Coal and Gangue Classification Based on Laser-Induced Breakdown Spectroscopy and Deep Learning.

Area of Science:

• Geoscience and Materials Science
• Spectroscopy and Analytical Chemistry
• Artificial Intelligence and Machine Learning

Background:

• Coal processing generates significant gangue (waste rock), comprising 15-20% of yield with low carbon and high ash content.
• Effective separation of coal and gangue is crucial for reducing waste and enhancing the efficiency of coal utilization.
• Current separation methods may lack the precision required for optimal resource management.

Purpose of the Study:

• To develop and validate a novel classification method for distinguishing coal from gangue.
• To leverage advanced spectroscopic techniques and deep learning for automated and accurate material sorting.
• To improve the efficiency and economic viability of coal processing operations.

Main Methods:

• Utilized laser-induced breakdown spectroscopy (LIBS) to obtain spectral data from coal and gangue samples.
• Transformed 1D spectral data into 2D time-series representations using Gramian angular summation fields (GASF).
• Developed and applied a novel deep learning model, GASF-CNN, incorporating SimAM attention and residual connectivity for classification.

Main Results:

• The GASF-CNN model achieved high performance across key evaluation metrics.
• Achieved classification accuracy of 98.33%, recall of 97.06%, precision of 100%, and an F1 score of 98.51%.
• Demonstrated superior performance compared to other conventional machine learning and deep learning models.

Conclusions:

• The proposed GASF-CNN method provides an accurate and effective approach for coal and gangue classification.
• This technique holds significant potential for optimizing coal processing and waste management in the mining industry.
• The integration of LIBS, GASF, and deep learning offers a powerful tool for material characterization and sorting.