Abstract: A quantitative analysis model for detecting Cr in soil with laser induced breakdown spectroscopy (LIBS) was established by using correlation vector machine (RVM) combined with principal component analysis (PCA). Fourteen soil samples with different Cr concentrations were prepared, of which ten were selected as training samples for model construction, and the other four as test samples for model performance evaluation. The results show that the prediction accuracy of PCA-RVM model is significantly better than that of RVM model for the measurement of Cr content in soil. The root mean square error (RMSE) of the whole prediction is reduced from 8.00% of RVM model to 3.21% of PCARVM model, and the prediction accuracy is improved by 59.9%. Compared with RVM model, the relative standard deviation of repeated prediction results of PCA-RVM model for all four samples in the test sample set is significantly reduced and is less than 1.89%, indicating that the prediction results of PCARVM model have better stability.

Key words: spectroscopy, laser-induced breakdown spectroscopy, principal component analysis, relevance vector machine, soil