Improving LIBS-based mineral identification with Raman imaging and spectral knowledge distillation

Area of Science:

• Multimodal spectral imaging
• Data-driven modeling
• Spectroscopy

Background:

• Combining data from different sensing modalities improves data-driven models.
• Multimodal spectral imaging enhances standalone spectroscopy through fusion, hyphenation, or knowledge distillation.
• Laser-induced Breakdown Spectroscopy (LIBS) is a valuable technique for material analysis.

Purpose of the Study:

• To enhance the performance of Laser-induced Breakdown Spectroscopy (LIBS) for mineral classification using knowledge distillation.
• To explore Raman spectroscopy as a supervisor for LIBS in a multimodal approach.
• To demonstrate the effectiveness of this method for challenging mineral identification tasks.

Main Methods:

• Implementation of a knowledge distillation pipeline where Raman spectroscopy acts as a supervisor for LIBS.
• Utilizing spectral imaging techniques to augment LIBS data.
• Case study focused on the identification of spodumene and petalite, challenging Li-bearing minerals.

Main Results:

• LIBS systems trained with Raman-derived labels showed enhanced classification performance compared to standalone LIBS.
• The knowledge distillation approach effectively improved the analytical capabilities of the LIBS system.
• The interpretability of the deployed model facilitated assisted feature discovery.

Conclusions:

• Knowledge distillation using Raman spectroscopy as a supervisor significantly improves LIBS performance for mineral classification.
• This multimodal strategy offers a promising avenue for enhancing single-technique systems in complex identification scenarios.
• The developed workflow has potential applications in both academic research and industrial settings for assisted feature discovery.