Application of deep learning on quantitative analysis of binary solid dispersions by UV Raman spectroscopy for planetary exploration

As a surface mineral analysis and detection technology, Raman spectroscopy is widely used in deep space exploration because of its advantages, such as non-destructiveness, high resolution, no need for sample preparation, and real-time analysis capability. Based on the unique fingerprint Raman spectra of different minerals and organics, Raman spectroscopy can be used to accurately identify the species and qualitatively analyze its chemical compositions, which would help to further explain the geological formation and alteration processes and determine the potential biological characteristics during the planetary exploration missions. In this study, a deep learning model, which is called Inception-ResNet-v1 model with a squeeze-and-excitation block (IRMSE), for the quantitative analysis of UV Raman spectra is proposed and investigated. This model can automatically learn and extract component information from Raman spectra through deep residual networks and use attention mechanisms to select the most critical information for the current task from redundant features. Experimental results indicate that the prediction accuracy of the proposed model is much better than that of traditional methods for the quantitative analysis of the solid dispersions of different contents between minerals or between minerals and organic compounds. Therefore, this study validates the feasibility of using deep learning for the quantitative analysis of minerals and organic materials by Raman spectra collected by planetary exploration missions.