Study On The Grey Correlation Model Of Soil Available Nitrogen Content In Hyperspectral

The content of available nitrogen in soil is one of the important nutritional indexes to measure soil fertility.It is of great significance to develop precision agriculture to detect the soil nitrogen content in a large range accurately and quickly by using hyperspectral technology.Although the traditional method of measuring soil nitrogen content is high precision,the process is tedious and time consuming,and it is difficult to meet the needs of monitoring soil nitrogen content in large space;The advantages of hyperspectral remote sensing,such as many bands,narrow band and large amount of information,make it a new technology of soil nitrogen content monitoring.The spectral measurement of soil will be disturbed by many factors,which leads to uncertainty in the spectral estimation of soil nitrogen content and the low accuracy of the estimation.Therefore,this paper takes 70 soil samples collected in Hengshan County of Shaanxi Province as the research object,and measures the data of indoor spectral reflectance and available nitrogen content of soil,and studies are carried out by statistical analysis and grey correlation analysis.The research contents and conclusions of the paper are as follows:(1)The spectral reflection characteristics of different available nitrogen contents in Hengshan County were studiedAccording to the collected soil samples,the content of available nitrogen and spectral reflection data of soil were measured;The spectral curve was pretreated and the spectral reflection characteristics of soil available nitrogen content were analyzed.The results showed that the content of available nitrogen in soil had a significant negative correlation with reflectance,which was consistent with the research results of many scholars and was in line with the scientific.The difference of the spectral curve interval of different available nitrogen content decreases with the increase of available nitrogen content in soil.(2)The sensitive band and characteristic factors of soil available nitrogen were determinedIn order to determine the sensitive band of soil available nitrogen,nine spectral combination transformation technologies,such as logarithm and logarithm,are used to transform the spectrum.The correlation analysis of soil available nitrogen content is carried out for the original and transform spectra.The basis of extracting characteristic factors is the principle of great correlation.The results show that the correlation between available nitrogen content and reflection spectrum can be improved by first order differential and simple combination transformation technology in visible light and near infrared,while other transformation methods have little effect on the correlation between reflection spectrum and soil available nitrogen content;Five characteristic factors are determined at 387 nm and1932nm of first order differential transformation of logarithm,2135 nm,2301nm of first order differential transformation of square and 2420 nm of square root.The correlation coefficients of five characteristic factors are-0.590,-0.583,-0.693,0.567 and 0.450 respectively.(3)A local regression estimation model for hyperspectral grey correlation analysis is establishedBased on the gray characteristics of soil available nitrogen content estimation,the paper constructs the weighted grey distance weight vector based on the correlation between the characteristic factors and dependent variables.On this basis,the grey correlation degree model of weighted distance is constructed,and the grey correlation degree between the sample to be estimated and the known model is calculated,and the gray correlation degree is sorted from large to small;On this basis,the grey correlation local regression estimation model is established,and the results are compared with the multivariate linear regression,BP neural network and support vector machine model.The results show that the accuracy of the local regression model of grey correlation degree is higher than that of the common estimation model,among which the decision coefficient R2 is 0.6935 and the average relative error is 22.530%.This shows that the grey correlation local regression model is feasible and effective for the Hyperspectral Estimation of soil available nitrogen content.