Heavy metal residue in vegetables is a big concern in the whole world.The aim of this work is to explore the effect of multivariable selection on analyzing Cd in Chinese cabbage polluted in lab by collecting the spectra of laser induced breakdown spectroscopy (LIBS) from the samples.At the same time,the actual Cd content in samples was obtained by anodic stripping voltammetry (ASV).The LIBS spectral range in partial least square (PLS) model was screened by standard normal variable transformation (SNV),first derivative (FD),second derivative (SD) and center treatment (CT) for preprocessing spectra and the optimized method was used for the analysis of interval partial least square (iPLS) and synergy interval partial least square (SiPLS).The results indicated that the method of CT was the best as a comparison with PLS,iPLS and SiPLS.And the intervals of wavelength were 214.72-215.82 nm,215.88-216.97 nm and 225.08 -226.35 nm by utilizing the optimized SiPLS.Here the root mean square error of cross validation (RMSECV) between real content and predicted ones was 1.487,the root mean squared error of prediction (RMSEP) was 1.094,the correlation coefficient (R) was 0.9942,and the average relative error (ARE) was 11.60％.The results displayed that LIBS could predict Cd in vegetables by multivariable selection of SiPLS and the accuracy could meet the requirement of rapid and green analysis of Cd in vegetables.

There are many noise signals in original laser induced breakdown spectroscopy (LIBS) spectra.To explore the effect of spectral pretreatment on LIBS information by different filter methods, the LIBS spectra of Pb-polluted cabbage in wavelength range of 400.45-410.98 nm was investigated and preprocessed by adjacent averaging, Savitzky-Golay (S-G) and fast Fourier transformation (FFT).Then partial least square (PLS) model was established for evaluating the spectral treatment effect.The result showed that the root mean square error of prediction (RMSEP) and average relative error of S-G method were 0.26 and 3.7%, suggesting a superior smoothing effect than other methods.Experimental results indicated that an appropriate filtering method could help to improve the spectral quality and raise the precision of model checkout.