Improving laser-induced breakdown spectroscopy regression modelsviatransfer learning

Abstract

Laser-induced breakdown spectroscopy (LIBS) is a well-established analytical tool with relevance in extra-terrestrial exploration. Despite considerable efforts towards the development of calibration-free LIBS approaches, these are currently outperformed by calibration-based...

Spectral library transfer between distinct laser-induced breakdown spectroscopy systems trained on simultaneous measurements

“…19 CT was extensively studied in various spectroscopic branches (mostly in NIR and IR 18 ) but has emerged in LIBS only relatively recently (see ref. 19 and 20). While the data complexity varies between distinct spectroscopic techniques ( e.g.…”

“…Last to mention is our previous work, 25 where we used an MLP to transfer spectra from the ChemCam to the SuperCam to improve calibration models for studied oxides. We were able to improve the RMSE over models that were trained solely on the SuperCam system.…”

Transfer learning based on dynamic time warping algorithms to improve qualitative analysis and quantitative prediction of rocks over multiple LIBS instruments

“…The Chemcam team 15 is also disturbed by this problem, the difference being that their Mars rover will generate a large amount of data during its active running, allowing them to enlarge the calibration dataset and recalibrate the model. Kepes 31 used a transfer learning method to improve the problem of model sharing between the Chemcam and Supercam instruments and the artificial neural networks (ANN) model was trained to predict the spectra of Supercam as the shape of the spectra of Chemcam, which was effective to improve the performance of the convolutional neural networks (CNN) regression model. Shabbir 32 introduced two important features based on transfer learning in the experiments, one of which was to combine the features of smooth and rough samples and then generate a group of common features with no significant differences to train the regression model.…”

Accuracy improvement of laser-induced breakdown spectroscopy coal analysis by hybrid transfer learning

“…A neural network was trained to find inter-domain transfer functions for spectra, and the average predicted root-mean-square error of prediction (RMSEP) for multiple oxides was reduced by 13.38% by a convolutional neural networks (CNN) models trained using the transferred data. 22 Sun et al introduced polished rock spectra in the LIBS classification of raw rocks. Both of the datasets were utilized in the feature selection and the modeling.…”