LIBS multi-component quantitative inversion method based on deep learning convolutional neural network

Abstract

The invention discloses an LIBS multi-component quantitative inversion method based on a deep learning convolutional neural network, and is suitable for the field of laser spectrum analysis. Accordingto the method, the unique advantages of the convolutional neural network algorithm in the aspect of image feature recognition are utilized, and the method is applied to LIBS spectrum quantitative inversion. According to the convolutional neural network construction scheme designed by the invention, feature extraction and deep learning can be performed on the LIBS spectral line form of the sample,and after the convolutional neural network is trained by using the LIBS spectrum of the known sample, the network can analyze and predict the contents of various chemical components of the unknown sample at the same time. The method has the advantages of being simple and convenient to operate, efficient in training, high in accuracy and good in robustness, is suitable for quantitatively analyzingthe LIBS spectrum, and is particularly suitable for analyzing the LIBS spectrum with relatively high spectral line form complexity and relatively large interference noise.

Description

technical field [0001] The invention relates to the technical field of laser spectrum analysis, in particular to a laser-induced breakdown spectrum analysis method based on a deep learning convolutional neural network algorithm, which can simultaneously quantitatively invert the chemical composition contents of various substances in a sample. Background technique [0002] Laser-induced breakdown spectroscopy (LIBS) is an in-situ, minimally destructive and efficient method for analyzing the chemical composition of substances. It is widely used in environmental monitoring, biomedicine, industrial testing, deep space exploration and other fields. LIBS is relatively mature in the qualitative analysis of material components, but there are still problems in the quantitative analysis of component content, such as insufficient accuracy, large errors, low stability, and low repeatability. This is mainly caused by matrix effects, self-absorption effects, and experimental parameter eff...

AI Technical Summary

Problems solved by technology

[0005] 1. For the traditional linear analysis method: 1) Its quantitative inversion method completely depends on the wavelength position and peak intensity of the characteristic peak of the plasma radiation, and to accurately identify and calibrate the characteristic peak position and peak, it is necessary to carry out a certain analysis of the spectrum. A series of conventional or special preprocessing, the preprocessing process not only adds additional time-consuming, but also has no uniform standard method, which hinders the result comparison and cross-validation

2) Since many physical processes of laser-induced plasma are actually highly nonlinear, they will also have intricate effects on the final LIBS spectral shape and spectral line intensity. In quantitative analysis at the same time, even after a series of spectral preprocessing, the accuracy of traditional linear methods often still cannot meet the needs of practical applications

[0006] 2. For the ordinary BPNN method: 1) Each layer of BPNN is a fully connected layer, so for a large network with many nodes, the number of weights that need to be trained is extremely large, and the training difficulty and time-consuming are unbearable

2) The robustness of the BPNN method is poor, and a high accuracy can only be guaranteed when the complexity of the LIBS spectral shape is low and the interference noise is small

2) It is necessary to perform various preprocessing on the spectrum, especially the principal component analysis method is required to reduce the dimensionality of the spectral data, and the additional time-consuming caused by preprocessing is more

3) The core structure of the CNN network used is single, and the analysis object in this paper is only a single component (potassium element), so this simple structure can still have high accuracy. The complexity of processing and the difficulty of training will increase significantly, and the accuracy will not be guaranteed

4) The training efficiency of the programming language (MATLAB) used for calculation is not high enough, it is not suitable for large-scale networks with a large number of nodes, it is not suitable for complex networks with diverse structures, and it cannot meet the needs of efficient deep learning

Images