Abstract: The SuperCam carried by the NASA's Perseverance rover can detect the surface material composition of Mars such as Mn. In order to determine the content of Mn on Mars, a quantitative method for Mn based on ensemble learning is proposed using the laser-induced breakdown spectroscopy (LIBS) dataset of geologic standards. A series of pre-processing such as spectral denosing and de-baselining are carried out firstly, then spectral deconvolution is performed to realize peak-fitting, and finally a quantitative method for Mn content prediction is established. The quantitative accuracy for Mn of the different quantitative methods were experimentally compared. The results show that, compared with the two traditional methods (LASSO and ElasticNet), the root-mean-square error of the proposed method based on ensemble learning is reduced by 49% and 30% on average, respectively, and the quantitative results of the new method are closer to the real values of the samples. This study shows that the ensemble learning based quantitative method is more suitable for Mars Mn quantification.

Key words: spectroscopy, spectral quantification methods, laser-induced breakdown spectroscopy, ensemble learning, Mars, Mn element