Feature recognition of irregular pellet images by regularized Extreme Learning Machine in combination with fractal theory

Abstract

As we know, the fractal theory is an effective method for irregular image processing and analysis, which is difficult to implement using other baseline methods. However, Extreme Learning Machine (ELM) algorithm in combination with fractal theory can provide a better solution than the single fractal theory in image processing, transformation, data prediction, and so on. The features including fractal dimension, area ratio, and average perimeter of hematite, magnetite, calcium silicate, calcium ferrite, and stomata in the microstructure of irregular pellet images are extracted by artificial intelligence (AI) technologies. An AL-based ELM model suitable for the features of pellet images was constructed, the ELM model was optimized by regularization, and the feature parameters of pellet images were predicted by improved ELM algorithm, the optimal number of neuron nodes in the hidden layer was obtained for every mineral. The results show that the efficiency of model training and prediction in the feature for irregular pellet images is highly comparable to baseline methods. The method can be used as a reference in the intelligent prediction, control, and optimization of pellet image quality.