Study On The Estimation Method Of Plant Chlorophyll Content Based On Fluorescence Spectrum

Chlorophyll is the“probe”of vegetation photosynthesis,which is closely related to the growth and photosynthesis of vegetation,and is an important indicator to describe the physiological status of vegetation.Chlorophyll fluorescence,as a unique photochemical phenomenon of plants,can reflect almost all changes of photosynthesis process,and chlorophyll fluorescence has been widely used in plant photosynthesis characterization,plant stress research,remote sensing and so on.The research on the estimation method of chlorophyll content of plant leaves using chlorophyll fluorescence remote sensing technology can not only solve the time-consuming and costly problems in the traditional measurement of chlorophyll,but also provide a reference for large-scale plant biochemical parameter mapping.In this study,a total of 299 sets of chlorophyll fluorescence spectra and chlorophyll content data were obtained from 26 sample trees of 11 tree species mainly distributed in the experimental site of Maoershan Experimental Forest in Shangzhi City,Heilongjiang Province.The regression analysis model,support vector machine model,support vector machine model optimized by particle swarm algorithm and support vector machine model optimized by genetic algorithm were established based on the vegetation index and the spectrum feature variables extracted by three algorithms.The optimal chlorophyll content estimation model was derived from the comparison analysis.The main research results include the following aspects.(1)Three vegetation indices,NDVI,RVI and DVI,were selected and calculated by combining them band by band,and correlated with chlorophyll content.The optimal band combinations for NDVI,DVI,and RVI were screened as 659 and 797 nm,651 and 724 nm,and 659 and 797 nm,respectively.Feature band extraction of spectral data using applied successive projection algorithm(SPA),competitive adaptive weighting algorithm(CARS),and random forest(RF).(2)One-dimensional regression models were established using single vegetation indices,and among all the one-dimensional regression models three quadratic models with single vegetation indices worked better,among which the NDVI-based quadratic model worked best with R~2 of 0.232 and RMSE of 0.781 in the training set and R~2 of 0.258 and RMSE of 0.762 in the validation set.The multiple regression models were established with the vegetation index combinations and the feature band data extracted by the three algorithms as input and the chlorophyll content as output,respectively.The multiple regression model based on CARS worked best with R~2 of 0.504 and RMSE of 0.591 in the training set and R~2 of 0.518 and RMSE of 0.476 in the validation set.(3)Support vector machine model,support vector machine model optimized by particle swarm algorithm and support vector machine model optimized by genetic algorithm based on the combination of vegetation indices and three algorithms to extract feature bands respectively.The results show that the genetic algorithm optimized support vector model based on CARS is the optimal support vector machine model with R~2 of 0.719 and RMSE of 0.234 for the training set and R~2 of 0.725 and RMSE of 0.225 for the validation set.a comparative analysis of all models shows that the GA-SVM estimation model of chlorophyll content based on the feature bands extracted by CARS works best.