Background reduction method for soil XRF spectrum based on XRF-EGAN model

A background reduction method for soil XRF based on an XRF-EGAN model, the background reduction method for soil XRF comprises:

• constructing a generator of a model by using a one-dimensional fully convolutional network layer and a residual connection, based on a design mode of a GAN model,
• constructing a discriminator of the model by using one-dimensional convolution and a fully connected layer, and
• obtaining a trained generator and a trained discriminator by training the XRF-EGAN model using an adversarial training mode,
• wherein, the generator is a soil XRF background reduction model, which in turn improves a correlation between a net peak area and content of element of soil XRF,
• wherein, XRF denotes X-ray Fluorescence, and the XRF-EGAN model denotes an XRF background reduction model built based on a Generative Adversarial Neural Network (GAN) model;
• and wherein the background reduction method for soil XRF comprises:
• step 1:
• collecting XRF spectra data of a soil sample by using an XRF analyzer, and manually subtracting the background of the XRF spectra data of the soil sample to finally obtain the soil XRF spectra data Datanoisy before background reduction and the soil XRF spectra data Dataclean without the background;
• step 2:
• training the XRF-EGAN neural network model via the collected Datanoisy data and Dataclean data, and saving network model parameters of the generator of an optimal XRF-EGAN model after completing the training;
• step 3:
• loading the network model of the generator of the XRF-EGAN model, performing XRF spectra background reduction by using the network of the generator of the XRF-EGAN model for new soil XRF spectra data measured by the XRF analyzer, and obtaining an output after the background reduction;
• wherein, before inputting soil XRF spectra data x with background into the XRF-EGAN neural network model in step 2, preprocessing the soil XRF spectra data x, and an expression for preprocessing the soil XRF spectra data x is as follows:
• y
• i
• =
• {
• log
• 2
• x
• i
• ,
• x
• i
• ≥
• 1
• 0
• ,
• x
• <
• 1
• ,
• x
• i
• ∈
• R
• ,
• i
• =
• 1
• ,
• 2
• ,
• …
• ,
• 2048
• where xi denotes a count value of the i-th channel of 2048 channels of the XRF spectra, logarithm of the input soil XRF spectra data x is taken, and then a logarithmically taken result y=[y1 y2 . . . y2048] is maximum-value and minimum-value normalized, and a mathematical expression is as follows:
• z
• i
• =
• y
• i
• max
• ⁢
• (
• y
• )
• -
• min
• ⁢
• (
• y
• )
• ,
• i
• =
• 1
• ,
• 2
• ,
• …
• ,
• 2048
• where y denotes an output result of the input soil XRF spectra data x after taking the logarithm;
• zi denotes a normalized result of a result yi of the i-th channel after taking the logarithm.