Using Convolutional Neural Networks to Develop State-of-the-Art Flotation Froth Image Sensors

Abstract

Convolutional neural networks provide a state-of-the-art approach to the development of froth image sensors. In this study, it is shown that a pretrained neural network architecture, namely VGG16, can be used to obtain significant improvements in froth image sensors. However, training of these networks is computationally demanding and require large data sets that may not be readily available. These problems can be circumvented by making use of transfer learning and partial retraining of the network. Likewise, minor modification of the network architecture can also expedite the development of the models. This is demonstrated in a case study involving an image data set from an industrial platinum group metals plant.