Research On Spectra Data Processing Methods In Laser Probe Technology

Since laser probe technology was invented, most researchers have mainly focused on the physical characteristics of laser-induced plasma, the physical and chemical properties of experimental samples, experimental parameters optimization, and instrument performance, etc. However, the research on spectra data processing has not been paid enough attention. As a powerful soft optimization method, spectra data processing is very effective on many aspects on improving the quality of spectra analysis. On one hand, data processing can replace some high-precision hardware equipments to meet the requirements, which can reduce the hardware cost. On the other hand, data processing can solve the technical problems while hardware can not overcome. Furthermore, data processing can significantly improve the accuracy of spectra analysis by extracting useful spectra information and reprocessing spectra data. The spectra data preprocessing and quantitative analysis methods had been researched in this dissertation to improve the accuracy of qualitative and quantitative analysis.Firstly, a peak detection algorithm based on continuous wavelet transform for laser probe was studied. By putting forward a new method to automatically calculate the noise, continuous wavelet transform combined with the signal-to-noise ratio threshold method was adopted to identify the spectral peaks of soil samples. The results showed that the approach can effectively eliminate interference from spike noise, identify strong peaks and overlapping peaks, and thus improved the accuracy of qualitative analysis. The peak detection method laid a solid foundation for the subsequent quantitative analysis.Secondly, a modified algorithm of background removal to correct the traditional background subtraction algorithm based on discrete wavelet transform was proposed and applied on the background correction of the four elements Cr, V, Cu and Mn in low alloy steel samples. The results showed that the background of the spectra reduced after applying the algorithm and avoided overestimation compared with the traditional method. In comparison with the original data, the methods of polynomial fitting and conventional wavelet transform, the proposed algorithm can effectively improve the analysis accuracy.Thirdly, a genetic algorithm and partial least squares hybrid model for quantitative analysis was proposed and applied to predict the concentrations of eleven soil compositions Mn, Cr, Cu, Pb, Ba, Al2O3, CaO, Fe2O3, MgO, Na2 O, and K2 O, respectively. The results demonstrated that, as a pretreatment method for optimizing the selection of spectral lines, genetic algorithm can effectively remove repetitive, redundant or irrelevant variables in spectra, and reduce the number of spectral lines for use in building partial least squares model, and hence reduced the modeling time and simplified the quantitative analysis model. More importantly, for most of the soil compositions, the hybrid model can significantly improve the prediction ability compared with the conventional partial least squares model.Finally, a partial least squares and artificial neural network hybrid model was presented and applied on the quantitative analysis of eleven soil compositions. The results indicated that the hybrid model can combine both the advantage of decreasing multicollinearity of independent variables of partial least squares and the ability of processing non-linear problem of artificial neural network, and thus improved the analysis accuracy of laser probe.