Estimation Of Chlorophyll Content In Rice Leaves Based On Hyperspectral Data

With the increase of population and the increasing demand for rice,the growth status and yield of rice have become a research hotspot.Among them,chlorophyll is the most important factor affecting crop growth.The content of chlorophyll can accurately reflect the growth status of rice and estimate the yield of rice.The traditional method of chlorophyll analysis is obtained by chemical experiments,which is destructive,time-consuming and costly.Hyperspectral technology uses hyperspectral band to retrieve the chlorophyll value of crops,which is fast,non-destructive and has a wide range of research.The development of hyperspectral technology provides a good solution to the above problems.In this study,hyperspectral technology was used to retrieve chlorophyll content in rice leaves.Redundancy of spectral band would cause inconvenience to data processing.This study focused on dimensionality reduction of hyperspectral data.Based on the spectral band after dimensionality reduction,a model for estimating chlorophyll content was established to improve the drawbacks of traditional problems.In this paper,the data obtained by Kalima Academician Workstation of Shenyang Agricultural University in Liaoning Province are analyzed.Firstly,the dimension of rice leaf hyperspectral data is reduced.Secondly,the regression analysis model for estimating chlorophyll content of leaves and BP,RBF and GA-BP neural network models are established based on the dimension reduction data.The main contents of this paper are as follows:(1)In view of the problems involved in the high-dimensional characteristics of hyperspectral remote sensing data,four dimension reduction methods,namely vegetation index,principal component analysis,SPA feature band extraction and base function expansion,are applied to reduce the dimension of 736 groups of chlorophyll sensitive 400-1000 nm bands studied in this paper.The methods of dimension reduction are as follows:The vegetation index NDVI,RVI,DVI and RDVI were constructed by random combination of two bands.It was found that the optimum spectral bands of NDVI and DVI were 538 and 698nm,the correlation coefficients were 0.699 and 0.679,the optimum spectral bands of DVI were 504 and 708 nm,and the correlation coefficients were 0.662;Principal component analysis was used to find the principal components of the band,and the cumulative contribution rate of the first nine principal components reached 99.928%.This shows that the nine principal components can well summarize the original variables and reduce the601-dimensional data of 400-1000 nm band to 9-dimensional;The bands selected by SPA feature band extraction method are 538,678,691 and 711 nm,which reduces the601-dimensional data from 400 to 1000 nm to 4-dimensional;Gram?Schmidt transform was used to find the projection space in chlorophyll sensitive 400-1000 nm band by base function expansion method,and the main base of concentrated band information was constructed to reduce the dimension of spectral data from 601 to 48 dimensions.(2)The regression analysis model for estimating chlorophyll content in rice leaves was established based on spectral data of four dimension reduction methods,and compared with BP,RBF and GA-BP neural network models.The results showed that the training set R~2was 0.688,RMSE was 1.26,validation set R~2 was 0.692,RMSE was 1.20,which was the best regression model for estimating chlorophyll content in rice leaves by spectral inversion.GA-BP neural network model based on base function expansion was the best one in neural network model.The training set R~2 is 0.736,the RMSE is 1.02,the validation set R~2 is 0.755and the RMSE is 1.01.Compared with the above two optimal models,the GA-BP neural network model based on the expansion of the basis function is the best in all models.