CN116030310A - Sample classification method and system based on laser-induced breakdown spectroscopy technology - Google Patents
Sample classification method and system based on laser-induced breakdown spectroscopy technology Download PDFInfo
- Publication number
- CN116030310A CN116030310A CN202310225705.2A CN202310225705A CN116030310A CN 116030310 A CN116030310 A CN 116030310A CN 202310225705 A CN202310225705 A CN 202310225705A CN 116030310 A CN116030310 A CN 116030310A
- Authority
- CN
- China
- Prior art keywords
- sample
- spectrum data
- laser
- induced breakdown
- adopting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a sample classification method and a system based on a laser-induced breakdown spectroscopy technology, which relate to the technical field of sample classification and comprise the following steps: measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method, and collecting original spectrum data; preprocessing the collected original spectrum data to obtain net spectrum data; performing dimensionality reduction on the net spectrum data by adopting a principal component analysis method, extracting characteristic variables, and constructing a training data set; adopting a machine learning algorithm to construct a sample classification model; training the sample classification model by taking the training data set as input and the corresponding class label of the known sample as expected output to obtain an optimal classification model; and identifying the class label of the sample to be detected by using the optimal classification model. The invention combines the laser-induced breakdown spectroscopy technology with the machine learning algorithm, can realize quick and accurate classification of unknown samples, improves the classification recognition effect, has better stability and reduces the calculated amount.
Description
Technical Field
The invention relates to the technical field of sample classification, in particular to a sample classification method and system based on a laser-induced breakdown spectroscopy technology.
Background
Laser Induced Breakdown Spectroscopy (LIBS) is a rapid chemical analysis tool where powerful laser pulses are focused on a sample to form a microplasma from which elemental or molecular emission spectra can be used to determine the elemental composition of the sample. With the rapid development of lasers and the generation of high-sensitivity optical detection equipment, LIBS technology is unprecedented and is widely applied to the field of material detection, such as water pollution, soil analysis, industrial evaluation, food safety, environmental monitoring, archaeological relics, medical analysis and the like, multiple elements can be analyzed simultaneously, almost all samples (solid, liquid and gas) can be analyzed, remote non-contact detection is realized, and classification of the samples can be realized by collecting spectral data of the samples and a classification algorithm.
For example, the invention patent with publication number CN111579491a, a planar laser-induced breakdown spectroscopy scanner, can rapidly acquire spectral information in a planar sample while performing laser marking operation, and further perform qualitative and quantitative analysis on the surface components of an object through a correction algorithm. However, the problem that the accuracy is low and accurate discrimination cannot be achieved still exists when the LIBS technology is used for classifying and identifying samples at present; the LIBS has large full spectrum data volume, and if the LIBS is processed by full spectrum data, the LIBS has large program operation volume and low operation speed, and can not realize rapid analysis.
With the development of machine learning and artificial intelligence algorithms in recent years, LIBS provides a new method for identifying and classifying substances with high similarity in combination with the machine learning algorithm. Therefore, how to extract the hidden data information in the laser-induced breakdown spectroscopy image to the maximum extent, and to realize the rapid and accurate classification of the unknown sample is a technical problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In view of the above, the invention provides a sample classification method and system based on a laser-induced breakdown spectroscopy technology, which can rapidly and accurately classify unknown samples, improve classification and identification effects, have better stability and reduce calculation amount.
In order to achieve the above object, the present invention provides the following technical solutions:
a sample classification method based on a laser-induced breakdown spectroscopy technology comprises the following steps:
measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method, and collecting original spectrum data;
preprocessing the collected original spectrum data to obtain net spectrum data;
performing dimensionality reduction on the net spectrum data by adopting a principal component analysis method, extracting characteristic variables, and constructing a training data set;
adopting a machine learning algorithm to construct a sample classification model; training the sample classification model by taking the training data set as input and the corresponding class label of the known sample as expected output to obtain an optimal classification model;
and identifying the class label of the sample to be detected by using the optimal classification model.
The technical effect that above-mentioned technical scheme reaches is: the laser-induced breakdown spectroscopy technology is combined with a machine learning algorithm, so that the unknown sample can be rapidly and accurately measured, the calculated amount is greatly reduced, and the recognition accuracy is improved.
Optionally, collecting the original spectrum data specifically includes the following steps:
nd: the YAG laser is used as an excitation source, the spark discharge device is used as an auxiliary excitation source, a known sample is fixed on a sample table capable of three-dimensional displacement, and the position of the known sample is changed after each measurement so that the pulse laser excites different positions of the known sample;
the pulse laser is converged and vertically focused on the surface of a known sample through a focusing lens to generate high-temperature plasma, the plasma spark is excited for the second time by using 1500V direct-current voltage discharge, and the outer flame of the flame acts on the laser focusing point of the known sample to obtain the excited and enhanced plasma;
and coupling the plasma emission spectrum into an optical fiber, and transmitting the plasma emission spectrum to a spectrometer for light splitting and detection to obtain original spectrum data.
The technical effect that above-mentioned technical scheme reaches is: the spark discharge device is used as an auxiliary excitation source to act on the surface of the sample, so that the enhancement of laser-induced breakdown spectrum signals can be realized, and the detection of low-content elements is facilitated.
Optionally, the collected raw spectrum data includes M sets of raw spectrum data obtained by repeatedly measuring M times for each known sample, and each set of raw spectrum data includes N raw spectra obtained by measuring N times at different positions of each known sample; the preprocessing of the collected raw spectral data specifically comprises the following steps:
setting a threshold value according to the peak intensities of a plurality of characteristic spectral lines, and eliminating any one of the spectral data when the peak intensity of the characteristic spectral line is lower than the set threshold value to obtain first spectral data;
performing smoothing treatment, background subtraction treatment and normalization treatment on the first spectrum data to obtain second spectrum data;
respectively extracting the characteristic spectral line intensities of N spectrums in M groups of spectrum data of each known sample, and calculating to obtain the mean value and the variance of the characteristic spectral line intensities of M groups of spectrum data of each known sample;
screening M groups of spectrum data of each known sample by using a significance checking method, removing the group containing the abnormal mean value or the abnormal variance, summing all spectrum data of the rest groups, and averaging to obtain final spectrum data, namely net spectrum data.
The technical effect that above-mentioned technical scheme reaches is: the spectral data after the abnormal values are measured and removed in multiple groups can better reflect the real property of the sample, the spectral data containing correct information is reserved, and the analysis precision of the laser-induced breakdown spectrum is improved.
Optionally, the main component analysis method is adopted to reduce the dimension of the net spectrum data, and the method specifically comprises the following steps:
adopting a principal component analysis method to respectively solve principal component scores of the net spectrum data of different types of known samples;
according to the main component scores of the net spectrum data of the different types of known samples, respectively calculating the spectrum information accumulation contribution rate of the main components of the net spectrum data of the different types of known samples;
and determining the number of the principal components of the net spectrum data for judging the sample category according to the spectrum information accumulation contribution rate of the principal components, and extracting a principal component score matrix.
Optionally, extracting the feature variable further includes the steps of:
adopting a K-fold cross validation algorithm to acquire an initial variable and a preset threshold value of mutual information, and performing mutual information processing on the net spectrum data to acquire a first characteristic variable;
and confirming the particle number, the acceleration coefficient and the maximum iteration number of the particle swarm algorithm by adopting a K-fold cross validation algorithm, and screening the first characteristic variable by adopting the particle swarm algorithm to obtain a second characteristic variable so as to construct a sample classification model.
The technical effect that above-mentioned technical scheme reaches is: redundant variables in the spectrum data can be eliminated, characteristic variables with higher prediction precision can be screened out, the precision of the sample classification model is improved, and the calculation cost is saved.
Optionally, the construction and training of the sample classification model specifically includes the following steps:
a sample classification model is established by adopting a multi-layer perceptron classifier, wherein the multi-layer perceptron classifier comprises an input layer, a hidden layer and an output layer; the hidden layer comprises a full connection layer and a Dropout layer; the number of the neurons contained in the input layer is the same as the dimension of the spectrum data in the training data set, and the number of the neurons contained in the output layer is the same as the total number of the sample categories;
calculating the cross entropy loss of the real label of the known sample in the training data set and the predicted output value of the multi-layer perceptron classifier, and optimizing the multi-layer perceptron classifier by taking the cross entropy loss as an optimization target;
when the classification accuracy of the multi-layer perceptron classifier is superior to that of the network model before optimization, updating model parameters until iteration is finished, and storing the model with the highest classification accuracy, wherein the obtained trained multi-layer perceptron classifier is the optimal classification model.
The technical effect that above-mentioned technical scheme reaches is: the recognition accuracy of the model can be improved, errors are reduced, and the accurate classification of unknown samples is realized.
Optionally, the method further comprises:
after the net spectrum data is obtained, selecting P characteristic spectral lines which have no self-absorption and self-inversion and have highest spectral line intensity from the characteristic spectral lines of the net spectrum data according to the element to be detected;
taking class labels of known samples as output and taking spectral line intensities of P characteristic spectral lines as input, establishing multiple regression models and selecting P optimal models from the multiple regression models;
establishing a comprehensive model by taking the prediction results of the P optimal models as input and the class labels of the known samples as output;
and inputting the P characteristic spectral line intensities of the sample to be tested into the comprehensive model to obtain the element content of the sample to be tested.
The technical effect that above-mentioned technical scheme reaches is: the element content of the sample to be detected can be detected rapidly, the operation is simple, and the cost is low.
The invention also provides a sample classification system based on the laser-induced breakdown spectroscopy technology, which comprises: the device comprises an acquisition module, a preprocessing module, a first construction module, a second construction module, a training module and an identification module;
the acquisition module is used for measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method and acquiring original spectrum data;
the preprocessing module is used for preprocessing the collected original spectrum data to obtain net spectrum data;
the first construction module is used for reducing the dimension of the net spectrum data by adopting a principal component analysis method, extracting characteristic variables and constructing a training data set;
the second construction module is used for constructing a sample classification model by adopting a machine learning algorithm;
the training module is used for training the sample classification model by taking the training data set as input and the class label of the corresponding known sample as output to obtain an optimal classification model;
and the identification module is used for identifying the class label of the sample to be detected by utilizing the optimal classification model.
Optionally, the device for collecting the original spectrum data is a laser-induced breakdown spectroscopy device, including: nd: YAG laser, spark discharge device, sample stage, focusing lens, optical fiber, and spectrometer;
nd: the YAG laser is an excitation source and is used for emitting pulse laser; the pulse laser is converged by a focusing lens and vertically focused on the surface of the sample;
the sample is placed on a sample stage, and the sample stage is positioned on an emergent light path of the focusing lens;
the spark discharge device is an auxiliary excitation source and is used for acting on the surface of the sample simultaneously with the pulse laser;
an optical fiber for transmitting the plasma obtained at the sample focal point to a spectrometer;
and the spectrometer is used for carrying out light splitting and detection on the plasma to obtain original spectrum data.
Compared with the prior art, the invention discloses a sample classification method and system based on a laser-induced breakdown spectroscopy technology, and has the following beneficial effects:
(1) The invention adopts the laser-induced breakdown spectroscopy technology and combines the machine learning algorithm, so that the category and the element content of an unknown sample can be rapidly and accurately measured, the calculated amount is greatly reduced, and the recognition accuracy is improved;
(2) The spark discharge device is used as an auxiliary excitation source and acts on the surface of the sample together with the pulse laser, so that the enhancement of laser-induced breakdown spectrum signals can be realized, and the detection of low-content elements is facilitated;
(3) According to the invention, a plurality of groups of measurement and pretreatment are carried out on the spectrum data of the sample, so that the real property of the sample can be better reflected, the spectrum data containing correct information is reserved, and the analysis precision of the laser-induced breakdown spectrum is improved; and redundant variables in the spectrum data are eliminated, characteristic variables with higher prediction precision are screened out, the precision of the sample classification model is improved, and the calculation cost is saved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a sample classification method based on laser induced breakdown spectroscopy provided by the invention;
fig. 2 is a block diagram of a sample classification system based on a laser-induced breakdown spectroscopy technology provided by the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The laser-induced breakdown spectroscopy technology has the characteristics of simultaneous multi-element analysis, remote measurement, rapidness, no need of complex sample pretreatment, micro loss, high sensitivity and the like, and thus has wide application prospect in the aspect of substance detection. Aiming at the problems that the existing LIBS technology still has low accuracy, large operation amount, low operation speed and incapability of realizing rapid analysis when the LIBS technology is used for sample classification and identification, the embodiment of the invention discloses a sample classification method based on a laser-induced breakdown spectroscopy technology, which is shown in a figure 1 and comprises the following steps:
measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method, and collecting original spectrum data;
preprocessing the collected original spectrum data to obtain net spectrum data;
performing dimensionality reduction on the net spectrum data by adopting a principal component analysis method, extracting characteristic variables, and constructing a training data set;
adopting a machine learning algorithm to construct a sample classification model; training the sample classification model by taking the training data set as input and the corresponding class label of the known sample as expected output to obtain an optimal classification model;
and identifying the class label of the sample to be detected by using the optimal classification model.
Based on the collected laser-induced breakdown spectroscopy data, a machine learning method is adopted to establish a classification model, so that the sample classification performance and the model classification accuracy can be improved, the calculation efficiency is improved, and the unknown sample can be classified rapidly and accurately.
Further, the method for acquiring the original spectrum data specifically comprises the following steps:
nd: the YAG laser is used as an excitation source, the spark discharge device is used as an auxiliary excitation source, a known sample is fixed on a sample table capable of three-dimensional displacement, and the position of the known sample is changed after each measurement so that the pulse laser excites different positions of the known sample;
the pulse laser is converged and vertically focused on the surface of a known sample through a focusing lens to generate high-temperature plasma, the plasma spark is excited for the second time by using 1500V direct-current voltage discharge, and the outer flame of the flame acts on the laser focusing point of the known sample to obtain the excited and enhanced plasma;
and coupling the plasma emission spectrum into an optical fiber, and transmitting the plasma emission spectrum to a spectrometer for light splitting and detection to obtain original spectrum data.
The sensitivity of the laser-induced breakdown spectroscopy technology directly determines the analyzed elements, the current method for enhancing the sensitivity of the LIBS technology has high complexity and high cost, and the technology takes a spark discharge device as an auxiliary excitation source, can enhance the laser-induced breakdown spectroscopy signal without adding an additional device, and can obtain obvious spectroscopy signals with low content elements in a sample.
Further, the collected raw spectrum data comprises M groups of raw spectrum data obtained by repeatedly measuring M times for each known sample, and each group of raw spectrum data comprises N raw spectra obtained by measuring N times at different positions of each known sample, wherein M is more than 0 and N is more than 0; the preprocessing of the collected raw spectral data specifically comprises the following steps:
setting a threshold value according to the peak intensities of a plurality of characteristic spectral lines, and eliminating any one of the spectral data when the peak intensity of the characteristic spectral line is lower than the set threshold value to obtain first spectral data;
performing smoothing treatment, background subtraction treatment and normalization treatment on the first spectrum data to obtain second spectrum data;
respectively extracting the characteristic spectral line intensities of N spectrums in M groups of spectrum data of each known sample, and calculating to obtain the mean value and the variance of the characteristic spectral line intensities of M groups of spectrum data of each known sample;
screening M groups of spectrum data of each known sample by using a significance checking method, removing the group containing the abnormal mean value or the abnormal variance, summing all spectrum data of the rest groups, and averaging to obtain final spectrum data, namely net spectrum data.
After abnormal data are removed, a spectrum with high spectral line intensity and clear spectral line can be obtained, but under the condition of extremely large relative standard deviation, the significance of single measurement is not great, so that the technology performs multiple measurements, removes the group with abnormal fluctuation of parameters, can reserve the spectrum data containing correct information, better reflects the real property of a sample, and improves the analysis precision of the laser-induced breakdown spectrum.
Further, the main component analysis method is adopted to reduce the dimension of the net spectrum data, and the method specifically comprises the following steps:
adopting a principal component analysis method to respectively solve principal component scores of the net spectrum data of different types of known samples;
according to the main component scores of the net spectrum data of the different types of known samples, respectively calculating the spectrum information accumulation contribution rate of the main components of the net spectrum data of the different types of known samples;
and determining the number of the principal components of the net spectrum data for judging the sample category according to the spectrum information accumulation contribution rate of the principal components, and extracting a principal component score matrix.
The principal component analysis method is adopted to analyze the spectrum data, so that effective information and environmental noise can be effectively distinguished, data reduction and feature extraction of the net spectrum data are realized, and interference of human subjective factors is avoided.
Further, extracting the characteristic variable, further comprising the steps of:
adopting a K-fold cross validation algorithm to acquire an initial variable and a preset threshold value of mutual information, and performing mutual information processing on the net spectrum data to acquire a first characteristic variable;
and confirming the particle number, the acceleration coefficient and the maximum iteration number of the particle swarm algorithm by adopting a K-fold cross validation algorithm, and screening the first characteristic variable by adopting the particle swarm algorithm to obtain a second characteristic variable so as to construct a sample classification model.
Because of the complexity of the sample, each laser-induced breakdown spectroscopy data comprises a lot of noise and redundant information besides the sample information, and the mutual information processing of the spectroscopy data by the technology can eliminate redundant variables in the spectroscopy data, and then the particle swarm algorithm is used for further screening, so that the accurate screening of characteristic variables can be realized, and the prediction accuracy of a classification model is improved.
Further, the construction and training of the sample classification model specifically comprises the following steps:
a sample classification model is established by adopting a multi-layer perceptron classifier, wherein the multi-layer perceptron classifier comprises an input layer, a hidden layer and an output layer; the hidden layer comprises a full connection layer and a Dropout layer; the number of the neurons contained in the input layer is the same as the dimension of the spectrum data in the training data set, and the number of the neurons contained in the output layer is the same as the total number of the sample categories;
calculating the cross entropy loss of the real label of the known sample in the training data set and the predicted output value of the multi-layer perceptron classifier, and optimizing the multi-layer perceptron classifier by taking the cross entropy loss as an optimization target;
when the classification accuracy of the multi-layer perceptron classifier is superior to that of the network model before optimization, updating model parameters until iteration is finished, and storing the model with the highest classification accuracy, wherein the obtained trained multi-layer perceptron classifier is the optimal classification model.
According to the technical scheme, the sample classification model is introduced into the multi-layer perceptron classifier, the class labels of the samples to be detected can be rapidly and accurately determined through the optimal classification model obtained through training, the recognition accuracy of the model is improved, and errors are reduced.
Further, the method further comprises:
after the net spectrum data is obtained, selecting P characteristic spectral lines which have no self-absorption and self-inversion and have highest spectral line intensity from the characteristic spectral lines of the net spectrum data according to the element to be detected;
taking class labels of known samples as output and taking spectral line intensities of P characteristic spectral lines as input, establishing multiple regression models (such as partial least square method, multiple linear regression, logistic regression, extreme learning machine and the like) and selecting P optimal models from the multiple regression models;
establishing a comprehensive model by taking the prediction results of the P optimal models as input and the class labels of the known samples as output;
and inputting the P characteristic spectral line intensities of the sample to be tested into the comprehensive model to obtain the element content of the sample to be tested.
The technical scheme can realize the rapid detection of the element content of the sample to be detected based on the laser-induced breakdown spectroscopy, and has the characteristics of low cost, simple method and rapid detection.
Corresponding to the method shown in fig. 1, the embodiment of the present invention further provides a sample classification system based on a laser-induced breakdown spectroscopy technology, which is used for implementing the method shown in fig. 1, and the sample classification system based on the laser-induced breakdown spectroscopy technology provided in the embodiment of the present invention may be applied to a computer terminal or various mobile devices, and its structural schematic diagram is shown in fig. 2, and specifically includes: the device comprises an acquisition module, a preprocessing module, a first construction module, a second construction module, a training module and an identification module;
the acquisition module is used for measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method and acquiring original spectrum data;
the preprocessing module is used for preprocessing the collected original spectrum data to obtain net spectrum data;
the first construction module is used for reducing the dimension of the net spectrum data by adopting a principal component analysis method, extracting characteristic variables and constructing a training data set;
the second construction module is used for constructing a sample classification model by adopting a machine learning algorithm;
the training module is used for training the sample classification model by taking the training data set as input and the class label of the corresponding known sample as output to obtain an optimal classification model;
and the identification module is used for identifying the class label of the sample to be detected by utilizing the optimal classification model.
Further, the device for collecting the original spectrum data is a laser-induced breakdown spectroscopy device, which comprises: nd: YAG laser, spark discharge device, sample stage, focusing lens, optical fiber, and spectrometer;
nd: the YAG laser is an excitation source and is used for emitting pulse laser; the pulse laser is converged by a focusing lens and vertically focused on the surface of the sample;
the sample is placed on a sample stage, and the sample stage is positioned on an emergent light path of the focusing lens;
the spark discharge device is an auxiliary excitation source and is used for acting on the surface of the sample simultaneously with the pulse laser;
an optical fiber for transmitting the plasma obtained at the sample focal point to a spectrometer;
and the spectrometer is used for carrying out light splitting and detection on the plasma to obtain original spectrum data.
The invention adopts the laser-induced breakdown spectroscopy technology and combines the machine learning algorithm, so that the category and the element content of an unknown sample can be rapidly and accurately measured, the calculated amount is greatly reduced, and the recognition accuracy is improved; the spark discharge device is used as an auxiliary excitation source and pulse laser is acted on the surface of the sample at the same time, so that the enhancement of laser-induced breakdown spectrum signals can be realized, and the detection of low-content elements is facilitated; redundant variables in the spectrum data are eliminated, characteristic variables with higher prediction precision can be screened out, and the calculation cost is saved.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (9)
1. The sample classification method based on the laser-induced breakdown spectroscopy technology is characterized by comprising the following steps of:
measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method, and collecting original spectrum data;
preprocessing the collected original spectrum data to obtain net spectrum data;
performing dimensionality reduction on the net spectrum data by adopting a principal component analysis method, extracting characteristic variables, and constructing a training data set;
adopting a machine learning algorithm to construct a sample classification model; training the sample classification model by taking the training data set as input and the corresponding class label of the known sample as expected output to obtain an optimal classification model;
and identifying the class label of the sample to be detected by using the optimal classification model.
2. The method for classifying samples based on the Laser Induced Breakdown Spectroscopy (LIBS) according to claim 1, wherein the step of collecting raw spectral data comprises the steps of:
nd: the YAG laser is used as an excitation source, the spark discharge device is used as an auxiliary excitation source, a known sample is fixed on a sample table capable of three-dimensional displacement, and the position of the known sample is changed after each measurement so that the pulse laser excites different positions of the known sample;
the pulse laser is converged and vertically focused on the surface of a known sample through a focusing lens to generate plasma, the plasma spark is excited for the second time by using 1500V direct-current voltage discharge, and the outer flame of the flame acts on the laser focusing point of the known sample to obtain the excited and enhanced plasma;
and coupling the plasma emission spectrum into an optical fiber, and transmitting the plasma emission spectrum to a spectrometer for light splitting and detection to obtain original spectrum data.
3. The method for classifying samples based on the laser induced breakdown spectroscopy according to claim 1, wherein the collected raw spectral data comprises M sets of raw spectral data obtained by repeatedly measuring M times for each known sample, each set of raw spectral data comprising N raw spectra obtained by measuring N times for each known sample at different positions; the preprocessing of the collected raw spectral data specifically comprises the following steps:
setting a threshold value according to the peak intensities of a plurality of characteristic spectral lines, and eliminating any one of the spectral data when the peak intensity of the characteristic spectral line is lower than the set threshold value to obtain first spectral data;
performing smoothing treatment, background subtraction treatment and normalization treatment on the first spectrum data to obtain second spectrum data;
respectively extracting the characteristic spectral line intensities of N spectrums in M groups of spectrum data of each known sample, and calculating to obtain the mean value and the variance of the characteristic spectral line intensities of M groups of spectrum data of each known sample;
screening M groups of spectrum data of each known sample by using a significance checking method, removing the group containing the abnormal mean value or the abnormal variance, summing all spectrum data of the rest groups, and averaging to obtain final spectrum data, namely net spectrum data.
4. The method for classifying samples based on the laser-induced breakdown spectroscopy according to claim 1, wherein the method for dimensionality reduction of the net spectral data by principal component analysis comprises the steps of:
adopting a principal component analysis method to respectively solve principal component scores of the net spectrum data of different types of known samples;
according to the main component scores of the net spectrum data of the different types of known samples, respectively calculating the spectrum information accumulation contribution rate of the main components of the net spectrum data of the different types of known samples;
and determining the number of the principal components of the net spectrum data for judging the sample category according to the spectrum information accumulation contribution rate of the principal components, and extracting a principal component score matrix.
5. The method for classifying samples based on the laser induced breakdown spectroscopy according to claim 1, wherein the feature variables are extracted, further comprising the steps of:
adopting a K-fold cross validation algorithm to acquire an initial variable and a preset threshold value of mutual information, and performing mutual information processing on the net spectrum data to acquire a first characteristic variable;
and confirming the particle number, the acceleration coefficient and the maximum iteration number of the particle swarm algorithm by adopting a K-fold cross validation algorithm, and screening the first characteristic variable by adopting the particle swarm algorithm to obtain a second characteristic variable so as to construct a sample classification model.
6. The method for classifying samples based on the laser-induced breakdown spectroscopy according to claim 1, wherein the construction and training of the sample classification model specifically comprises the following steps:
a sample classification model is established by adopting a multi-layer perceptron classifier, wherein the multi-layer perceptron classifier comprises an input layer, a hidden layer and an output layer; the hidden layer comprises a full connection layer and a Dropout layer; the number of the neurons contained in the input layer is the same as the dimension of the spectrum data in the training data set, and the number of the neurons contained in the output layer is the same as the total number of the sample categories;
calculating the cross entropy loss of the real label of the known sample in the training data set and the predicted output value of the multi-layer perceptron classifier, and optimizing the multi-layer perceptron classifier by taking the cross entropy loss as an optimization target;
when the classification accuracy of the multi-layer perceptron classifier is superior to that of the network model before optimization, updating model parameters until iteration is finished, and storing the model with the highest classification accuracy, wherein the obtained trained multi-layer perceptron classifier is the optimal classification model.
7. The method of claim 1, further comprising:
after the net spectrum data is obtained, selecting P characteristic spectral lines which have no self-absorption and self-inversion and have highest spectral line intensity from the characteristic spectral lines of the net spectrum data according to the element to be detected;
taking class labels of known samples as output and taking spectral line intensities of P characteristic spectral lines as input, establishing multiple regression models and selecting P optimal models from the multiple regression models;
establishing a comprehensive model by taking the prediction results of the P optimal models as input and the class labels of the known samples as output;
and inputting the P characteristic spectral line intensities of the sample to be tested into the comprehensive model to obtain the element content of the sample to be tested.
8. A sample classification system based on laser-induced breakdown spectroscopy, comprising: the device comprises an acquisition module, a preprocessing module, a first construction module, a second construction module, a training module and an identification module;
the acquisition module is used for measuring different types of known samples by adopting a laser-induced breakdown spectroscopy method and acquiring original spectrum data;
the preprocessing module is used for preprocessing the collected original spectrum data to obtain net spectrum data;
the first construction module is used for reducing the dimension of the net spectrum data by adopting a principal component analysis method, extracting characteristic variables and constructing a training data set;
the second construction module is used for constructing a sample classification model by adopting a machine learning algorithm;
the training module is used for training the sample classification model by taking the training data set as input and the class label of the corresponding known sample as output to obtain an optimal classification model;
and the identification module is used for identifying the class label of the sample to be detected by utilizing the optimal classification model.
9. The system for classifying samples based on the laser induced breakdown spectroscopy of claim 8, wherein the means for collecting raw spectral data is a laser induced breakdown spectroscopy device comprising: nd: YAG laser, spark discharge device, sample stage, focusing lens, optical fiber, and spectrometer;
nd: the YAG laser is an excitation source and is used for emitting pulse laser; the pulse laser is converged by a focusing lens and vertically focused on the surface of the sample;
the sample is placed on a sample stage, and the sample stage is positioned on an emergent light path of the focusing lens;
the spark discharge device is an auxiliary excitation source and is used for acting on the surface of the sample simultaneously with the pulse laser;
an optical fiber for transmitting the plasma obtained at the sample focal point to a spectrometer;
and the spectrometer is used for carrying out light splitting and detection on the plasma to obtain original spectrum data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310225705.2A CN116030310A (en) | 2023-03-09 | 2023-03-09 | Sample classification method and system based on laser-induced breakdown spectroscopy technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310225705.2A CN116030310A (en) | 2023-03-09 | 2023-03-09 | Sample classification method and system based on laser-induced breakdown spectroscopy technology |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116030310A true CN116030310A (en) | 2023-04-28 |
Family
ID=86072580
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310225705.2A Pending CN116030310A (en) | 2023-03-09 | 2023-03-09 | Sample classification method and system based on laser-induced breakdown spectroscopy technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116030310A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116559092A (en) * | 2023-07-04 | 2023-08-08 | 北京理工大学 | Tumor microscopic gene mutation detection system based on macroscopic spectrum element component analysis |
| CN116878407A (en) * | 2023-09-08 | 2023-10-13 | 法博思(宁波)半导体设备有限公司 | Epitaxial wafer thickness measuring method and device based on infrared interference |
-
2023
- 2023-03-09 CN CN202310225705.2A patent/CN116030310A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116559092A (en) * | 2023-07-04 | 2023-08-08 | 北京理工大学 | Tumor microscopic gene mutation detection system based on macroscopic spectrum element component analysis |
| CN116559092B (en) * | 2023-07-04 | 2023-09-26 | 北京理工大学 | Tumor microscopic gene mutation detection system based on macroscopic spectrum element component analysis |
| CN116878407A (en) * | 2023-09-08 | 2023-10-13 | 法博思(宁波)半导体设备有限公司 | Epitaxial wafer thickness measuring method and device based on infrared interference |
| CN116878407B (en) * | 2023-09-08 | 2023-12-01 | 法博思(宁波)半导体设备有限公司 | Epitaxial wafer thickness measuring method and device based on infrared interference |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116030310A (en) | Sample classification method and system based on laser-induced breakdown spectroscopy technology | |
| CN101995395B (en) | Method for online detecting material by laser induction multiple spectrum united fingerprint network | |
| CN113155809B (en) | Novel spectral detection method for ore classification and real-time quantitative analysis | |
| CN112051256B (en) | CNN model-based LIBS (laser induced breakdown spectroscopy) measurement method and system for content of element to be measured | |
| CN110717368A (en) | Qualitative classification method for textiles | |
| CN104730041B (en) | Method and apparatus for improving plastic identification precision of laser probe | |
| CN104483292B (en) | A kind of method that use multiline ratio method improves laser microprobe analysis accuracy | |
| CN107220625B (en) | LIBS spectrum automatic identification and classification method for sample | |
| US10718713B2 (en) | Unknown sample determining method, unknown sample determining instrument, and unknown sample determining program | |
| CN110763660A (en) | LIBS quantitative analysis method based on ensemble learning | |
| CN105300954A (en) | Method for Raman spectrum characterization of heterogeneous foam layer minerals in antimony ore flotation | |
| CN113095188A (en) | Deep learning-based Raman spectrum data analysis method and device | |
| CN111832477A (en) | Novel coronavirus detection method and system | |
| CN112505010A (en) | Transformer fault diagnosis device and method based on fluorescence spectrum | |
| CN109142251B (en) | LIBS quantitative analysis method of random forest auxiliary artificial neural network | |
| Ruan et al. | A modified backward elimination approach for the rapid classification of Chinese ceramics using laser-induced breakdown spectroscopy and chemometrics | |
| CN112834481B (en) | Raman spectrum enhancement measurement system and measurement method | |
| CN111272735B (en) | Detection method of laser-induced breakdown spectroscopy | |
| CN116858822A (en) | Quantitative analysis method for sulfadiazine in water based on machine learning and Raman spectrum | |
| CN116026808A (en) | Raman spectrum discrimination method and system | |
| CN113791062A (en) | Method for judging fixed substance type based on Raman spectrum | |
| CN112378892B (en) | Cosmetic authenticity nondestructive identification method based on Raman spectrum | |
| CN112816458B (en) | Steel classification method and system based on laser-induced breakdown spectroscopy technology | |
| CN111044504B (en) | Coal quality analysis method considering uncertainty of laser-induced breakdown spectroscopy | |
| CN114240896A (en) | Method for detecting welding quality of stud welding based on convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |