Abstract. Fast and effective measures to determine heavy metals play an important role in ensuring food quality and safety. In this experiment, laser-induced breakdown spectroscopy (LIBS) was used to detect copper content (Cu) in tobacco ( L.) leaves. The experimental parameters for detection, including laser energy, delay time, and camera gate width, were optimized by response surface methodology (RSM). Univariate analysis and multivariate analysis, including partial least squares regression (PLSR) and extreme learning machine (ELM), were used to establish calibration models. In addition, different preprocessing methods were used to eliminate the signal variations and further improve the calibration performance, including baseline correction, background normalization, area normalization, and standard normal variate (SNV) normalization. The results showed that LIBS combined with both univariate and multivariate methods could be used to detect copper content in tobacco leaves. SNV and area normalization were efficient in dealing with signal variations and improving the calibration performance. The ELM model with SNV normalized variables in the spectral region of 324.02 to 325.98 nm achieved the best performance (R2 = 0.9552 and RMSE = 4.8416 mg kg-1 in the testing set). The results provide the first proof-of-principle data for fast determination of copper content in tobacco leaves. Keywords: Copper content, Laser-induced breakdown spectroscopy, Multivariate calibration, Response surface methodology, Tobacco leaves, Univariate calibration.
