| Mineral classification is an important part of geological survey and many engineering applications.Traditional mineral analysis methods include physical and chemical methods,in which chemical methods rely too much on a single chemical property,the composition of more complex minerals is not universal,the physical method is artificial mineralization,there is a subjective deviation,and the workload of the identification cycle is long.This study through the raman spectra of minerals and then qualitative analysis of minerals,Raman spectral advantages are: Raman spectral peaks clear sharp,identifiable,more suitable for quantitative research,as well as further differentiation analysis for qualitative research,can also build a spectral database for retrieval;Combined with computer technology,the characteristics of the spectral spectrum can be excavated more effectively,resulting in faster and more objective results,and more importantly,the damage to the mineral itself caused by the collection of the mineral Raman spectrum is negligible.This study used web crawler technology to collect high-quality Raman spectral data on more than 200 minerals in open databases such as RUFF and the University of Palma,with a total of about 3,700 samples.Collected Raman spectrum there are many kinds,high characteristic dimensions,data acquisition standards are not uniform,noise,fluorescence background,and so on,in view of these problems,this study uses multiple pre-treatment methods to deal with one by one,including but not limited to smoothing,interpolation,remove baseline,etc.,and then normalization,PCA decomposition,for classification models,this study attempted a variety of machine learning classification algorithms(e.g.SVM,KNN,logistics,etc.),Finally,the best similarity learning method based on the Siamese neural network of comprehensive performance(evaluation criteria and model efficiency)was chosen.Its accuracy rate is 92.07 percent,which is significantly higher than that of traditional machine learning algorithms,and the Hungarian algorithm was used to optimize the negative sample during the training phase.The Siamese neural network calculates the similarity between minerals,which,in addition to identifying minerals,can also provide some reference for alternative materials for the mineral. |
