Estimating Soil Total Nitrogen And Organic Matter Based On Hyperspectral

Soil is the basis of the growth of crops.Total nitrogen and organic matter are the main nutrient contents,which plays an important role in crop growth.Traditional methods for measuring soil nutrient are time-consuming,laborious,and time-inefficient,while the quantitative estimation model of soil total nitrogen and organic matter is constructed by using hyperspectral remote sensing technology,which can effectively improve the monitoring efficiency of soil nutrients and provide an important basis for soil informatization management and accurate fertilization of crops.In this study,347 samples were collected from four main soil types in Henan Province.The high spectral reflectance curves of soil samples were obtained by using hyperspectral remote sensing technology,Then The original spectral curve was transformed by 10 spectral forms such as the first derivative,the reciprocal,and the first derivative of the reciprocal et al.It use that correlation analysis(CA)and genetic algorithm(GA)to select the characteristic band of soil total nitrogen and organic matter,combined with Four modeling methods of Partial Least-squares Regression(PLSR),Support Vector Machine(SVM),Back Propagation Neural Network(BPNN)and Random Forest(RF)to constructed hyperspectral estimation models for soil total nitrogen and organic matter respectively.The specific results are as follows:(1)According to the statistical analysis of total nitrogen and organic matter of five sampling area samples by the traditional statistical method,the TN of Luohe and Shangshui county is above the middle level,and the TN of the Huaxian,Xinyang and Pingdingshan country is below the middle level;the SOM of Shangshui country is above the middle level,and the SOM of four areas of luohe,chuanxian,xinyang and pingdingshan is below the middle level.In terms of the coefficient of variation,both TN and SOM are at medium variation level in the five regions.(2)In addition to obvious absorption peaks near 1400 nm,1900 nm,and 2200 nm,the overall spectrum of the soil spectrum was relatively flat.The spectrum curve of different soil types is roughly the same,but the reflectivity has a certain degree of difference.The nutrient content also affects the reflectivity.For example,Shajiang black soil has high organic matter content but the overall reflectivity is relatively low.The trend of the original spectral curve of fluvo-aquic soil,paddy soil,and cinnamon soil has no obvious difference,but there are multiple intersections.(3)The characteristic bands of soil total nitrogen and organic matter were determined.The correlations between the original spectrum and its various transformation forms and the contents of total nitrogen and organic matter were respectively analyzed,and The best performance was the reciprocal derivative of the reflectance logarithm(1/lgR),which was the most and most obvious the peak point of the correlation coefficient of total nitrogen and organic matter,and the correlation coefficients with total nitrogen and organic matter reached 0.79 and 0.81 respectively.The waveband of the 10 peak point of the correlation coefficient of(1/lgR)and total nitrogen and organic matter is used as the characteristic band.When GA is used to select the characteristic band of total nitrogen and organic matter for the original spectrum,the size of spectral interval is 10 nm.According to the frequency of the band selection and composition of contribution,spectral interval band are arranged in descending order,and finally the original spectral band of the spectral band of the top 10 is used as the characteristic band.(4)Based on the characteristic bands,four modelling methods which are named as PLSR,SVM,BPNN and RF were used to construct the estimation models for total nitrogen and organic matter content.The validation precision RPD of the model built using RF model is greater than 2,indicating that RF can be used in the construction of high spectral estimation model of total nitrogen and organic matter and has good estimation ability.Except that the accuracy of the model constructed based on GA-RF is less than that of the model constructed by CA-PLSR,the accuracy of the estimation models of the three nonlinear modeling methods of SVM,BPNN and RF is higher than that of the linear model PLSR.And except that the accuracy RPD of the model based on CA-PLSR is less than 2,the accuracy RPD of other models is greater than 2,which indicates that these model has good estimation ability for total nitrogen and organic matter.In addition,the model based on BPNN has the highest estimation accuracy in all models,therefore BPNN is the best model of the overall modeling results.