Construction Of Soil Organic Matter And Alkali Hydrolyzable Nitrogen Rapid Detection Model Based On Hyperspectral

As an important indicator of soil fertility,soil nutrient content has a significant impact on the growth and development of crops.The traditional methods of obtaining nutrient contents such as organic matter and alkali hydrolyzable nitrogen have a series of problems,such as difficult detection,slow speed and high cost.As an important technology of quantitative remote sensing,hyperspectral technology has become a research hotspot in the quantitative estimation of soil nutrient content.At present,the national demand for soil testing and formulated fertilization is higher and higher.It is expected that the soil nutrient content can be obtained quickly and accurately,and the monitoring efficiency of soil nutrient can be improved,so as to provide an important basis for crop growth and the development of precision agriculture.This study took the farmland soil in Pingdu District,Qingdao City,Shandong Province as the research object.116 soil samples were collected in June 2020,and the nutrient content data of soil organic matter and alkali-hydrolyzable nitrogen were measured in the laboratory using traditional chemical methods,and used The ASD Field4 surface spectrometer acquires the reflectance curve of soil samples indoors.Based on the analysis of soil nutrient content and spectral reflectance distribution characteristics,combined with the(Savitzky-Golay,SG)smoothing algorithm,the spectral reflectance is preprocessed in 10 transformation forms(logarithm,reciprocal,square root,logarithmic reciprocal,The first-order differential of the original spectrum,the first-order differential of the logarithm,the first-order differential of the reciprocal,the first-order differential of the square root and the first-order differential of the logarithm),and the correlation coefficient method is used to screen out the characteristics of soil organic matter and alkali-hydrolyzable nitrogen Band,using BP neural network(BPNN),multiple linear regression(MLR)and partial least squares(PLSR)three modeling methods to construct soil organic matter and alkaline hydrolyzable nitrogen soil nutrient content prediction models,and verify the accuracy of the model.The main findings are as follows:(1)Based on the statistical analysis of the contents of soil organic matter and alkali hydrolyzable nitrogen in the experimental area,the results show that the contents of soil organic matter and alkali hydrolyzable nitrogen in the experimental area are in the third level(above medium level);from the results of coefficient of variation,the coefficient of variation of organic matter and alkali hydrolyzable nitrogen in all soil samples belongs to the medium level.(2)The spectral reflectance curves of all soil samples are consistent,and the spectral reflectance values are in the range of 0-0.6.The reflectivity curve is unstable between 350nm and 2500nm,and there is a relatively good positive correlation between the front wave and the back wave.(3)When using correlation analysis to select characteristic bands.The results showed that among the 10 kinds of spectral pretreatment,the(1/lg R)’pretreatment was the best.The correlation coefficient of spectral reflectance with soil organic matter and alkali hydrolyzable nitrogen content was the most obvious,and the maximum correlation number was 0.61 and0.59,respectively.Finally,(1/lg R)’and the peak of correlation coefficient of organic matter and alkali hydrolyzable nitrogen content were selected as the characteristic bands of soil organic matter and alkali hydrolyzable nitrogen content respectively.(4)Based on the selected characteristic bands,MLR,BPNN and PLSR methods were used to build the nutrient content estimation models of soil organic matter and alkali hydrolyzable nitrogen.From the modeling results,the descending order of modeling accuracy and validation accuracy R~2of the three models of the two nutrient contents is BPNN>MLR>PLSR.It can be seen that the prediction accuracy of the nonlinear modeling method BPNN is better than that of the linear model PLSR and MLR.The accuracy R~2,RMSE and RPD of BPNN based organic matter estimation model were 0.8086,3.7772 and 2.2630,respectively,and the accuracy R~2,RMSE and RPD of BPNN based alkali hydrolyzable nitrogen estimation model were 0.7706,16.7541 and 2.0542,respectively.The relative analysis error(RPD)of the validation set of the two nutrient content estimation models constructed by BPNN is greater than 2,which indicates that the model constructed by BPNN has good estimation ability for soil organic matter and alkali hydrolyzable nitrogen content in the experimental area,and BPNN is the best model.