Hyperspectral Estimation Of Maize Physiological And Biochemical Parameters Based On Fractional Differential

As one of the major food crops in China,it is of great significance to ensure the high-quality production of maize for our country’s food security.This study takes summer maize in Qinan Village,Qian County,Xianyang City,Shaanxi Province,China as the research object,based on the SVC HR-1024i and the S185 hyperspectral camera equipped with DJI UAV M600 Pro,we obtained the hyperspectral data of maize at the jointing,tasseling,milk-ripe and ripening stages in the study area,and measured the anthocyanin content and chlorophyll content of maize canopy leaves simultaneously using a Dualex scientific+^TM and SPAD-502,respectively.The fractional order differential spectra of order 0～2 in steps of 0.2 and the construction of 10 common vegetation indices were calculated to analyze the correlations with anthocyanin and chlorophyll,and simple single-factor functional regression models and sparrow search algorithms were developed to optimize the random forest（SSA-RFR）multi-factor estimation models,and the accuracy of various models was analyzed and compared.The main results obtained were as follows:（1）The hyperspectral reflectance curves of maize leaves at different fertility stages were consistent,with the high to low reflectance in the visible band for the milk-ripe,jointing,tasseling and ripening stages,and the high to low reflectance in the NIR band for the milk-ripening,tasseling,jointing,and ripening stages,respectively.Anthocyanin content was directly proportional to the spectrum,the“red edge”shifted towards the short wave direction when the content of anthocyanin increased;chlorophyll content was inversely proportional to the spectrum,the“red edge”shifted towards the long wave direction when the content of chlorophyll increased.The SVC hyperspectral and the UAV hyperspectral were transformed by fractional order differential spectroscopy with the same pattern,with the increase of order,the spectral reflectance gradually decreases and finally converges to 0.（2）The correlation between fractional order differential transform spectra and anthocyanin content and chlorophyll content of maize was elevated at different stages by at least 7.77%for anthocyanin and 22.14%for chlorophyll.The order of the maximum absolute values of correlation coefficients differed among different stages,with the order of 1.2,1.6,1.4 and 0.8 for the maximum absolute values of anthocyanin correlation coefficients,and the order of 1,1.6,1.2 and 2 for the maximum values of chlorophyll correlation coefficients.The optimal sensitive vegetation indices with both anthocyanins and chlorophylls were GDVI and GNDVI,and the correlation enhancement was not as good as the fractional order differencing technique,but the construction method was simpler and more stable.（3）Single-factor estimation models for anthocyanin and chlorophyll showed consistency across different stages,with models based on fractional differential spectrum characteristic band better than those based on optimal vegetation indices.The optimal model for each fertility stage was mainly based on a binomial function,with milk-ripen and ripening stage being the single-factor models with the highest accuracy.Therefore,a binomial function model could be preferred for estimating maize physiological and biochemical parameters,and fertility stages could be preferred to the milk-ripe and ripening stages.（4）The accuracy of multi-factor models based on fractional order differential spectra was significantly higher than those of multi-factor models based on primary spectra and single-factor models.The best estimation model for anthocyanin content of maize leaves was of order1.6 at all fertility stages,with the best performance at the jointing stage（R²=0.818,RMSE=0.007,MRE=5.899%）and the highest enhancement over the primary spectrum at the tasseling stage.The best estimation model for chlorophyll content of maize leaves varied in order of 0.8,1.6,1.2 and 0.8 for each fertility stage,again with the best performance at nodulation（R²=0.775,RMSE=1.475,MRE=2.345%）and the highest improvement over the primary spectrum.（5）Based on UAV imaging hyperspectral combined with all transformed spectral features bands to establish maize canopy chlorophyll content not only can integrate all fractional order differential spectral response features,but also significantly improve model prediction accuracy and visualize the spatial distribution pattern of maize canopy chlorophyll content.