Hyperspectral Inversion Of Maize Physiological And Biochemical Parameters Based On Different Sensors From Air-Ground

Maize is one of the main food crops in China.It is of great significance to obtain the physiological and biochemical parameters of maize in Guan Zhong area in real time for agricultural development and food security in China.In this study,maize in Guan Zhong area was taken as the research object.The portable ground object spectrometer（SVC HR-1024i）and hyperspectral camera（S185）were used to obtain the two types of spectral remote sensing data of maize canopy in different growth stages of Guan Zhong ground and low-altitude UAV respectively.The nitrogen balance index（NBI）and anthocyanin content（Anth value）of maize canopy were measured synchronously by Dualex scientific+^TM.Single-factor modeling parameters were extracted from the primary spectrum and its transform spectrum,and multi-factor modeling parameters were extracted from different types of spectrum with the combination of successive projections algorithm and different types of spectrum.NBI and Anth values were estimated by different inversion methods,and an inversion model construction method based on the fusion data of different sensors was proposed.The results are as follows:（1）SVC HR-1024i and S185 had similar characteristics of primary spectral curves of maize canopy leaves at different growth stages,and SVC HR-1024i had stronger anti-interference ability.The NBI and Anth values of maize leaves had different indicators for the primary spectral curves of SVC HR-1024i and S185.The reflectance is lower in visible region（450-700 nm）and higher in near infrared region（750-850 nm）,which are connected by Red Edge（700-750 nm）.The primary spectral curve of SVC HR-1024i is stable with small error,while the primary spectral curve of S185 is susceptible to the influence of soil background factors.When the maize canopy is sparse,the addition of soil information improves the reflectivity of the primary spectral curve in the visible region as a whole.The NBI of maize leaves is inversely proportional to the reflectance of the visible region of the primary spectral curve,and directly proportional to the reflectance of the near infrared region,that is,with the increase of NBI,the reflectance of the visible region decreases while that of the near infrared region increases.The Anth values of maize leaves is proportional to the reflectance of the primary spectral curve,that is,when the Anth values increases,the reflectance of the visible and near infrared regions will increase.（2）Spectral transformation has a certain potential in the inversion of maize leaf NBI and Anth values.In terms of inversion of maize leaf NBI,based on the different growth stages of the SVC HR–1024i best characteristic wavelengths are in the transform spectrum,tasseling stage from get rid of the spectral envelope,multi-factor modeling parameters based on S185tasseling stage of multi-factor modeling parameters from the first derivative spectrum,data fusion,the optimal single factor model parameters in the first derivative spectrum;In inversion of maize leaf Anth values,based on the different growth stages of the SVC HR–1024i maize best characteristic wavelengths are located on the first derivative spectrum,jointing stage,the flare opening stage and tasseling stage multi-factor modeling parameters from transform spectrum,based on the best features of jointing stage of S185 band and multi-factor modeling parameters in transform spectrum,data fusion,The optimal single-factor modeling parameters are located in the envelope removal spectrum.（3）The new spectral index and characteristic band are superior to vegetation index,spectral area parameter and spectral location parameter;The successive projections algorithm has a better dimension reduction effect,and S185 has a better dimension reduction effect than SVC HR-1024i.The algorithm does not modify the spectral values,and the selected parameters have clear physical meaning and good interpretation.The modeling parameters of optimal NBI and Anth values in different growth stages were new spectral indices and characteristic bands.Based on SVC HR-1024i,the NBI and Anth values of maize in different growth stages were 4-27,and the dimension reduction ratio was above 73%.Multi-factor modeling parameters of NBI and Anth values of maize at different growth stages based on S185 ranged from 3 to 9,with dimension reduction ratios above 91%.After data fusion,the number of multi-factor modeling parameters extracted by the algorithm ranges from 3 to 25,and the dimension reduction ratio was above 75%.（4）The best inversion periods of maize leaf NBI and Anth values were located in the middle stage of maize growth stages,i.e.,flare opening stage and tasseling stage.The multi-factor model is better than the single factor model and the machine learning algorithm is better than the multiple linear regression.In the NBI inversion models of maize leaves,the optimal model based on SVC HR-1024i was the SSA-ELMR model in the flare opening stage,and its modeling R² and validation R² were 0.91 and 0.89,and RPD were 3.17 and 2.99,respectively.The optimal model based on S185 was SSA-ELMR model at tasseling stage,and its modeling R² and validation R² were 0.70 and 0.69,and RPD was 1.54 and 1.51,respectively.In the inversion models of maize leaf Anth values,the optimal model based on SVC HR-1024i was SSA-ELMR model at tasseling stage,with modeling R² and validation R² of 0.84 and 0.89,and RPD were 2.28 and 2.71,respectively.The optimal model based on S185 was SSA-ELMR model at tasseling stage.The R² of modeling and validation were 0.69 and 0.71,and the RPD was 1.51 and 1.73,respectively.The results of RPD show that the optimal inversion model of NBI and Anth values established based on SVC HR-1024i has excellent prediction ability,and the optimal inversion model of NBI and Anth values established based on S185 has rough prediction ability.（5）It is feasible to invert NBI and Anth values of maize leaves based on the fusion data of different sensors.By integrating the data of SVC HR-1024i and S185 at tasseling stage,the modeling R² and validation R² of the optimal NBI inversion model for maize leaves were 0.86and 0.82,respectively,and the corresponding RPD was 2.48 and 1.86,indicating that the model had good prediction ability.The modeling R² and validation R² of the optimal Anth value inversion model for maize leaves were 0.89 and 0.79,respectively,and the corresponding RPD was 2.80 and 2.18,indicating that the model had excellent prediction ability.