Abstract
Abstract A laser-induced breakdown spectroscopy (LIBS) technique coupled with random forest (RF) was developed and used to classify ceramics from different dynasties. LIBS spectra of the ceramic surfaces were collected, and the major elements (Fe, Ca, Si, Al and Mg) in the ceramics identified using the NIST database. To obtain a better classification results, the LIBS spectra were subjected to five different pre-processing techniques (normalized by maximum integrated intensity, by extremum integrated intensity, mean centering, first-order derivative and second-order derivative). The input variables for RF modeling were selected and optimized by different variable importance threshold values (from 0 to 0.20) and four assessment criteria of out-of-bag (OOB) error, sensitivity, specificity and accuracy. LIBS spectra pre-processed by mean centering with a variable importance threshold value of 0.02 as the input variable were used to construct a RF classification model for different dynasty ceramics. Finally, the classification performance of the RF model was verified by the train set, OOB data and test set. The sensitivity, specificity and accuracy obtained by the RF model for the ceramics samples of test set were 0.8528, 0.9710 and 0.9433, respectively, which indicated a good classification performance.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call