基于高光谱和BP神经网络的双子叶植物叶片叶绿素遥感估算
|
摘要
针对传统叶绿素分析方法具有破坏性且耗费人力、时间长、成本高的弊端,依据LOPEX’93数据集中双子叶植物的高光谱数据和叶绿素值,构建了双子叶植物基于高光谱的叶绿素含量最佳估算模型,利用Pearson相关性分析一阶微分光谱、高光谱特征参数与叶绿素的相关关系,发现724nm波段处一阶导数与双子叶植物叶绿素值的相关性最大,其相关性为0.509;高光谱特征参数RVI、NDVI、TCAR与叶绿素的相关性达到0.7以上,构建基于一阶微分光谱、高光谱特征参数和BP神经网络的叶绿素估算模型,并对模型进行验证;再结合一元线性模型、指数模型、对数模型和幂函数模型与BP神经网络模型进行比较。结果表明:叶绿素值与一阶微分光谱在724nm处的光谱数据作为自变量建立的传统回归模型可用于双子叶植物叶绿素的估算,最优建模样本R~2和最优验证样本R_V~2分别为0.541和0.745,RMSE为6.16;基于高光谱特征参数RVI、NDVI、TCAR建立的叶绿素估算回归模型,最优建模样本R~2和最优验证样本R_V~2分别为0.618,0.708;0.632,0.866;0.594,0.654,RMSE分别为6.65,5.61,7.07,将基于高光谱特征参数变量构建传统回归模型时筛选到的光谱参数作为输入,实测叶绿素值作为输出,构建BP神经网络模型,其最优建模R2与最优验模R_V~2分别为0.692和0.874,最优验证样本RMSE为5.23,与其他回归模型相比,BP神经网络模型预测精度最高。研究表明基于高光谱数据的模型具有较好的预测能力,是估算双子叶植物叶绿素值的一种高效的方法。
Aiming at solving the disadvantages of traditional chlorophyll analysis methods, such as destructive, manpower consuming, long time and high cost, the optimal estimation model of chlorophyll content in dicotyledon was established based on hyperspectral data and chlorophyll values of dicotyledon in LOPEX 93. Pearson correlation was used to analyze the correlation among first order differential spectra, hyperspectral characteristic parameters and chlorophyll. It was found that the first derivative at 724 nm had the greatest correlation with chlorophyll value of dicotyledons, and the correlation was 0.509. The correlation among hyperspectral parameters RVI, NDVI, TCAR and chlorophyll was more than 0.7.A, chlorophyll estimation model based on first order differential spectra, hyperspectral characteristic parameters and BP neural network was constructed and validated. Then,the model was compared with the BP neural network model, including one linear model, exponential model, logarithmic model and power function model. These results showed that the chlorophyll values had the largest correlation coefficient with the first derivative spectra at 724 nm. The traditional regression model based on the spectral data of this wavelength could be used for the estimation of chlorophyll in dicotyledonous plants. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.541 and 0.745 and the RMSE in validation samples was 6.16. Regression models for chlorophyll estimation based on hyperspectral characteristic parameters RVI, NDVI and TCAR were established. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.618, 0.708, 0.632, 0.866, 0.594, 0.654, and RMSE was 6.65,5.61 a nd 7.07, respectively.The selected spectral parameters during the traditional spectral model based on hyperspectral characteristic parameter variables are used as input parameters, and the chlorophyll values as the outputs parameters, BP neural network model was built. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.692 and 0.874. The RMSE in optimal validation samples was 5.23. Compared with other regression models, BP neural network model had the highest prediction accuracy. The research showed that hyperspectral model based on hyperspectral data had better prediction ability, and it was an efficient method for estimating chlorophyll value of dicotyledon.
Aiming at solving the disadvantages of traditional chlorophyll analysis methods, such as destructive, manpower consuming, long time and high cost, the optimal estimation model of chlorophyll content in dicotyledon was established based on hyperspectral data and chlorophyll values of dicotyledon in LOPEX 93. Pearson correlation was used to analyze the correlation among first order differential spectra, hyperspectral characteristic parameters and chlorophyll. It was found that the first derivative at 724 nm had the greatest correlation with chlorophyll value of dicotyledons, and the correlation was 0.509. The correlation among hyperspectral parameters RVI, NDVI, TCAR and chlorophyll was more than 0.7.A, chlorophyll estimation model based on first order differential spectra, hyperspectral characteristic parameters and BP neural network was constructed and validated. Then,the model was compared with the BP neural network model, including one linear model, exponential model, logarithmic model and power function model. These results showed that the chlorophyll values had the largest correlation coefficient with the first derivative spectra at 724 nm. The traditional regression model based on the spectral data of this wavelength could be used for the estimation of chlorophyll in dicotyledonous plants. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.541 and 0.745 and the RMSE in validation samples was 6.16. Regression models for chlorophyll estimation based on hyperspectral characteristic parameters RVI, NDVI and TCAR were established. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.618, 0.708, 0.632, 0.866, 0.594, 0.654, and RMSE was 6.65,5.61 a nd 7.07, respectively.The selected spectral parameters during the traditional spectral model based on hyperspectral characteristic parameter variables are used as input parameters, and the chlorophyll values as the outputs parameters, BP neural network model was built. The R~2 in optimal modeling samples and the R_V~2 in optimal validation samples were 0.692 and 0.874. The RMSE in optimal validation samples was 5.23. Compared with other regression models, BP neural network model had the highest prediction accuracy. The research showed that hyperspectral model based on hyperspectral data had better prediction ability, and it was an efficient method for estimating chlorophyll value of dicotyledon.
引文
[1]丁永军,张晶晶,孙红,等.玻璃温室环境下番茄叶绿素含量敏感光谱波段提取及估测模型[J].光谱学与光谱分析,2017,37(1):194-199.
[2]徐新刚,赵春江,王纪华,等.新型光谱曲线特征参数与水稻叶绿素含量间的关系研究[J].光谱学与光谱分析,2011,31(1):188-191.
[3]岳学军,全东平,洪添胜,等.柑橘叶片叶绿素含量高光谱无损检测模型[J].农业工程学报,2015,31(1):294-302.
[4] FERET J B,FRANCOIS C,ASNER G P,et al.PROSPECT-4 and 5:Advances in the leaf optical properties model separating photosynthetic pigments[J].Remote Sensing of Environment,2008,112(6):3030-3043.
[5] MAIN R,CHO M A,MATHIEU R,et al.An investigation into robust spectral indices for leaf chlorophyll estimation[J].ISPRS Journal of Photogrammetry and Remote Sensing,2011,66(6):751-761.
[6]宫兆宁,赵雅莉,赵文吉,等.基于光谱指数的植物叶片叶绿素含量的估算模型[J].生态学报,2014,34(20):5736-5745.
[7]王锦坚,洪添胜,岳学军,等.基于高光谱的柑橘树红边特征及叶绿素和LAI的监测[J].中国科学:信息科学,2010,40(S1):125-132.
[8]李敏夏,张林森,李丙智,等.苹果叶片高光谱特性与叶绿素含量和SPAD值的关系[J].西北林学院学报,2010,25(2):35-39.
[9]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.
[10]于丰华.基于无人机高光谱遥感的东北粳稻生长信息反演建模研究[D].沈阳:沈阳农业大学,2017.
[11]尹健康,陈昌华,邢小军,等.基于BP神经网络的烟田土壤水分预测[J].电子科技大学学报,2010,39(6):891-895.
[12]徐小逊,张萍,徐精文,等.基于被动微波遥感和BP神经网络的土壤水分反演研究——以川中丘陵区为例[J].西南农业学报,2014,27(6):2478-2484.
[13]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,2009,25(S2):123-129.
[14] 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.
[15]王洋,肖文,邹焕成,等.基于PROSPECT模型的植物叶片干物质估测建模研究[J].沈阳农业大学学报,2018,49(1):121-127.
[16]王纪华,赵春江,黄文江.农业定量遥感基础与应用[M].北京:科学出版社,2008.
[17]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[J].光谱学与光谱分析,2013,33(10):2838-2842.
[18]吴长山,项月琴,郑兰芬,等.利用高光谱数据对作物群体叶绿素密度估算的研究[J].遥感学报,2000,25(3):228-232.
[19]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[J].农业工程学报,2005,31(1):36-40.
[20]房贤一,朱西存,王凌,等.基于高光谱的苹果盛果期冠层叶绿素含量监测研究[J].中国农业科学,2013,46(16):3504-3513.
[21] YU F H,XU T Y,CAO Y L,et al.Models for estimating the leaf NDVI of japonica rice on a canopy scale by combining canopy NDVI and multisource environmental data in Northeast China[J].International Journal of Agriculture and Biology Engineering,2016,9(5):132-142.
[22]季云.BP算法及其应用实例[J].中国新技术新产品,2010,15(9):29.
[23]夏天,吴文斌,周清波,等.冬小麦叶面积指数高光谱遥感反演方法对比[J].农业工程学报,2013,29(3):139-147.
[24] AZIMI-SADJADI M R,GHALOUM S,ZOUGHI R.Terrain classification in SAR images using principal components analysis and neural network[J].Geoscience&Remote Sensing IEEE Transactions on,1993,31(2):511-515.
[25]刘晓莉,杨灵娥,宋春玲.提高多目标输出神经网络模型泛化能力和预测精度的方法[J].佛山科学技术学院学报(自然科学版),2008,10(1):31-33.
[2]徐新刚,赵春江,王纪华,等.新型光谱曲线特征参数与水稻叶绿素含量间的关系研究[J].光谱学与光谱分析,2011,31(1):188-191.
[3]岳学军,全东平,洪添胜,等.柑橘叶片叶绿素含量高光谱无损检测模型[J].农业工程学报,2015,31(1):294-302.
[4] FERET J B,FRANCOIS C,ASNER G P,et al.PROSPECT-4 and 5:Advances in the leaf optical properties model separating photosynthetic pigments[J].Remote Sensing of Environment,2008,112(6):3030-3043.
[5] MAIN R,CHO M A,MATHIEU R,et al.An investigation into robust spectral indices for leaf chlorophyll estimation[J].ISPRS Journal of Photogrammetry and Remote Sensing,2011,66(6):751-761.
[6]宫兆宁,赵雅莉,赵文吉,等.基于光谱指数的植物叶片叶绿素含量的估算模型[J].生态学报,2014,34(20):5736-5745.
[7]王锦坚,洪添胜,岳学军,等.基于高光谱的柑橘树红边特征及叶绿素和LAI的监测[J].中国科学:信息科学,2010,40(S1):125-132.
[8]李敏夏,张林森,李丙智,等.苹果叶片高光谱特性与叶绿素含量和SPAD值的关系[J].西北林学院学报,2010,25(2):35-39.
[9]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.
[10]于丰华.基于无人机高光谱遥感的东北粳稻生长信息反演建模研究[D].沈阳:沈阳农业大学,2017.
[11]尹健康,陈昌华,邢小军,等.基于BP神经网络的烟田土壤水分预测[J].电子科技大学学报,2010,39(6):891-895.
[12]徐小逊,张萍,徐精文,等.基于被动微波遥感和BP神经网络的土壤水分反演研究——以川中丘陵区为例[J].西南农业学报,2014,27(6):2478-2484.
[13]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,2009,25(S2):123-129.
[14] 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.
[15]王洋,肖文,邹焕成,等.基于PROSPECT模型的植物叶片干物质估测建模研究[J].沈阳农业大学学报,2018,49(1):121-127.
[16]王纪华,赵春江,黄文江.农业定量遥感基础与应用[M].北京:科学出版社,2008.
[17]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[J].光谱学与光谱分析,2013,33(10):2838-2842.
[18]吴长山,项月琴,郑兰芬,等.利用高光谱数据对作物群体叶绿素密度估算的研究[J].遥感学报,2000,25(3):228-232.
[19]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[J].农业工程学报,2005,31(1):36-40.
[20]房贤一,朱西存,王凌,等.基于高光谱的苹果盛果期冠层叶绿素含量监测研究[J].中国农业科学,2013,46(16):3504-3513.
[21] YU F H,XU T Y,CAO Y L,et al.Models for estimating the leaf NDVI of japonica rice on a canopy scale by combining canopy NDVI and multisource environmental data in Northeast China[J].International Journal of Agriculture and Biology Engineering,2016,9(5):132-142.
[22]季云.BP算法及其应用实例[J].中国新技术新产品,2010,15(9):29.
[23]夏天,吴文斌,周清波,等.冬小麦叶面积指数高光谱遥感反演方法对比[J].农业工程学报,2013,29(3):139-147.
[24] AZIMI-SADJADI M R,GHALOUM S,ZOUGHI R.Terrain classification in SAR images using principal components analysis and neural network[J].Geoscience&Remote Sensing IEEE Transactions on,1993,31(2):511-515.
[25]刘晓莉,杨灵娥,宋春玲.提高多目标输出神经网络模型泛化能力和预测精度的方法[J].佛山科学技术学院学报(自然科学版),2008,10(1):31-33.