Study On Migration Of Soil Total Nitrogen And Organic Matter Near-infrared Spectroscopy Analysis Model

Soil quality is an important factor of ensuring crop yield and quality in agricultural production.Total nitrogen content and organic matter content in the soil are two important criteria for assessing soil quality.Organic matter and nitrogen in soil play an irreplaceable key role in the process of crops growth,development and harvesting.Rapidly obtaining the data of total nitrogen content and organic matter content in soil is of great significance for scientific and precise fertilization and speeding up the development of smart agriculture.The near-infrared spectral component analysis technology has the advantages of rapid non-destructive analysis,which could be applied to large-scale,large-area soil component detection.According to the different ways of beam splitting,near-infrared spectroscopic instruments fall into three categories:dispersive type,filter type and interferometric type.At present,there is no available near-infrared spectroscopy model database of soil organic matter and total nitrogen content at home and abroad.There are two main reasons:Firstly,the number of soil samples is not large enough to build the near-infrared spectroscopy analysis models.Secondly,the existing near-infrared spectroscopy analysis model can only be used for the same model or even the same instrument,and it cannot be transferred to general use,which also makes the application of near-infrared spectroscopy analysis face high cost and repeated modeling issues.Aiming at the above problems,this thesis conducted a systematic and in-depth study on the quantitative near-infrared spectroscopy analysis model of soil organic matter and total nitrogen.Based on two different spectrometers with different principles and structures,this article studies the possibility and accuracy of different kinds of spectral transformation.What we have done are as follows(1)We analyzed the principles and differences between the two spectrometers(Avantes fiber spectrometer and Bruker Fourier spectrometer),and compared the spectral analysis models of the near-infrared fiber spectrometer and the Fourier transform near-infrared spectrometer.The differences and causes of the spectral data of the two spectrometers are summarized.The original data of two different spectrometers were used for modeling,and it was verified that the spectroscopic data obtained by different spectrometers can obtain a very similar prediction effect.(2)The mapping relationship was established between the two spectrometers,so that the data obtained by one spectrometer can be predicted using the prediction model established by the other spectral data to obtain a better prediction effect.Because the resolution of the two spectrometers is different,it is necessary to transform the spectral data.In this paper,empirical mode decomposition method was used to process Avantes spectral data.The processed data were subjected to principal component analysis to extract features,which were transformed into the principal component analysis features of Bruker original data by BP neural network.(3)We used different spectral prediction models to compare the prediction error and accuracy of Bruker spectrum and the transformed Avantes spectrum.Experimental results show that the predicted effect of the Avantes spectral data set after transformation is closer to that of Bruker spectral data.However,there are still some differences between the transformed Avantes spectral data and the Bruker spectral data.In the future research,we can try to improve the transformation effect by eliminating these differences,or use these differences to improve the prediction accuracy.