Hyperspectral PSO-BPNN Estimation Model Of Soil Organic Matter Based On Grey Interval

Soil organic matter is an important component of soil composition and plays an important role in plant growth and development,soil fertility maintenance and the normal operation of ecosystems.Hyperspectral remote sensing has become an important method for estimating soil organic matter because of its multi-band,narrow band and abundant information.However,there is still a problem of unsatisfactory estimation accuracy.In order to improve the accuracy of spectral estimation of soil organic matter,92 samples of brown soil collected in Tai'an City,Shandong Province,were taken as the research object.The content of organic matter and the outdoor spectral reflectance data were measured.The grey interval and particle swarm optimization neural network algorithm was used to establish the Hyperspectral Estimation Model of soil organic matter.The main research contents and conclusions are as follows:(1)The characteristic factors of soil organic matter were extracted.Firstly,spectral reflectance was processed,including smoothing and eliminating abnormal samples.On this basis,in order to improve the correlation between organic matter and spectral reflectance,12 kinds of mathematical transformations such as square,square root,reciprocal,logarithmic,first-order differential and their combination were carried out,and the characteristic factors were extracted by the principle of maximum correlation.The results showed that the correlation between organic matter and spectral reflectance can be significantly improved by first-order differential and second-order differential transformation.The first-order differential at575 nm,1379 nm,2127 nm,2341 nm and 793 nm of second-order differential were selected as the characteristic factors of organic matter content inversion.(2)A high-spectral PSO-BPNN estimation model for soil organic matter based on grey interval was established.Firstly,aiming at the defect that the neural network is easy to fall into local minimum,a hyperspectral estimation model of soil organic matter based on particle swarm optimization(PSO)neural network was established,which can search for the global optimal solution.Then,the grey interval was calculated by the predicted value obtained by the particle swarm optimization neural network algorithm,and the modelsamples were further screened by the grey interval and correlation principle.Then,the particle swarm optimization neural network model based on the grey interval was established to quantitatively estimate the organic matter content,and the results were compared with those of the traditional modeling method.The results showed that the decision coefficient of PSO neural network model based on grey interval is 0.8826 and the average relative error is 3.572%,while that of traditional PSO neural network model is 0.853 and the average relative error is 4.34%.The average relative errors of BP neural network model,support vector machine and multiple linear regression model are 8.79%,6.717% and 9.468%,respectively.This indicates that the particle swarm optimization neural network model based on grey interval is effective and provides a feasible method for hyperspectral estimation of soil organic matter content.