3% -20%) compared with MLR, and the RMSE and MAE were approximately decreased by 50%-75%. The machine learning methods showed relatively poor performance in apple diameter simulation and might have overfitting problems. 【Conclusion】 Compared to statistics and machine learning approaches, the LSTM model demonstrated higher accuracy and robust performance because of the incapability of considering the complex non-linear correlations in the fruit growth simulation. Keywords:apple diameter;growth model;meteorological factors;deep learning;Long Short-Term Memory (LSTM)
PDF (4762KB)元数据多维度评价相关文章导出EndNote|Ris|Bibtex收藏本文 本文引用格式 孙擎, 赵艳霞, 程晋昕, 曾厅余, 张祎. 基于多种算法的果树果实生长模型研究—以云南昭通苹果为例. 中国农业科学, 2021, 54(17): 3737-3751 doi:10.3864/j.issn.0578-1752.2021.17.015 SUN Qing, ZHAO YanXia, CHENG JinXin, ZENG TingYu, ZHANG Yi. Fruit Growth Modelling Based on Multi-Methods - A Case Study of Apple in Zhaotong, Yunnan. Scientia Acricultura Sinica, 2021, 54(17): 3737-3751 doi:10.3864/j.issn.0578-1752.2021.17.015
开放科学(资源服务)标识码(OSID):
0 引言
【研究意义】云南昭通市是云南苹果种植的主要区域,也是当地农民主要的经济收入来源之一。2019年,昭通苹果种植规模达4.51万hm2,总产量65万 t,总产值60亿元[1]。云南地形复杂,呈立体气候特征,气象灾害种类多、发生频率高、危害强度大[2],对苹果产量造成较大影响,亟需在模拟果实生长规律的同时,建立气候因子对苹果果实生长的影响模型。随着物联网在农业上的广泛应用,积累了大量的观测数据,为尝试探索应用新的技术和方法来模拟果实生长变化带来了机会。本研究尝试提出模拟苹果果实直径变化的一种新方法,以弥补目前研究方法的不足。研究结果可精确模拟和预测苹果果实直径,对适时进行果树灾害防御、灌溉施肥、生长环境调控等提供重要参考和应用价值。【前人研究进展】果树果实直径每日膨胀和收缩量是果树生长的重要指标,与作物生长环境如温度、日照、土壤水分等环境因子密切相关[3]。大部分关于果树生长的研究主要针对不同灌溉量[4]、施肥量[5]对果实生长发育、品质、直径等的影响,或茎干微变的研究[6,7]。对苹果果实生长的研究来说,主要是关于不同施肥量对苹果果树生长及果实产量的影响[8],或不同干旱程度对苹果果实、茎干直径和叶片水势的影响[9],径流与环境因子的关系 [10]。传统的果实生长发育模型包括作物生理模型、多元线性回归(multi-linear regression,MLR)、移动平均模型(moving average model,MA)、自动回归模型(autoregressive moving average model,ARMA),或基于机器学习方法如BP神经网络、人工神经网络(artificial neural network,ANN)、支持向量机(support vector machines,SVM)等。例如,LI等[11]使用作物生理模型模拟了苹果果实的生长发育情况;夏桂敏等[10]建立了苹果果树径流与环境因子的多元回归模型;张海辉等[12]使用机器学习方法对苹果霉心透射光谱进行研究,结果表明机器学习方法在修正果实直径对透射光谱的影响方面表现出良好的精度;张彪等[13]使用BP神经网络评价了苹果制干的适宜性。但传统作物生理模型、统计模型或机器学习方法存在一定局限,如模型过于简单导致模拟精度较低,或将动态时间建模问题转化为静态空间建模研究,忽略了过去时间的输入对于预测的影响[14]。部分研究使用了深度学习中的递归神经网络(recurrent neural network,RNN),表现出良好的精度[15],但RNN在训练过程中存在梯度消失和梯度爆炸等缺陷。长短期记忆模型(long short-term memory,LSTM)作为一种特殊的RNN模型可以有效规避上述问题,特别是LSTM在时间序列预测方面的优良特性,促使越来越多的研究者将LSTM模型应用在各个方面,如面部识别[16]、质量控制[17]、文本识别[18]、病虫害识别[19]。【本研究切入点】对于果树生长发育中果实或茎干直径微变来说,仅有的研究主要涉及的是作物茎干微变;同时,目前关于深度学习应用在果树方面的研究很少,且主要应用在机器视觉方面[20,21],对与经济价值直接关联的果实直径的监测及模拟研究鲜见报道。【拟解决的关键问题】本研究首先分析苹果果实直径在生长过程中的变化趋势,进而分析苹果果实直径与生长过程中环境因子的相关性,最后使用多元回归模型、机器学习模型中的决策树和随机森林以及深度学习中的LSTM模型分别模拟昭通苹果果实直径在生长过程中的变化情况,并对模拟及预测结果进行对比评估,以验证模型在果实直径监测及模拟中的精度和可靠性。
果实直径模拟结果评估使用的指标为:均方根误差(root mean square error,RMSE)、相关系数(correlation coefficient,R)和均方误差(mean square error, MSE)。其中,一般认为0.9<R≤1.0为极高相关,0.7<R≤0.9为高度相关,0.5<R≤0.7为中度相关,0.3<R≤0.5为低相关,0.0<R≤0.3为几乎不相关[26,27,28]。
$R M S E=\sqrt{\frac{\sum_{i=1}^{n}\left(S_{i}-O_{i}\right)^{2}}{n}}$ $R=\frac{\sum\nolimits_{i=1}^{n}{({{S}_{i}}-\overline{S})({{O}_{i}}-\overline{O})}}{\sqrt{\sum\nolimits_{i=1}^{n}{{{({{S}_{i}}-\overline{S})}^{2}}\sum\nolimits_{i=1}^{n}{{{({{O}_{i}}-\overline{O})}^{2}}}}}}$ $M S E=\frac{1}{n} \sum_{i=1}^{n}\left(S_{i}-O_{i}\right)^{2}$ 其中Oi为果实直径的观测值,Si为模拟的果实直径,$\bar{O}$为果实直径的平均值,$\bar{S}$为模拟的果实直径平均值。
总体来看,所有采样方法以LSTM模型模拟结果最好,MLR模型模拟结果较为稳健,使用3种采样方法对果实生长后期果实直径的模拟结果相近,但均有一定程度低估。使用LSTM模型在果实生长后期预测的结果比采样方法1和2的结果稍好,更接近观测值。采样方法2中MLR和LSTM模型对苹果果实生长中期的模拟效果较好,采样方法1和3对果实生长后期的果实直径有一定程度的低估,这可能是由于使用了2年的果实生长前期和中期数据进行训练,使用1年的果实生长后期数据进行训练所导致。LSTM模型在果实生长前期和后期的模拟效果高于其他模型,说明LSTM模型可以自动学习出果实在不同生长阶段的直径与环境气象因子的关系。有研究表明LSTM模型的RMSE比使用生理生长模型[10](physiological development time,PDT模型)会降低50%以上。
WANGS F, ZHANGD, XIAM, WANGC Y, LIUY, DENGF, YANGY Q, LANGP Y, QUANY, LIY G. The experience and effectiveness of apple industry in promoting poverty alleviation program in Zhaotong Journal of Yunnan Agricultural University (Social Science), 2020, 14(6): 36-41. (in Chinese) [本文引用: 1]
ZHOUG H, WUH B. 2016 Yunnan major natural disasters and the characteristics of the mitigation measures research Journal of Catastrophology, 2017, 32(3): 112-119. (in Chinese) [本文引用: 1]
CONEJEROW, ALARCÓNJ J, GARCÍA-ORELLANAY, ABRISQUETAJ M, TORRECILLASA. Daily sap flow and maximum daily trunk shrinkage measurements for diagnosing water stress in early maturing peach trees during the post-harvest period , 2007, 27(1): 81-88. DOI:10.1093/treephys/27.1.81URL [本文引用: 1]
CHENY J, CHENL, LOUY S, QINZ G, DONGX, MAC, MIAOY B, ZHANGC X, XUW P, WANGS P. Determination of thresholds to trigger irrigation of ‘Kyoho’ grapevine during berry development periods based on variations of shoot diameter and berry projected area Journal of Fruit Science, 2019, 36(5): 612-620. (in Chinese) [本文引用: 1]
WANGG D, XIAOY S, PENGF T, ZHANGY F, GAOH F, SUNX W, HEY. Effects of urea application combined with different amounts of nano-carbon on plant growth along with nitrogen absorption and utilization in young peach trees Scientia Agricultura Sinica, 2018, 51(24): 4700-4709. (in Chinese) [本文引用: 1]
WANGX S, MENGZ J, DUANA W, LIUZ G. Mechanism and methodology for monitoring tomato water content based on stem diameter variation Transactions of the Chinese Society of Agricultural Engineering, 2010, 26(12): 107-113. (in Chinese) [本文引用: 2]
LINL, TANGY, YANW L, ZHANGJ T, JIZ S. Overview of precise irrigation scheduling based on stem diameter variations Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(5): 1-13. (in Chinese) [本文引用: 1]
HEQ Q, GUOX H, YANGK, SHIM J, MAJ J, SUNX H, LEIT, LIUY W. Effects of phosphorus and potassium fertilizer on physiological growth and yield of dwarf rootstock apple trees under drip irrigation Water Saving Irrigation, 2019(8): 24-27. (in Chinese) [本文引用: 1]
GAOM W, NIEP X, SUNY, JIANGY M, ZHANGJ C, ZHANGJ X. Response of water status indicators in apple saplings to drought stress under shelter from rain Chinese Journal of Applied Ecology, 2016, 27(8): 2459-2466. (in Chinese) [本文引用: 1]
XIAG M, SUNY Y, WANGW Z, WUQ, CHID C. The characteristics of sap flow of Hanfu apple trees and its response to environmental factors Scientia Agricultura Sinica, 2019, 52(4): 701-714. (in Chinese) [本文引用: 3]
LIM, CHENM X, ZHANGY, FUC X, XINGB, LIW Y, QIANJ P, LIS, WANGH, FANX D, YANY J, WANGY A, YANGX T. Apple fruit diameter and length estimation by using the thermal and sunshine hours approach and its application to the digital orchard management information system , 2015, 10(4): e0120124. DOI:10.1371/journal.pone.0120124URL [本文引用: 1]
ZHANGH H, TIANS J, MAM J, ZHAOJ, ZHANGJ H, ZHANGZ J. Detection method of moldy core in apples using modified transmission spectrum based on size of fruit Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(1): 313-320. (in Chinese) [本文引用: 1]
ZHANGB, LIUX, BIJ F, WUX Y, JINX, LIX, LIX. Suitability evaluation of apple for chips-processing based on BP artificial neural network Scientia Agricultura Sinica, 2019, 52(1): 129-142. (in Chinese) [本文引用: 1]
CHENS Y. Study on the equivalent modeling of wind power farm based on LSTM multi-type wind turbine generators Journal of Shandong Agricultural University (Natural Science Edition), 2020, 51(2): 294-297. (in Chinese) [本文引用: 1]
WANGZ W. Use of deep learning and fuzzy decision making in automated sorting of apple [D]. Harbin: Harbin Institute of Technology, 2019. (in Chinese) [本文引用: 1]
XUEW D, CHAIY, LIQ G, HONGY Q, ZHENGG F. Short-term relay quality prediction algorithm based on long and short-term memory , 2018, 5(4): 46-54. [本文引用: 1]
LIP, LIX, PANH, KHYAMO, NOOR-A-RAHIMM. Text-based indoor place recognition with deep neural network , 2020, 390: 239-247. DOI:10.1016/j.neucom.2019.02.065URL [本文引用: 1]
SHENL Y, JIANGH Y, HUB, XIEY C. A study on joint entity recognition and relation extraction for rice diseases pests weeds and drugs Journal of Nanjing Agricultural University, 2020, 43(6): 1151-1161. (in Chinese) [本文引用: 1]
ZHENGY M, HUANGG Y, LUB B, WANGY M. An apple leaf image segmentation algorithm using deep learning and guided filtering Journal of China Agricultural University, 2020, 25(9): 112-120. (in Chinese) [本文引用: 1]
QUZ L. Application of deep learning network in recognition of apple surface lesions [D]. Nanjing: Nanjing University of Posts and Telecommunications, 2017. (in Chinese) [本文引用: 1]
MTIBAAF, NGUYENK K, AZAMM, PAPACHRISTOUA, VENNEJ S, CHERIETM. LSTM-based indoor air temperature prediction framework for HVAC systems in smart buildings , 2020, 32(23): 17569-17585. DOI:10.1007/s00521-020-04926-3URL [本文引用: 1]
BLANQUEROR, CARRIZOSAE, MOLERO-RIOC, MORALESD R. Sparsity in optimal randomized classification trees 2020, 284(1): 255-272. [本文引用: 1]
DEANDRADE C F, DELGADOR C, BARBOSAM L, TEODOROP E, DA SILVAJ C A, WANDERLEYH S, CAPRISTO-SILVAG F. Fire regime in Southern Brazil driven by atmospheric variation and vegetation cover , 2020, 295: 108194. DOI:10.1016/j.agrformet.2020.108194URL [本文引用: 1]
MUKAKAM M. Statics corner: A guide to appropriate use of correlation coefficient in medical research , 2012, 24(3): 69-71. [本文引用: 1]
VAJSOVÁB, FASBENDERD, WIRNHARDTC, LEMAJICS, DEVOSW. Assessing spatial limits of Sentinel-2 data on arable crops in the context of checks by monitoring , 2020, 12(14): 2195. DOI:10.3390/rs12142195URL [本文引用: 1]
LIUC W, KANGS Z, LIF S, LIS E, DUT S, TONGL. Relationship between environmental factor and maximum daily stem shrinkage in apple tree in arid region of northwest China , 2011, 130(1): 118-125. DOI:10.1016/j.scienta.2011.06.022URL [本文引用: 1]
ORTUNOM F, CONEJEROW, MORENOF, MORIANAA, INTRIGLIOLOD S, BIELC, MELLISHOC D, PEREZ-PASTORA, DOMINGOR, RUIZ-SANCHEZM C, CASADESUSJ, BONANYJ, TORRECILLASA. Could trunk diameter sensors be used in woody crops for irrigation scheduling? A review of current knowledge and future perspectives , 2010, 97(1): 1-11. DOI:10.1016/j.agwat.2009.09.008URL [本文引用: 1]
EGEAG, PAGÁNE, BAILLEA, DOMINGOR, NORTESP A, PÉREZ-PASTORA. Usefulness of establishing trunk diameter based reference lines for irrigation scheduling in almond trees , 2009, 27(6): 431-441. DOI:10.1007/s00271-009-0157-0URL [本文引用: 1]
BERMANM E, DEJONGT M. Seasonal patterns of vegetative growth and competition with reproductive sinks in peach (Prunus persica) , 2003, 78(3): 303-309. DOI:10.1080/14620316.2003.11511622URL [本文引用: 1]
SUNH, LIS, LIM Z, LIUH J, QIAOL, ZHANGY. Research progress of image sensing and deep learning in agriculture Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(5): 1-17. (in Chinese) [本文引用: 1]
SHAOZ Z, YAOQ, TANGJ, LIH Q, YANGB J, LVJ, CHENY. Research and development of the intelligent identification system of agricultural pests for mobile terminals Scientia Agricultura Sinica, 2020, 53(16): 3257-3268. (in Chinese) [本文引用: 1]
LIUH F, HEZ, JIAB, LIUZ, LIZ Z, FUJ P, MUR R, KANGJ H. Photosynthetic response characteristics of maize under drip irrigation based on machine learning Scientia Agricultura Sinica, 2019, 52(17): 2939-2950. (in Chinese) [本文引用: 1]