Estimation of Apple leaf SPAD Value Based on Hyperspectrum and BP Neural Network

Jiao-yang YU,Qing-rui CHANG,Ming-ming YOU,Zhuo-ran ZHANG,Dan LUO

DOI: https://doi.org/10.3969/j.issn.1001-7461.2018.02.26

2018-01-01

Abstract:Growth status of vegetation was characterized by leaf chlorophyll content.In order to provide a scientific basis for the growth monitoring of large scale coverage,lossless and real-time processing of apple trees at northwest region,the models for estimating chlorophyll content of apple leaves at different growth stages based on hyperspectrum were constructed.The field experiments were conducted in Shaozhai Village of Fufeng,Shaanxi Province.During different growth periods,the hyperspectral reflectance of apple leaf measurements was collected by SVC HR-1024Ⅰ field-portable spectroradiometer,and at the same time,chlorophyll relative content (soil and plant analyzer development,SPAD) of apple leaves was obtained by using SPAD-502.There were totally 120 samples collected at different period,three fourths of which were utilized as the training set and the remaining one quarter as validation set.The model constructed relied on the training set and the validation set was evaluated,respectively.We analyzed the rules between the different growth stages and SPAD value,hyper-spectral reflectance,the correlations between spectral reflectance,17 spectral characteristic parameters and SPAD values of apple leaves at different growth stages.Then single factor regression models and multiple stepwise regression models based on spectral characteristic parameters were established respectively to estimate SPAD value.And 17 spectral characteristic parameters selected by stepwise regression analysis as the input parameters,the measured SPAD values as the output parameters,BP neural network model for each stage was respectively built.Then we compared the predictive power of traditional regression models and multiple stepwise regression models to BP neural network model.The results showed that 1) from shoot-growing stage to fruit maturity stage,SPAD value of apple leaves rose in the first stage,and then decreased.At the same time,the leaf spectral reflectance was gradually getting smaller in the visible light region with the increase of SPAD value,while the leaf spectral reflectance rose in the near infrared region.2) The single factor regression models,multiple stepwise regression models based on spectral characteristic parameters and BP neural network based on stepwise regression analysis were approved by significant testing,which had the highest modeling precision and validation precision at autumn shoot pause growth period.3) The single factor regression models based on blue edge amplitude and green peak area respectively had the highest modeling and prediction accuracy at different growth stages.4) Compared to single factor regression models,multiple stepwise regression models,BP neural network model had the best modeling and verification accuracy in each growth period.The coefficient of determination (R2) for the modeling was higher than 0.90,and the coefficient of determination was greater than 0.84 for the validation set,the corresponding value of root mean square error (RMSE) were lower than 4.41,the relative error (RE) was less than 8.42 ％.Therefore,BP neural network model is an optimal model for the estimation of apple leaf SPAD value and may provide a theoretical basis for the improvement of remote sensing inversion accuracy of apple chlorophyll content.