Study On High-throughput Inversion Model Of Hyperspectral Biochemical Parameters Of Apple Leaves

Plant growth is usually represented by the apparent parameters and biochemical parameters of leaves,among which nitrogen content,water content and chlorophyll of vegetation leaves can better reflect their health characteristics.The application of spectroscopic techniques in the inversion of biochemical parameters of apple leaves is of great practical importance for the growth detection and scientific management of apple trees.However,most of the current inversions of apple leaf biochemical parameters are for a single variety or a single period,and the modeling process is not uniformly standardized and complex and tedious.In order to solve the above problems,the main biochemical parameters(nitrogen content,water content,chlorophyll)and their hyperspectral data were collected from apple leaves at different phenological stages(budding,stable growth,fruit expansion and fruit ripening)in this study,using ten varieties of apple trees as samples in Hong Qipo area.The correlation between biochemical parameters and spectral features analysis,and different strategies were selected to construct inverse models of apple leaf content nitrogen content,water content,and chlorophyll,and the main research contents are as follows:(1)Correlation analysis:The original spectral reflectance of the leaves were calculated separately for red-edge,yellow-edge and blue-edge positions,vegetation index,processed by first-order derivative and wavelet transform,and the results were correlated with leaf nitrogen content,water content and chlorophyll content to determine the characteristic waveband range.(2)Hyperspectral data pre-processing:Baseline correction using iterative polynomial fitting with first-order derivatives,scattering correction selection of multivariate scattering correction with standard normal transformation,filtering and noise reduction using SG smoothing and discrete wavelet transform,Feature analysis using continuous projection algorithm.Pre-processing scheme screening by visual inspection method,combined with correlation analysis results to determine the characteristic waveform.(3)Model establishment:The combination of LPF-SG-MSC-SPA-XGboost algorithm in the inversion model of apple leaf nitrogen content,water content and chlorophyll content has the best inversion effect.The mean R~2value of the modeling set is 83.6%and the mean R~2value of the prediction set is 77.1%.The preprocessing has less influence on the inversion results,and the choice of regression model has a large degree of influence on the inversion results.In the inversion results,the effect of nitrogen content and chlorophyll content is good,but the inversion effect of water content is poor.The inversion results were better for N content and chlorophyll content.The model accuracy was reduced by introducing vegetation index variables into the hyperspectral inversion model of biochemical parameters of apple leaves.In summary,the strategy of baseline correction,multiple scattering correction,smoothing filtering,feature extraction as pre-processing process combined with regression model is feasible in the inversion of hyperspectral biochemical parameters of apple leaves.