[Research in Magnesite Grade Classification Based on Near Infrared Spectroscopy and ELM Algorithm]

Area of Science:

• Mineralogy and Materials Science
• Spectroscopy
• Artificial Intelligence

Background:

• Industrial development necessitates efficient magnesite grade determination.
• Variability in magnesite content and distribution complicates traditional grading methods.
• Accurate and rapid classification is crucial for resource management and utilization.

Purpose of the Study:

• To develop a rapid and accurate magnesite grade classification model.
• To integrate near-infrared spectroscopy (NIR) with an Extreme Learning Machine (ELM) algorithm.
• To compare the proposed model's performance against traditional and other AI-based methods.

Main Methods:

• Near-infrared (NIR) spectroscopy was employed to analyze magnesite samples, leveraging differential absorption by H groups.
• Principal Component Analysis (PCA) was used for dimensionality reduction of NIR spectral data.
• An Extreme Learning Machine (ELM) model was established for quantitative analysis, with variations including optimized ELM and integration-Featured ELM.

Main Results:

• The PCA reduced data dimensionality, retaining over 99.99% of element contribution rate with 10 characteristic variables.
• The ELM model, trained and tested on magnesite samples, demonstrated high classification accuracy exceeding 90%.
• Improved ELM models showed enhanced classification performance compared to traditional ELM, artificial, chemical, and BP neural network methods.

Conclusions:

• The combination of NIR spectroscopy and ELM provides a cost-effective, time-efficient, and accurate solution for magnesite grade classification.
• This integrated approach offers a novel and superior method for the rapid assessment of magnesite quality.
• The developed model presents a significant advancement in mineral resource analysis and industrial application.