Estimating Ore Particle Size Distribution using a Deep Convolutional Neural Network
Laurentz E. Olivier, Michael G. Maritz, I.K. Craig
Abstract
In this work the ore particle size distribution is estimated from an input image of the ore. The normalized weight of ore in each of 10 size classes is reported with good accuracy. A deep convolutional neural network, making use of the VGG16 architecture, is deployed for this task. The goal of using this method is to achieve accurate results without the need for rigorous parameter selection, as is needed with traditional computer vision approaches to this problem. The feed ore particle size distribution has an impact on the performance and control of minerals processing operations. When the ore size distribution undergoes significant changes, operational intervention is usually required (either by the operator or by an automatic controller).