| In rock logging engineering,rock samples are used to determine rock formation lithology,fluid properties and rock position,and then to determine oil and gas field storage;During the evolution of the earth’s environment,rocks and minerals will variation or change to adapt to the changes of the earth’s environment;In addition,reinforcing the research on the important factors affecting geological disasters such as geological structure and lithological conditions on the earth is conducive to improving the ability to predict geological disasters.Therefore,rock lithology identification plays an irreplaceable role in many aspects,such as exploration and development of oil and gas fields,research on the genesis and evolution of the earth,analysis and prediction of geological disasters.And the identification and classification of rocks is very important for geological exploration and analysis.At present,the traditional rock classification method in geological work is mainly physical experiment observation method,which requires high professional knowledge and experience of the staff,and has strong subjectivity and low repeatability.Other chemical analysis methods are not suitable for field analysis because of their complex preparation process,long analysis period,high detection limit,and unsatisfied real time and accuracy,such as X-ray fluorescence spectrometry(XRF),X-ray diffraction analysis(XRD).In recent years,the rapid development of machine learning algorithm for rock and mineral analysis method has high accuracy,weak requirements for the geological professional level of staff,and makes the whole identification process more intelligent and automatic.The method of using laser-induced breakdown spectroscopy(LIBS)to collect the spectral data of rock and using machine learning algorithm to analyze the rock minerals has the advantages of LIBS technology,such as simple sample pretreatment,small destructiveness,short analysis time,real-time online and remote monitoring,and the advantages of using machine learning algorithm to analyze the rock minerals.It has great potential in the application of rock classification.In this paper,based on LIBS rock classification as the application background,the rock classification experiment system is designed and built independently,and the spectral data collection of different rock samples is completed.The influence of experimental parameters such as laser pulse energy and acquisition delay time on spectrum data acquisition is analyzed and the system parameters are set.Finally,the interval setting of laser bombardment position in the experimental scheme is explained.For the experiment of rock two classification,firstly,the characteristics of six rock samples are analyzed,and the spectral intensity and normalization of the spectral data are preprocessed.Based on the difference of mineral composition among the rock samples,the characteristic spectral lines for rock classification are determined and the element fingerprint is generated.The sample data set is formed by overlapping and cutting the element fingerprint and getting the average value of spectral lines.The classification model of spectral mean is established in rock classification.The shortcomings of the classification method using the average spectrum are analyzed by combining the average spectrum of the two types of rocks in the classification result set.In order to overcome the above shortcomings,a method of rock two classification based on the fingerprint of rock surface is proposed,and the classification method of average spectrum is used for comparative analysis.The results show that the accuracy of rock classification by fingerprint is significantly improved compared with that by average spectrum.The reason is that the fingerprint in the second classification model keeps the mineral spatial structure characteristics of the rock while using the content information of mineral elements in the rock,so the classification effect is better.Based on the two classification of rock,the four classification of rock is studied.Firstly,the SVM classification method of multi-dimensional fingerprint fusion proposed in the two classification of rock is adopted,and the classification effect is not ideal through the analysis of the experimental results.In order to improve the accuracy of rock classification,a four classification method is proposed,which combines SVM classification model of multi-dimensional fingerprint and KNN classification model of average spectrum.This method not only makes use of the mineral spatial structure characteristics of the rock retained in the rock fingerprint,but also makes use of the great difference of the spectral lines of conglomerate and shale in the element types and element contents.Therefore,the accuracy of rock classification is generally higher than that of SVM four classification method based on multi-dimensional fingerprint.Based on the study of two classification and four classification of different rock samples,two methods of rock classification according to different rock characteristics are proposed.These two methods improve the classification accuracy of the corresponding rock classification.It has great potential to use LIBS technology to collect spectral data of rocks and machine learning algorithm to classify the collected spectral data. |
