基于红边参数和人工神经网络的苹果叶片叶绿素含量估算
|
摘要
【目的】应用人工神经网络来解决苹果叶片叶绿素含量与高光谱数据之间的非线性关系,建立估测苹果叶片叶绿素含量的最佳模型,为苹果叶片叶绿素含量的快速无损监测提供参考。【方法】以位于陕西扶风杏林镇的树龄为4～5年的15棵苹果树为研究对象,2015年分别于果树花期(4月27日)、幼果期(5月30日)、果实膨大期(7月6日)、果实着色期(8月5日)、果实成熟期(9月11日)采集叶片样本,利用SVC HR-1024i型高光谱仪和SPAD-502叶绿素仪同步获取苹果叶片光谱值和对应的叶绿素含量,对原始光谱和一阶导数光谱与叶绿素含量之间的关系进行分析,从一阶导数光谱中提取苹果叶片光谱的5个红边参数,从5个红边参数中筛选出相关性好的红边参数,使用传统单变量回归算法、反向传播(back propagation,BP)神经网络和径向基函数(radial basis function,RBF)神经网络,建立叶绿素含量估测模型,用决定系数(R2)、均方根误差(RMSE)和相对误差(RE)来验证模型的准确性。【结果】与原始光谱相比,一阶导数光谱与苹果叶片叶绿素含量的相关性更高。5个红边参数中,红边位置、峰度系数、偏度系数与叶绿素含量的相关系数均较高,且均达极显著水平。建立的传统单变量回归模型中,基于红边位置、峰度系数和偏度系数的R2均大于0.7,其中基于偏度系数建立的多项式模型模拟精度最高,验证结果R2为0.872,RMSE为4.631,RE为8.81%。选取红边位置、峰度系数和偏度系数为人工神经网络的输入变量,分别优化BP神经网络的隐含层节点数和网络类型以及RBF神经网络的扩展系数(SPREAD值)来提高预测精度,结果发现,建立隐含层节点数为4的双隐含层BP神经网络最优模型R2为0.891,RMSE为3.844,RE为7.55%;当SPREAD值为0.6时,建立的RBF神经网络最优模型的R2为0.955,RMSE和RE分别为2.517和3.69%。【结论】估算苹果叶片叶绿素含量时,人工神经网络模型比传统单变量模型精度高,其中RBF神经网络模型学习速度快、精度高,拟合结果更加可靠。
【Objective】The nonlinear relationship between hyperspectral data and chlorophyll content of apple leaf was solved using artificial neural network(ANN)and the best model for estimating chlorophyll content of apple leaves was established to provide basis for rapid and non-destructive monitoring of chlorophyll content in apple leaves.【Method】A total of 15 apple trees at the age of 4-5years in Xinglin,Fufeng,Shaanxi were selected and leaf samples were collected in flowering(April 27),young fruit(May30),young fruit expansion(July 6),coloring(August 5),and fruit maturation(September 11)stages in2015.Hyperspectral reflectance was collected by spectrometer(SVC HR-1024i),and relative chlorophyll content was measured by chlorophyll meter SPAD-502.The corrections between original spectrum,first derivative spectrum and chlorophyll content were analyzed.Five red edge parameters were calculated as the first derivative.The traditional single variable empirical regression algorithms,back propagation(BP)and radial basis function(RBF)based on the well-related red edge parameters were applied to establish chlorophyll content estimated models.The accuracy of models was determined based on determination coefficient(R2),root mean square error(RMSE)and relative error(RE).【Result】Compared with the original spectrum,the first derivative spectra had higher relevance with chlorophyll content of apple leaves.The red edge position,kurtosis and skewness had higher correlation among the 5red edge parameters(P<0.01).The correlation coefficients of models based on red edge position,kurtosis and skewness reached over 0.7.The skewness based polynomial model was the best with R2,RMSE and RE of 0.872,4.631 and 8.81%,respectively.Red edge position,kurtosis and skewness were selected as input variables of ANN to build model for estimating chlorophyll content.The node of hidden layer and network type in BP-ANN and the SPREAD value in ANN-RBF were optimized to improve precision.The R2 based on the BP neural network optimal model with 4node of double hidden layer was 0.891,and RMSE and RE were 3.844 and 7.55%.When SPREAD was 0.6,R2,RMSE and RE of the optimized RBF-ANN were 0.955,2.517 and 3.69%,respectively.【Conclusion】Compared to the models with single variable,ANN can improve estimation accuracy of chlorophyll content prediction.RBF-ANN had the advantages of rapid,accurate,and reliable.
【Objective】The nonlinear relationship between hyperspectral data and chlorophyll content of apple leaf was solved using artificial neural network(ANN)and the best model for estimating chlorophyll content of apple leaves was established to provide basis for rapid and non-destructive monitoring of chlorophyll content in apple leaves.【Method】A total of 15 apple trees at the age of 4-5years in Xinglin,Fufeng,Shaanxi were selected and leaf samples were collected in flowering(April 27),young fruit(May30),young fruit expansion(July 6),coloring(August 5),and fruit maturation(September 11)stages in2015.Hyperspectral reflectance was collected by spectrometer(SVC HR-1024i),and relative chlorophyll content was measured by chlorophyll meter SPAD-502.The corrections between original spectrum,first derivative spectrum and chlorophyll content were analyzed.Five red edge parameters were calculated as the first derivative.The traditional single variable empirical regression algorithms,back propagation(BP)and radial basis function(RBF)based on the well-related red edge parameters were applied to establish chlorophyll content estimated models.The accuracy of models was determined based on determination coefficient(R2),root mean square error(RMSE)and relative error(RE).【Result】Compared with the original spectrum,the first derivative spectra had higher relevance with chlorophyll content of apple leaves.The red edge position,kurtosis and skewness had higher correlation among the 5red edge parameters(P<0.01).The correlation coefficients of models based on red edge position,kurtosis and skewness reached over 0.7.The skewness based polynomial model was the best with R2,RMSE and RE of 0.872,4.631 and 8.81%,respectively.Red edge position,kurtosis and skewness were selected as input variables of ANN to build model for estimating chlorophyll content.The node of hidden layer and network type in BP-ANN and the SPREAD value in ANN-RBF were optimized to improve precision.The R2 based on the BP neural network optimal model with 4node of double hidden layer was 0.891,and RMSE and RE were 3.844 and 7.55%.When SPREAD was 0.6,R2,RMSE and RE of the optimized RBF-ANN were 0.955,2.517 and 3.69%,respectively.【Conclusion】Compared to the models with single variable,ANN can improve estimation accuracy of chlorophyll content prediction.RBF-ANN had the advantages of rapid,accurate,and reliable.
引文
[1] Filella I,Penuelas J.The red edge position and shape as indicators of plant chlorophyll content,biomass and hydric status.[J].International Journal of Remote Sensing,1994,15(7):1459-1470.
[2] Blackburn G A.Quantifying chlorophylls and carotenoids at leaf and canopy scales:an evaluation of some hyperspectral approaches[J].Remote Sensing of Environment,1998,66(3):273-285.
[3] Horler D N H,Barber J,Barringer A R.Effects of heavy metals on the absorbance and reflectance spectra of plants[J].International Journal of Remote Sensing,1980(2):121-136.
[4]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[J].农业工程学报,2009(S2):123-129.Yao F Q,Zhang Z H,Yang R Y,et al.Hyperspectral models for estimating vegetation chlorophyll content based on red edge parameter[J].Transactions of the Chinese Society of Agricultural Engineering,2009(S2):123-129.
[5]黄文江,王纪华,刘良云,等.冬小麦红边参数变化规律及其营养诊断[J].遥感技术与应用,2003,18(4):206-211.Huang W J,Wang J H,Liu L Y,et al.The red edge parameters diversification disciplinarian and its application for nutrition diagnosis[J].Remote Sensing Technology and Application,2003,18(4):206-211.
[6]谭昌伟,王纪华,陆建飞,等.基于红边参数的夏玉米长势监测及其营养诊断研究[J].中国生态农业学报,2007,15(1):82-86.Tan C W,Wang J H,Lu J F,et al.Summer maize growth supervision and nutrition diagnosis with red edge parameters[J].Chinese Journal of Eco-Agriculture,2007,15(1):82-86.
[7]王秀珍,王人潮,李云梅,等.不同氮素营养水平的水稻冠层光谱红边参数及其应用研究[J].浙江大学学报(农业与生命科学版),2001,27(3):301-306.Wang X Z,Wang R C,Li Y M,et al.Study on red edge characteristics of rice spectral caused by nitrogen level[J].Juurnal of Zhejiang University(Agriculture&Life Science),2001,27(3):301-306.
[8]郑立华,李民赞,潘娈,等.基于近红外光谱技术的土壤参数BP神经网络预测[J].光谱学与光谱分析,2008,28(5):1160-1164.Zheng L H,Li M Z,Pan L,et al.Estimation of soil organic matter and soil total nitrogen based on NIR spectroscopy and BP neural network[J].Spectroscopy and Spectral Analysis,2008,28(5):1160-1164.
[9]高如泰,陈焕伟,李保国,等.基于BP神经网络的土壤水力学参数预测[J].土壤通报,2005,36(5):641-646.Gao R T,Chen H W,Li B G,et al.Prediction of soil hydraulic characteristic parameters based on BP neural network[J].Chinese Journal of Soil Science,2005,36(5):641-646.
[10]Minasny B,McBratney A B.The method for fitting neural network parametric pedotransfer functions[J].Soil Science Society of America Journal,2002,66(2):352-361.
[11]樊迪.基于BP网络的小麦病害诊断专家系统的研究与实现[D].陕西杨凌:西北农林科技大学,2015:3-6.Fan D.Research and implementation of wheat disease diagnosis expert system based on BP network[D].Yangling,Shaanxi:Northwest A&F University,2015:3-6.
[12]Chen L,Huang J F,Wang F M,et al.Comparison between back propagation neural network and regression models for the estimation of pigment content in rice leaves and panicles using hyperspectral data[J].International Journal of Remote Sensing,2007,28(16):3457-3478.
[13]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.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.
[14]朱新开,盛海君,顾晶,等.应用SPAD值预测小麦叶片叶绿素和氮含量的初步研究[J].麦类作物学报,2005,25(2):46-50.Zhu X K,Sheng H J,Gu J,et al.Primary study on application of SPAD value to estimate chlorophyll and nitrogen content in wheat leaves[J].Journal of Triticeae Crops,2005,25(2):46-50.
[15]Ruffin C,King R L,Younan N H.A combined dericative spectroscopy and Savitaky-Golay filtering method for the analysis of hyperspectral data[J].Mapping Sciences&Remote Sensing,2008,45(1):1-15.
[16]Tsai F,Philpot W.Derivative analysis of hyperspectral data[J].Remote Sensing of Environment,1998,66(1):41-51.
[17]李友坤.BP神经网络的研究分析及改进应用[D].安徽淮南:安徽理工大学,2012:14-20.Li Y K.Analysis and improvement applications of BP neural network[D].Huainan,Anhui:Anhui University of Science and Technology,2012:14-20.
[18]周志华,曹存根.神经网络及其应用[M].北京:清华大学出版社,2004:101-110.Zhou Z H,Cao C G.Neural network and application[M].Beijing:Tsinghua University Press,2004:101-110.
[19]唐延林,黄敬峰,王秀珍,等.水稻、玉米、棉花的高光谱及其红边特征比较[J].中国农业科学,2004,37(1):29-35.Tang Y L,Huang J F,Wang X Z,et al.Comparison of the characteristics of hyperspectra and the red edge in rice,corn and cotton[J].Scientia Agricultura Sinica,2004,37(1):29-35.
[20]朱咏莉,李萍萍,毛罕平,等.生菜叶片光谱红边参数对氮营养的响应特征分析[J].农业机械学报,2011,42(11):174-177.Zhu Y L,Li P P,Mao H P,et al.Response features of red edge parameters for lettuce leaf spectra under different nitrogen levels[J].Transactions of the Chinese Society for Agricultural Machinery,2011,42(11):174-177.
[21]姚付启,蔡焕杰,李亚龙,等.基于红边参数的冬小麦SPAD高光谱遥感监测[J].中国农村水利水电,2015(3):84-87.Yao F Q,Cai H J,Li Y L,et al.Monitoring winter wheat SPAD based on red edge parameter derived from hyperspectral reflectance[J].China Rural Water and Hydropower,2015(3):84-87.
[22]王晓星,常庆瑞,刘梦云,等.冬小麦冠层水平叶绿素含量的高光谱估测[J].西北农林科技大学学报(自然科学版),2016,44(2):48-54.Wang X X,Chang Q R,Liu M Y,et al.Hyper-spectral estimation of chlorophyll content in canopy of winter wheat[J].Journal of Northwest A&F University(Natural Science Edition),2016,44(2):48-54.
[23]谌爱文.基于BP和RBF神经网络的数据预测方法研究[D].长沙:中南大学,2007:6-10.Chen A W.Study on data prediction methods based on BP and RBF neural network[D].Changsha:Central South University,2007:6-10.
[24]冯伟.基于高光谱遥感的小麦氮素营养及生长指标监测研究[D].南京:南京农业大学,2007:9-12.Feng W.Monitoring nitrogen status and growth characters with canopy hyperspectral remote sensing in wheat[D].Nanjing:Nanjing Agricultural University,2007:9-12.
[25]姚霞,田永超,刘小军,等.不同算法红边位置监测小麦冠层氮素营养指标的比较[J].中国农业科学,2010,43(13):2661-2667.Yao X,Tian Y C,Liu X J,et al.Comparative study on monitoring canopy leaf nitrogen status on red edge position with different algorithms in wheat[J].Scientia Agricultura Sinica,2010,43(13):2661-2667.
[26]罗丹,常庆瑞,齐雁冰,等.基于光谱指数的冬小麦冠层叶绿素含量估算模型研究[J].麦类作物学报,2016,36(9):1225-1233.Luo D,Chang Q R,Qi Y B,et al.Estimation model for chlorophyll content in winter wheat canopy based on spectral indices[J].Journal of Triticeae Crops,2016,36(9):1225-1233.
[27]丁永军,张晶晶,李修华,等.基于光谱红边位置提取算法的番茄叶片叶绿素含量估测[J].农业机械学报,2016,47(3):292-297.Ding Y J,Zhang J J,Li X H,et al.Estimation of chlorophyll content of tomato leaf using spectrum red edge position extract algorithm[J].Transactions of the Chinese Society for Agricultural Machinery,2016,47(3):292-297.
[28]吴倩,林蕾,王学军,等.福海叶绿素含量的人工神经网络反演模型[J].地理与地理信息科学,2004,20(4):27-30.Wu Q,Lin L,Wang X J,et al.The neural network model on chlorophyll concentration in the Fuhai lake[J].Geography and Geo-Information Science,2004,20(4):27-30.
[29]罗定贵,郭清,王学军.地表水质评价的径向基神经网络模型设计[J].地理与地理信息科学,2003,19(5):77-81.Luo D G,Guo Q,Wang X J.Neural network model design of surface water environmental quality assessment[J].Geography and Geo-Information Science,2003,19(5):77-81.
[2] Blackburn G A.Quantifying chlorophylls and carotenoids at leaf and canopy scales:an evaluation of some hyperspectral approaches[J].Remote Sensing of Environment,1998,66(3):273-285.
[3] Horler D N H,Barber J,Barringer A R.Effects of heavy metals on the absorbance and reflectance spectra of plants[J].International Journal of Remote Sensing,1980(2):121-136.
[4]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[J].农业工程学报,2009(S2):123-129.Yao F Q,Zhang Z H,Yang R Y,et al.Hyperspectral models for estimating vegetation chlorophyll content based on red edge parameter[J].Transactions of the Chinese Society of Agricultural Engineering,2009(S2):123-129.
[5]黄文江,王纪华,刘良云,等.冬小麦红边参数变化规律及其营养诊断[J].遥感技术与应用,2003,18(4):206-211.Huang W J,Wang J H,Liu L Y,et al.The red edge parameters diversification disciplinarian and its application for nutrition diagnosis[J].Remote Sensing Technology and Application,2003,18(4):206-211.
[6]谭昌伟,王纪华,陆建飞,等.基于红边参数的夏玉米长势监测及其营养诊断研究[J].中国生态农业学报,2007,15(1):82-86.Tan C W,Wang J H,Lu J F,et al.Summer maize growth supervision and nutrition diagnosis with red edge parameters[J].Chinese Journal of Eco-Agriculture,2007,15(1):82-86.
[7]王秀珍,王人潮,李云梅,等.不同氮素营养水平的水稻冠层光谱红边参数及其应用研究[J].浙江大学学报(农业与生命科学版),2001,27(3):301-306.Wang X Z,Wang R C,Li Y M,et al.Study on red edge characteristics of rice spectral caused by nitrogen level[J].Juurnal of Zhejiang University(Agriculture&Life Science),2001,27(3):301-306.
[8]郑立华,李民赞,潘娈,等.基于近红外光谱技术的土壤参数BP神经网络预测[J].光谱学与光谱分析,2008,28(5):1160-1164.Zheng L H,Li M Z,Pan L,et al.Estimation of soil organic matter and soil total nitrogen based on NIR spectroscopy and BP neural network[J].Spectroscopy and Spectral Analysis,2008,28(5):1160-1164.
[9]高如泰,陈焕伟,李保国,等.基于BP神经网络的土壤水力学参数预测[J].土壤通报,2005,36(5):641-646.Gao R T,Chen H W,Li B G,et al.Prediction of soil hydraulic characteristic parameters based on BP neural network[J].Chinese Journal of Soil Science,2005,36(5):641-646.
[10]Minasny B,McBratney A B.The method for fitting neural network parametric pedotransfer functions[J].Soil Science Society of America Journal,2002,66(2):352-361.
[11]樊迪.基于BP网络的小麦病害诊断专家系统的研究与实现[D].陕西杨凌:西北农林科技大学,2015:3-6.Fan D.Research and implementation of wheat disease diagnosis expert system based on BP network[D].Yangling,Shaanxi:Northwest A&F University,2015:3-6.
[12]Chen L,Huang J F,Wang F M,et al.Comparison between back propagation neural network and regression models for the estimation of pigment content in rice leaves and panicles using hyperspectral data[J].International Journal of Remote Sensing,2007,28(16):3457-3478.
[13]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.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.
[14]朱新开,盛海君,顾晶,等.应用SPAD值预测小麦叶片叶绿素和氮含量的初步研究[J].麦类作物学报,2005,25(2):46-50.Zhu X K,Sheng H J,Gu J,et al.Primary study on application of SPAD value to estimate chlorophyll and nitrogen content in wheat leaves[J].Journal of Triticeae Crops,2005,25(2):46-50.
[15]Ruffin C,King R L,Younan N H.A combined dericative spectroscopy and Savitaky-Golay filtering method for the analysis of hyperspectral data[J].Mapping Sciences&Remote Sensing,2008,45(1):1-15.
[16]Tsai F,Philpot W.Derivative analysis of hyperspectral data[J].Remote Sensing of Environment,1998,66(1):41-51.
[17]李友坤.BP神经网络的研究分析及改进应用[D].安徽淮南:安徽理工大学,2012:14-20.Li Y K.Analysis and improvement applications of BP neural network[D].Huainan,Anhui:Anhui University of Science and Technology,2012:14-20.
[18]周志华,曹存根.神经网络及其应用[M].北京:清华大学出版社,2004:101-110.Zhou Z H,Cao C G.Neural network and application[M].Beijing:Tsinghua University Press,2004:101-110.
[19]唐延林,黄敬峰,王秀珍,等.水稻、玉米、棉花的高光谱及其红边特征比较[J].中国农业科学,2004,37(1):29-35.Tang Y L,Huang J F,Wang X Z,et al.Comparison of the characteristics of hyperspectra and the red edge in rice,corn and cotton[J].Scientia Agricultura Sinica,2004,37(1):29-35.
[20]朱咏莉,李萍萍,毛罕平,等.生菜叶片光谱红边参数对氮营养的响应特征分析[J].农业机械学报,2011,42(11):174-177.Zhu Y L,Li P P,Mao H P,et al.Response features of red edge parameters for lettuce leaf spectra under different nitrogen levels[J].Transactions of the Chinese Society for Agricultural Machinery,2011,42(11):174-177.
[21]姚付启,蔡焕杰,李亚龙,等.基于红边参数的冬小麦SPAD高光谱遥感监测[J].中国农村水利水电,2015(3):84-87.Yao F Q,Cai H J,Li Y L,et al.Monitoring winter wheat SPAD based on red edge parameter derived from hyperspectral reflectance[J].China Rural Water and Hydropower,2015(3):84-87.
[22]王晓星,常庆瑞,刘梦云,等.冬小麦冠层水平叶绿素含量的高光谱估测[J].西北农林科技大学学报(自然科学版),2016,44(2):48-54.Wang X X,Chang Q R,Liu M Y,et al.Hyper-spectral estimation of chlorophyll content in canopy of winter wheat[J].Journal of Northwest A&F University(Natural Science Edition),2016,44(2):48-54.
[23]谌爱文.基于BP和RBF神经网络的数据预测方法研究[D].长沙:中南大学,2007:6-10.Chen A W.Study on data prediction methods based on BP and RBF neural network[D].Changsha:Central South University,2007:6-10.
[24]冯伟.基于高光谱遥感的小麦氮素营养及生长指标监测研究[D].南京:南京农业大学,2007:9-12.Feng W.Monitoring nitrogen status and growth characters with canopy hyperspectral remote sensing in wheat[D].Nanjing:Nanjing Agricultural University,2007:9-12.
[25]姚霞,田永超,刘小军,等.不同算法红边位置监测小麦冠层氮素营养指标的比较[J].中国农业科学,2010,43(13):2661-2667.Yao X,Tian Y C,Liu X J,et al.Comparative study on monitoring canopy leaf nitrogen status on red edge position with different algorithms in wheat[J].Scientia Agricultura Sinica,2010,43(13):2661-2667.
[26]罗丹,常庆瑞,齐雁冰,等.基于光谱指数的冬小麦冠层叶绿素含量估算模型研究[J].麦类作物学报,2016,36(9):1225-1233.Luo D,Chang Q R,Qi Y B,et al.Estimation model for chlorophyll content in winter wheat canopy based on spectral indices[J].Journal of Triticeae Crops,2016,36(9):1225-1233.
[27]丁永军,张晶晶,李修华,等.基于光谱红边位置提取算法的番茄叶片叶绿素含量估测[J].农业机械学报,2016,47(3):292-297.Ding Y J,Zhang J J,Li X H,et al.Estimation of chlorophyll content of tomato leaf using spectrum red edge position extract algorithm[J].Transactions of the Chinese Society for Agricultural Machinery,2016,47(3):292-297.
[28]吴倩,林蕾,王学军,等.福海叶绿素含量的人工神经网络反演模型[J].地理与地理信息科学,2004,20(4):27-30.Wu Q,Lin L,Wang X J,et al.The neural network model on chlorophyll concentration in the Fuhai lake[J].Geography and Geo-Information Science,2004,20(4):27-30.
[29]罗定贵,郭清,王学军.地表水质评价的径向基神经网络模型设计[J].地理与地理信息科学,2003,19(5):77-81.Luo D G,Guo Q,Wang X J.Neural network model design of surface water environmental quality assessment[J].Geography and Geo-Information Science,2003,19(5):77-81.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |