光学学报, 2019, 39 (9): 0930003, 网络出版: 2019-09-09
基于高光谱和BP神经网络的棉花冠层叶绿素含量联合估算 下载: 873次
Combined Estimation of Chlorophyll Content in Cotton Canopy Based on Hyperspectral Parameters and Back Propagation Neural Network
光谱学 棉花 叶绿素含量 高光谱 BP神经网络 spectroscopy cotton chlorophyll content hyperspectral parameters back propagation neural network
摘要
冠层叶绿素能够有效反映植被的生长状况。为了基于高光谱精确估算冠层的叶绿素含量,以棉花为研究对象,实测棉花冠层光谱反射率和叶绿素含量,然后进行原始光谱数据转换,计算高光谱参数,分析叶绿素含量与高光谱参数之间的相关关系,构建估算棉花冠层叶绿素含量的BP神经网络模型。结果表明:包络线去除处理后,冠层反射率和叶绿素含量的相关性在560~740 nm波段范围内提高了10.7%,效果优于原始光谱和一阶微分光谱得到的结果;基于原始光谱和去除包络线光谱建立的植被指数mSR、mND、NDI、DD与叶绿素含量表现出较高的相关性,相关系数均在0.8左右;在所建的BP神经网络模型中,基于包络线光谱指数建立的模型的决定系数为0.85,均方根误差和相对误差分别为1.37、1.97%,这一结果优于基于红边参数、原始光谱植被指数和一阶微分光谱指数建立的模型。本研究可为作物叶绿素含量估算的实际应用提供理论依据和技术支持。
Abstract
Chlorophyll content in canopy plays an important role in reflecting the growing status of vegetation. To achieve high accuracy of chlorophyll content estimation based on hyperspectral data, the spectral reflectance and chlorophyll content in cotton canopy are measured from field observation. Original spectral data is transformed to calculate the hyperspectral parameters. The correlation between hyperspectral parameters and chlorophyll content is analyzed and a back propagation (BP) neural network model for estimating chlorophyll content in cotton canopy is established. Results show that after continuum-removal transformation, the correlation between canopy reflectance and chlorophyll content improves by 10.7% in the spectral bands of 560-740 nm, which is better than that of the original spectrum and the first-order differential spectrum. Vegetation indices, such as mSR, mND, NDI, and DD, which are established using the original spectrum and continuum-removal spectrum, show a high correlation with chlorophyll content under both spectral conditions with a correlation coefficient of approximately 0.8. In the BP neural network model, the model determination coefficient based on continuum spectral indices is 0.85, and the root-mean-square error and relative error are 1.37 and 1.97%, respectively. This result is better than that of the model based on red-edge parameters, original spectral vegetation indices, and first-order differential spectral indices. This study provides important theoretical basis and technical support for practical application of chlorophyll content estimation in crops.
依尔夏提·阿不来提, 白灯莎·买买提艾力, 买买提·沙吾提, 安申群. 基于高光谱和BP神经网络的棉花冠层叶绿素含量联合估算[J]. 光学学报, 2019, 39(9): 0930003. Ablet Ershat, Maimaitiaili Baidengsha, Sawut Mamat, Shenqun An. Combined Estimation of Chlorophyll Content in Cotton Canopy Based on Hyperspectral Parameters and Back Propagation Neural Network[J]. Acta Optica Sinica, 2019, 39(9): 0930003.