Objective: To establish back-propagation artificial neural networks quantitative models by Raman spectroscopy technique

the model can get a rapid determination for the content of CaF2 in Fluoritum. Method: Raman spectra of 54 batches of samples including Fluoritum and the mixture which had different content of CaF2 were collected

and were preprocessed with Fourier self-deconvolution (FSD)

first-order derivative (FD)

vector normalization (VN) and 9 point smoothing by OPUS 7.0 software.Modeling spectrum interval was optimized by siPLS

the quantitative models were established by BP-ANN algorithm in MATLAB 2014a software.The content of CaF2 was determined by EDTA titration method

which was used as a reference. Result: Three-layer BP-ANN quantitative models were established

the characteristic modeling spectrum intervals were 1 675-1 625

1 525-1 475

850-800

750-700

650-600 cm-1

RMSEP was 2.73%with R2 of 85.64%.The maximum relative deviation of prediction result was 5.55%

the average relative deviation was 2.30% and the average recovery was 99.74%. Conclusion: The three-layer BP-ANN quantitative model established on the basis of siPLS can get a rapid and accurate determination for the content of CaF2 in Fluoritum.