Spectral Feature Extraction And Partial Least Squares Quantitative Inversion Based On EMD And Reverse Spectral Absorption Integral

In recent years,with the development of remote sensing technology,hyperspectral remote sensing has the characteristics of "high spectral resolution and large number of bands",which provides the possibility of rapid,fine and accurate monitoring of chromium content in soil.At the same time,the high dimensional spectral structure brings new challenges to data processing-the high correlation between adjacent bands leads to the redundancy of band information,and takes up a large amount of storage space and increases the computational burden.On the basis of the spectral formation principle of ground objects and the quantitative inversion theory of remote sensing,this paper studies the spectral feature extraction,characteristic parameter construction and quantitative expression of soil.Using indoor spectral curve of the northeast black soil zone,starting from the spectral absorption image interpretation,in order to solve the problem of the strong correlation between adjacent band,draw lessons from the spectral integral,constructs the new reverse spectral absorption spectrum characteristic parameter,integral,and for the first time with the aid of empirical mode decomposition to extract soil spectral characteristics,through the characteristics of the method of stepwise screening,finally based on partial least-squares regression to establish the multiple inversion model of chromium content in the soil,the contrast analysis of the different spectral feature extraction method to build the influence of the inversion model.The main research contents and results are as follows:(1)Study feature extraction and quantitative inversion based on spectral transformation.The results show that spectral transformation can improve the correlation between spectral variables and chromium content,and the characteristic bands of soil are mainly located near 1420,1920,2210,2251 and 2425 nm.Compared with other partial least square inversion models,square-root first-order differential transformation can improve the precision of soil chromium inversion model.(2)Feature extraction and quantitative inversion based on spectral characteristic parameter--absorption area were studied.According to the structure absorption area at the absorption position of the normalized spectral curve of soil after envelopment treatment,the absorption area at 565~810 and 1863~2153nm has a significant influence on the content of chromium.However,the inversion model results are not ideal,and the content of chromium in soil can only be roughly estimated.(3)Study feature extraction and quantitative inversion based on reverse spectral absorption integral.Starting from the image interpretation of spectral absorption,in order to overcome the soil spectral curve between adjacent band strong correlation,spectrum integral,construct new reverse spectral absorption spectrum characteristic parameter,integral,which had a great effect on the chromium content of extract of band position is mainly focused on the 468,1757,1874,1923,2205,2250,2425 nm,near the PLSR model can accurately predict content of chromium in soil in the study area.(4)Research EMD-based feature extraction and quantitative inversion.Based on the fact that hyperspectral data are "non-linear" and "non-stationary" signals,empirical mode decomposition is introduced to de-noising and feature extraction of soil spectral curves.The basic principle of empirical mode decomposition and the formation of eigenmode function and its significance in spectral curve decomposition are deeply studied.Through comparative analysis,it was found that the intrinsic modal functions IMF3 and IMF4 obtained from the soil spectral curve after EMD decomposition could best reflect the spectral characteristics.On the one hand,the values corresponding to the absorption positions in IMF3 and IMF4 were extracted from the residual R2,and the characteristic bands were extracted as 2012,411,1972,2388 and 2420 nm,to establish the partial least squares inversion model of chromium.On the other hand,in the residual R2,the inverse spectral absorption integral is constructed with the absorption positions in IMF3 and IMF4 as the central bands,and the characteristic bands are selected,and the partial least squares inversion model is established.The precision and stability of the R2-based inverse spectral absorption integral model have been significantly improved by synthesizing the evaluation indexes of various models,indicating that EMD has obvious advantages and potentials in the application of soil spectral feature extraction and soil chromium inversion.