Convolutional neural networks applied to quantify the process behaviour of complex individual particles in froth flotation

The field of automated mineralogy has largely contributed to our understanding of mineral processing. Lately, by evaluating the particle information collected with automated mineralogy using statistical learning methods, it became possible to quantify the process behaviour of individual particles with consideration to their size, shape, liberation, and mineral association. Yet, automated mineralogy still requires a large intervention from operators for constructing an ore-specific mineral list and performing a series of image processing tasks. Here, we propose a method to quantify the process behaviour of individual particles using convolutional neural networks on the raw data collected with automated mineralogy: backscattered electrons and characteristic X-Ray signals. The flotation of a complex copper porphyry ore is used as a case study. The accuracy of the method is compared to the current standard procedure: manually processing the automated mineralogy data followed by particle-based modelling with a logistic regression.