Inversion Of Chlorophyll Content In Citrus Canopy Based On Hyperspectral Data

Chlorophyll is an important index in crop growth.Accurate,rapid and nondestructive detection of chlorophyll content in citrus leaves is of great significance for the health monitoring of citrus fruit trees,agricultural irrigation and fertilization regulation.UAV hyperspectral and ground ASD hyperspectral have vegetation reflection information of 270 bands and 2,151 bands respectively,which can better characterize crop nutrient content and growth information,and is an important means to detect vegetation chlorophyll content.After spectral transformation and feature optimization of UAV and ASD hyperspectral data,The machine learning model can accurately invert the chlorophyll content of citrus canopy.Through sensitivity analysis and continuous optimization of input parameters,PROSAIL model can simulate spectra of the same band as ground ASD data and improve the inversion ability.Therefore,based on UAV hyperspectral,ASD hyperspectral and PROASAIL simulated hyperspectral data,a series of spectral transformations such as multiple scattering correction and continuous wavelet transform were carried out respectively in this paper,and feature optimization was carried out and inversion model was established to achieve accurate inversion of citrus canopy chlorophyll content.The main research contents and conclusions are as follows:(1)Construct inversion model of citrus canopy chlorophyll content index based on UAV hyperspectrum: SNV-D1 spectral transform can effectively improve the influence of spectral noise and baseline drift.RF-RFE feature primary and PCA feature optimization methods not only avoid the multicollinearity interference of spectral data,but also retain the original spectral information of hyperspectral data,which provides good input data for the establishment of high-precision SVR prediction model.(2)Construct orange canopy chlorophyll inversion model based on ASD hyperspectral combined with continuous wavelet transform: The regression model established by using the wavelet coefficients near the band 710～760nm can better predict the citrus canopy chlorophyll,and the wavelet transform can better eliminate the correlation between bands.When the canopy ASD spectrum is modeled by using the SVR algorithm after the cgau8 wavelet basis function transform,the model accuracy is the highest,and the coefficient of determination is 0.8687.It showed that the model could better retrieve the chlorophyll content of citrus canopy.(3)Using PROSAIL model to simulate citrus canopy spectrum and invert citrus canopy chlorophyll,combining with machine learning deviation compensation method to improve inversion accuracy: EFAST sensitivity analysis was used to determine the input parameters and verify the validity of the simulation spectrum.The simulation spectrum was preprocessed and feature extracted based on the optimal wavelet basis function.The machine learning algorithm SVR model was used to compensate the deviation of PROSAIL model.Finally,the SVR-PROSAIL model was constructed to realize the inversion study of citrus canopy chlorophyll content.The inversion accuracy of the model was better than that of the model constructed by UAV hyperspectral and ASD hyperspectral,and the coefficient of determination was 0.9183,which further verified the validity of the SVR-PROSAIL model.The results provided the basis for rapid,accurate and nondestructive detection of chlorophyll content in citrus canopy,so as to achieve quantitative chlorophyll inversion.