LIBS结合最大相关最小冗余特征选择的镁合金快速分类识别OA

Magnesium alloys,owing to their low density,high specific strength,and excellent corrosion resistance,are widely applied in aerospace,automotive,and electronics industries.Laser-induced break-down spectroscopy(LIBS)offers advantages such as rapid analysis and minimal sample preparation,mak-ing it highly promising for magnesium alloy detection.However,LIBS spectra exhibit significant fluctua-tions across repeated measurements,while the spectral similarity among different types of magnesium al-loys is high.In addition,the data contain redundant information,which limits the performance of direct classification.In this work,a rapid classification method for magnesium alloys based on feature selection was proposed.Three feature selection strategies maximum relevance minimum redundancy(mRMR),random forest(RF),and spectral indices were systematically compared and combined with three classifi-ers,including support vector machine(SVM),back propagation neural network(BPNN),and k-nearest neighbor(KNN),to construct multiple classification models for magnesium alloys.Experimental results demonstrate that the mRMR-BPNN model achieved accuracies of 99.4%and 92.5%on the first-day and second-day test datasets,respectively,using only 180 selected features.This performance significantly outperforms other feature selection classifier combinations as well as direct classification using raw spectra.The proposed method effectively improves both classification accuracy and generalization capability with-out requiring complex preprocessing,providing a reliable approach for rapid online detection and quality control of magnesium-aluminum alloys.This work highlights the practical potential of LIBS technology for industrial on-site applications.