Study On The Grey Relation Degree Estimation Model Of Soil Organic Matter Based On Hyper-spectral

As an important component of soil,soil organic matter plays an important role in crop growth,soil water and fertilizer conservation and land ecosystem.The traditional measuring method of soil organic matter is sampling and testing,which possesses a high precision,but time-consuming and difficult to implement.Compared with conventional remote sensing,Hyper-spectral remote sensing can accurately capture the subtle spectral reflectance information,so as to realize accurate recognition and retrieval of objects.The extensive use of hyper-spectral remote sensing will be of great significance for the evaluation of crop growth monitoring,land use and precision agriculture.This study takes Tai'an city of Shandong Province as the study area,and takes the organic matter content and outdoor reflection spectrum of 92 soil samples as the research object.According to the grey characteristics of soil organic matter estimation,a grey soil organic matter hyper-spectral estimation model is established by using grey system theory.By comparing with the classical estimation method,the effectiveness of grey prediction model is verified.The main research contents are as follows:(1)The sensitive bands and the inversion factors of organic matter of fluvo aquic brown soil in tai'an were determined.The reflective spectrum is transformed by using 10 kinds of spectral transform technology such as square root,reciprocal,logarithmic,first-order differential and combination transformation and continuum removal.Then organic sensitive bands are determined through correlation analysis between original or transform spectra and soil organic matter content,and the inversion factors is extracted based on the maximum correlation.The results show that,among 10 kinds of transformation techniques,first-order differential and logarithmic-reciprocal-differential technique can significantly improve the correlation between reflectance spectra and the organic matter in the visible and near IR region,while the square root,reciprocal and logarithmic is not conducive to improve the correlation;Spectral characteristics of organic matter are mainly located in the visible light band of 485~760nm,and near infrared band of 1375~1382nm,2121~2133nm,2336-2347 nm which near water absorption band;the 5 inversion factors selected are respectively located in the original spectra of 665 nm,first derivative spectra of 575 nm and 2341 nm and the logarithm reciprocalfirst-order differential spectra of 1378 nm and 2128 nm,and their correlation are all greater than 0.55.(2)The grey estimation model of soil organic matter based on hyper-spectral is established.Based on the grey characteristics of soil organic matter estimation and non time series characteristics of spectral factors,the classic grey relation degree is improved to weighted distance grey relation degree and grey weighted relation degree by using the correlation coefficient between the characteristic indexes and the dependent variable and the standard deviation of the characteristic indexes.Then,two kinds of grey relation estimation model with residual modification are obtained according to the residual modification model which established from recognition error.On this basis,the standard deviation of corrected value residuals are used to extend the point estimation to interval estimation.Finally,the grey estimation model proposed in this paper is compared with multiple linear regression model,BP neutral network model and support vector machine model to verify it's validity.The results show that,both the weighted distance grey relation degree and grey relation degree can be used to estimate soil organic matter based on hyper-spectral,residual correction model can improve the estimation accuracy effectively,and interval estimation can not only reduce effect the uncertainty,but reflect the dynamic characteristics of organic matter;Among the five kinds of point estimation models,grey weighed relation degree estimation model with residual modification and weighed distance grey relation degree estimation model with residual modification possess the highest accuracy,who's average relative error is 6.79% and7.94% respectively,followed by support vector machine,who's average relative error is12.94%.BP neural network and multivariate linear regression model shows poor performance and their average relative error are all beyond 14%.he results show that the grey correlation estimation model has great potential in the estimation of soil organic matter.