High-performance identification of insulating materials by using generalized spectrum in laser-induced breakdown...

Area of Science:

• Materials Science
• Analytical Chemistry
• Spectroscopy

Background:

• Accurate identification of insulating materials is vital for waste reduction, pollution control, and recycling initiatives.
• Traditional methods often struggle to retain comprehensive spectral information during dimensionality reduction.

Purpose of the Study:

• To introduce and evaluate a novel Generalized Spectrum Method for Laser-Induced Breakdown Spectroscopy (GSM-LIBS).
• To demonstrate GSM-LIBS's superiority over Principal Component Analysis (PCA) in spectral analysis for material identification.

Main Methods:

• Developed GSM-LIBS by integrating multiple spectral features (peak intensity, integral intensity, etc.) to preserve global and local spectral information.
• Applied GSM-LIBS to classify seven types of insulating materials.
• Validated performance using machine learning models: k-nearest neighbor (KNN), support vector machine (SVM), and neural network (NN).

Main Results:

• GSM-LIBS effectively reduced spectral dimensions while retaining critical information like plasma state and element concentration.
• Classification accuracy improved significantly: KNN from 0.935 to 0.979, SVM from 0.965 to 0.996, and NN from 0.984 to 0.994.
• GSM-LIBS demonstrated superior performance compared to PCA-LIBS across all tested machine learning models.

Conclusions:

• GSM-LIBS offers a more effective approach to spectral dimensionality reduction and feature extraction for material identification.
• The method significantly enhances the accuracy of insulating material classification, supporting recycling and waste management.
• GSM-LIBS shows broad applicability and effectiveness in analytical spectroscopy.