Application of deep learning in laser-induced breakdown spectroscopy: a review

Due to its remarkable element detection capability, laser-induced breakdown spectroscopy (LIBS) has been extensively utilized for element-related analysis. Although the spectral peaks corresponding to different elements can be manually identified by the LIBS spectra library, fully extracting information from LIBS spectra remains challenging due to measurement uncertainty interference. To improve the performance of LIBS analysis, various machine learning (ML) methods have been proposed to compensate for the measurement uncertainty. Among these, deep learning (DL), the most cutting-edge topic in artificial intelligence, has been applied in spectroscopy analysis in recent years. This work presents the first review of DL approaches in LIBS spectra analysis, where the principles and applications are introduced and summarized. A comprehensive discussion on current applications, challenges, and future perspectives is conducted to provide guidelines for future research and applications. The reviewed papers demonstrate that DL exhibits great potential and a promising future in LIBS analysis.