Comparison of Convolutional and Conventional Artificial Neural Networks for Laser-Induced Breakdown Spectroscopy...

Area of Science:

• Artificial Intelligence
• Chemometrics
• Spectroscopy

Background:

• Deep learning algorithms have enabled advanced AI applications in digital imaging and diagnostics.
• These methods are being explored for complex chemometric analyses, including laser-induced plasmas.

Purpose of the Study:

• To evaluate the advantages and disadvantages of convolutional neural networks (CNNs) in laser-induced breakdown spectroscopy (LIBS).
• To assess the practical potential of "deep LIBS" for classification and quantitative analysis.
• To compare deep learning approaches with traditional shallow artificial neural networks.

Main Methods:

• Utilizing convolutional neural networks (a deep learning technique) for LIBS data analysis.
• Comparing the performance of deep LIBS against shallow artificial neural networks.
• Case study involving the analysis of six bronze samples with known compositions.

Main Results:

• Convolutional neural networks offer significant potential for advanced LIBS applications.
• Deep LIBS demonstrates competitive or superior performance compared to shallow networks in classification and quantitative analysis.
• The study provides insights into the practical applicability of deep learning in LIBS.

Conclusions:

• Deep learning, particularly CNNs, presents a powerful tool for enhancing LIBS capabilities.
• The findings support the integration of deep LIBS for routine analytical tasks.
• Further research can expand the application of deep LIBS to diverse materials and analyses.