Inversion Predict Large Coking Base Soil PAHs Content Based On The Hyperspectral Data

So far,only a handful of scholars in the world used hyperspectral remote sensing technology to predict the contents of PAHs in soil.Using hyperspectral remote sensing technology to predict the contents of PAHs in soil can effectively avoid the shortcomings which include high technology equipment shortage,time-consuming and high cost when analysing contents of PAHs in soil in laboratory.At the same time,it can quickly predict the content of surface soil PAHs in a wide range of research area at a low price,which will provide a scientific basis for soil pollution prevention and management.In the process of inversion,dealing with high dimensional hyperspectral data is inevitable.Using a simple mathematical statistics methods could result the curse of dimensionality,and using data mining requires the user has highly knowledge reserve.Therefore,to establish a graphical user interface which is simple and easy to use is an effective method to solve this problem.Such a graphical user interface can be more convenient and efficient to deal with hyperspectral data for many users.In this paper,agricultural surface soils were sampled in a large coking base located in the middle of Shanxi Province,and a total of 36 samples were collected.16 kinds of optimal control Soil PAHs were analyzed by a GC-MS.Hyperspectral reflectance data of soil samples were collected indoors by ASD FieldSpec4 spectrometer.Spectrum data were transformed to 9 kinds of forms,for example,the first derivative,second derivative,reciprocal,logarithmic,etc.Hyperspectral prediction models which can predict the total content of PAHs in soil are respectively establish by the principal component regression,BP neural network,support vector machine(SVM)and extreme learning machine based on the above data.In addition,using a tool named "guide" created a graphical user interface in MATLAB to establish hyperspectral prediction models by hyperspectral data and one physical and chemical indicator.And the model can be used to inversion that physical and chemical indicator.In this paper,the main research results are as follows:(1)Among the 9 kinds of data form,when reflectivity data is transformed into the first derivative of the reciprocal of reflectivity,the second derivative of the reciprocal of reflectivity or the first derivative of logarithm of the reciprocal of reflectivity,it can prominently display useful information which it contains.(2)Four optimal models are respectively established by the BP neural network,principal component regression,support vector machine(SVM)and the extreme learning machine.All their R2 are greater than 0.7,all RPD are greater than 1.4.It indicates that those four kinds of models can be used to predict PAHs in soil.Among all models,the best model is established by BP neural network with the first derivative of the reciprocal of reflectivity.Its R2 is 0.8840,and RPD is the highest,at 2.03.(3)Graphical user interface is established which integrates data conversion with model establishment and inversion.It is mainly composed of modeling interface,inversion interface and all kinds of model Parameter settings interfaces.It implements a random combination of 9 kinds of data transfer form,3 kinds of proportion of training set and testing set and 4 kinds of modeling method.The graphical user interface can quickly and easily establish hyperspectral prediction models and invoke it for inversion.