Froth image feature engineering-based prediction method for concentrate ash content of coal flotation

Abstract

Machine vision and machine learning have been researched widely in froth flotation and the technology continues to benefit from advances in computer technology. The feature engineering leads to a predicted resistance in the machine learning pipelines, especially in coal flotation. This study sought to examine the performance of feature engineering of coal flotation froth image on ash content prediction with an industrial dataset. In order to evaluate the practical use in industry, the morphoscopic (3 features), statistical (gray levels histogram (5 features), gray level co-occurrence matrix (24 features), statistical modeling (48 features)) and color spaces (18 features) are used to prepare feature engineering. Correlation matrix are used to investigate the relationship between features and ash content. The support vector regression is used to predict ash content. The evaluation of the model performance shows that the principal component analysis can effectively improve the accuracy. When the feature dimension is reduced to 14 by the principal component analysis, the optimal RMSE is 0.6331, the R2 value is 0.78. The feature engineering of coal flotation froth image in this paper can make a good prediction of the coal flotation concentrate ash content. Furthermore, the results can be used as the theoretical basis for the intelligent construction of flotation.