Research On Extraction Methods Of Black Soil Nutrient Information By Hyperspectral Remote Sensing

With the development of information technology in the field of soil science,hyperspectral remote sensing plays a more and more important role in the quantitative remote sensing monitoring of soil composition,with its high spectral resolution,strong band continuity and high spatial resolution.Soil nutrients are mineral nutrients that can be absorbed directly or through the roots of the plant,including nitrogen,phosphorus,potassium,calcium,magnesium,sulfur,iron,boron,molybdenum,zinc,manganese,copper and chlorine.The organic matter contained in the black soil of Northeast China is also a universal nutrient.In this paper,the objective,application direction,data acquisition,processing,analysis and extraction models of hyperspectral technology in the extraction of soil nutrients are explored.The key technologies of hyperspectral soil nutrient extraction are summarized,and the comprehensive technical solutions of "theoretical layer-layer-object layer-reasoning layer-storage layer" are formed and summarized from the aspects of soil spectral acquisition,processing,analysis and modeling.(1)The nutrient prediction method based on information content in black soil was studied.The model accuracy based on three methods based on mechanism,band standard deviation and information entropy is compared.The study shows that organic matter and nitrogen use information entropy,characteristic bands are 705 nm,714 nm,733 nm,657 nm and 743 nm;phosphorus and potassium use band standard deviation,characteristic bands are 915 nm,924 nm,905 nm,886 nm and 895 nm,and the model essence is built.The degree is relatively high.S3,N3,P2 and K2 methods have high accuracy in the inversion of organic matter,nitrogen,phosphorus and potassium,which proves the quantitative effect of information method on the extraction of nutrients in black soil.(2)A new method for predicting nutrient content by spectral parameters has been established.The 18 sub indexes of spectral statistics,spectral eigenvalues and spectral information can reflect the comprehensive spectrum of soil spectrum,which is an effective spectral training data.The algorithm with the highest precision of extracting the information of organic matter and total potassium is the neural network method,the error is 1.21% and 0.81% respectively.When the support vector machine algorithm extracts the information of total nitrogen and total phosphorus,the measured mean and the mean value of the sample are in perfect agreement,and the accuracy is the highest.The comprehensive accuracy of soil nutrient extraction by aero hyperspectral extraction was evaluated.The extraction errors of organic matter,total nitrogen,total phosphorus and total potassium were 5.25%,6.05%,2.74% and 8.90% respectively,and the accuracy was the highest in the total phosphorus inversion.(3)The response relationship between spectral transformation method and nutrient content was evaluated.The original spectral reflectance data are processed as re sampling RE,logarithmic reciprocal LR,first order differential FD,envelope removal CR and multivariate scatter correction MSC.The transformation method of the optimum extraction precision of each black soil and the difference of the extraction precision of the five spectral transformation methods were obtained,which provided a quantitative basis for mastering the relationship between the spectral transformation and the response of the nutrient content in the black soil.The results showed that MSC,MSC,LR and RE spectral transformation methods were applied to the combination of organic matter,nitrogen,phosphorus and potassium,and the spatial distribution precision of the soil nutrient content was the highest.(4)A frequency domain identification method for nutrient prediction in black soil is proposed.Based on the analysis of the characteristics of the black soil spectrum,an adaptive Gauss low pass filtering algorithm is designed.By eliminating the data of different energy levels gradually,the purpose of classification and classification of black land is achieved.The method combines the spatial context information and spectral information of the hyperspectral pixel of black land,and breaks through the spatial information of the image to break through the performance bottleneck of the traditional pixel by pixel spectral classification.The cut-off frequency optimal model is established,which smoothes the contradiction between variance and residuals,and balances image smoothing with detail preservation.(5)A machine learning method for nutrient extraction from hyperspectral black soil is explored.The intelligent extraction of black soil nutrient information is introduced by introducing the strategy of machine learning and solving problems into the intelligent extraction of nutrient information of black soil.By building feature vectors,logical statements,rules,semantic networks and frameworks,the intelligent extraction of nutrient information of black soil is realized.The extraction technology has shown good application value in intelligent reduction of nutrient information based on spectrum,intelligent clustering,intelligent classification,intelligent regression and so on.In the field of soil quantitative remote sensing,hyperspectral remote sensing technology has been in the forefront.It is pointed out that the information indicated by the spectrum can not only provide rapid information for soil composition,but also the information extraction model based on the measured data,which is the theoretical basis of software R & D,instrument development and soil quality evaluation.With the continuous development of hyperspectral technology,the research of soil optics will be characterized by fine differentiation,specialization,integration of hardware and software,small modularization,and real-time information extraction.In order to adapt to this trend,the establishment of intelligent method for extracting nutrients from black soil will be the focus of this field in the future.