Research Of The Anti-interference Model Of Soil Organic Matter By The Characteristics Of Visible Spectroscopy

Soil is the material basis of agricultural production,and soil organic matter is one of the important indicators to measure soil fertility.In recent decades,it has become a mature and effective method to indirectly determine soil organic matter by using visible / near-infrared nondestructive testing technology.Therefore,it is of great practical significance to establish a prediction model of soil organic matter with high accuracy and adaptability based on the obtained visible / near-infrared spectral data,Moreover,the research on soil organic matter is rarely in the visible light range,and the materials for developing the visible spectroscopy range are easy to obtain.Therefore,this paper based on the spectral data analysis of the visible spectroscopy range of 380 ~ 780 nm,through various spectral data processing methods,establishes the organic matter prediction model with accuracy and robustness,which provides a reference for the practical application of hand-held soil organic matter spectrometer supply theoretical model support.Main research contents and conclusions:(1)Two batches of soil were collected.The first batch of 51 soil samples were collected in Shanxi Province(Taigu,Guandi mountain,Youyu),the first 36 samples were from Taigu,and the last 15 soil samples were from Guandi mountain and Youyu.The second batch of soil samples collected 107 samples from 6 different regions,Guandi mountain,Zhongyang,Fangshan,Loufan,Ningwu and Taigu.158 original soil samples were collected from the two batches.The soil samples were prepared by natural air drying,screening and drying steps.Then,one part of the soil samples were measured by GB9834-88 standard chemical method and the other part was used for spectrum measurement and processing,the field spec3 spectrometer is used to scan the collected soil samples to obtain the original spectrum curve,and the spectrum data is processed in the visible spectroscopy range.The spectrum processing software is mtalb2014 b.(2)Firstly,Savitzky-Golay convolution smoothing,multi variable scattering correction(MSC),first and second derivative,wavelet denoising are used to preprocess soil spectral data.After S-G smoothing and wavelet denoising pretreatment,the original soil spectrum obviously eliminates the edge noise of the original spectral curve of the soil sample,and improves the signal-to-noise ratio to a certain extent;after the first and second derivative(1-der and 2-der)treatment,it is found that the peaks and valleys of the soil spectral curve are sharpened,which can significantly enhance the spectral resolution and reduce the overlap of adjacent spectral bands,thus greatly To eliminate the interference of background and baseline drift during spectrum acquisition,It can also effectively eliminate the irrelevant noise in the spectral information.Taking the average spectral curve of soil as the standard spectral line,after multiple scattering correction,it is found that the spectral curve is well concentrated near the standard spectral line,which indicates that multiple scattering correction can effectively remove the influence of particle scattering and surface photometric change on the spectrum of soil samples,and enhance the correlation between the spectral data and the measured data Then,the least square model(PLS)is established for the soil spectral data obtained by various preprocessing methods.Compared with the model parameters,it is found that the spectral data model after wavelet denoising has the best effect.The prediction correlation coefficient Rp of the model is 0.8188,and the prediction root mean square error RMSEP is 1.2830.Therefore,the later extraction and modeling of soil spectral characteristic wavelength are based on wavelet denoising noise preprocessing.(3)Firstly,KS(Kennard stone)and SPXY(sample set partitioning based on joint X-Y distance)are used to divide the sample set.Then principal component(PCA),successive projections algorithm(SPA)and Adaptive reweighted sampling(CARS)are used to extract the characteristic wavelengths.The characteristic wavelengths of PCA extraction are 560 nm,390nm and 420 nm.The characteristic wavelengths extracted by SPA are 427 nm,574nm,626 nm,780nm and 640 nm.The characteristic wavelengths obtained by CARS are 536 nm,540nm,601 nm and 604 nm.420nm,531 nm,532nm,539 nm,545nm,612 nm,613nm,615 nm,616nm,617 nm,620nm,then establish least squares(PLS)and multiple linear regression(MLR)models are used to obtain the model statistical parameters of the model.It is found that the optimal prediction correlation coefficient of KS +CARS+ MLR model parameters is 0.8703,the prediction root mean square error is 0.5418,followed by KS +SPA + MLR model,the prediction correlation coefficient is 0.8545,and the prediction root mean square error is 0.5812.(4)In order to retain more original spectral information and make the model more robust,the area around the extracted central wavelength of about 5nm is taken as the independent variable of the prediction model of soil organic matter,and the prediction effect of the model is more significant.The prediction correlation coefficient of KS + CARS+ MLR model is 0.8869,the prediction root mean square error is 0.52874,while the prediction correlation coefficient of KS + SPA + MLR model is 0.8668,and the prediction root mean square error is 0.55224.Compared with the central wavelength area as an independent variable,the prediction correlation coefficient is improved,and the prediction root mean square error is reduced.It shows that the accuracy of organic matter prediction model can be effectively improved by using the approximate area substitution of independent variable.(5)Through the extraction of the effective wavelength of soil organic matter,it is found that although there is no obvious water absorption peak in the visible light range,there is a second-order frequency doubling of O-H bond in the visible light range,so it is necessary to remove the influence of water on the prediction of soil organic matter to a certain extent,and establish the anti-interference model in the visible light range.In this paper,50 samples from two batches of dry soil samples were respectively configured as 0%(dry soil),4%,5%,8%,10%,12%,15%,16% and 17% step moisture content.The spectral curves of different moisture content soil samples were reconstructed by using MDI algorithm.It was found that the reconstructed spectral curves of samples were very close to the original spectral curves of dry soil samples.The modeling results showed that the water branch positive energy It can effectively reduce the impact of water on the prediction of organic matter.After the first batch of soil reconstruction,the model prediction correlation coefficient is 0.7836,and the prediction root mean square error is 1.0672;after the second batch of soil reconstruction,the model prediction correlation coefficient is 0.7508,and the prediction root mean square error is 1.1798.Compared with the parameters of the original soil prediction model,the prediction correlation coefficient of the reconstructed model is higher,and the root mean square error is lower.