A new method for cross-calibration between LIBS systems based on a domain adaptive fully connected network

Abstract

As an important elemental analysis method, laser-induced breakdown spectroscopy (LIBS) enables the qualitative and quantitative determination of the compositions of geological targets on Mars. In order to analyze the chemical compositions of Martian rocks and soils at Gale crater, the ChemCam team built a LIBS spectral library of 408 geological standards using a ChemCam's ground replica under a simulated Martian atmosphere. However, the lack of spectra data with component labels makes it difficult to establish a composition inversion model for predicting the collected spectra. The Mars Surface Composition Detector (MarSCoDe) carried by China's Mars rover, Zhurong, also employs LIBS technology and has successfully conducted related composition detection work. Therefore, this paper proposes a domain-adaptive fully connected neural network (DAFCN) capable of transferring prior knowledge between ChemCam and SDU-LIBS (a LIBS system developed by Shandong University). The model extracts the features of the source domain and target domain through a three-layer fully connected network and uses a domain adaptation function to reduce the difference in feature distribution between the source domain and target domain. Experimental results show that the proposed method is superior to the traditional method without migration. Further, we tried this method to establish the connection between MarSCoDe and other LIBS devices, and then quantitatively analyzed the spectra collected by MarSCoDe.