Visual ore quality assessment by image analysis

In order to estimate the amount of oil that can be recovered from oil sands slurry, technically referred to as processability number, we propose a method based on image processing in this paper. Our study begins with a review of human observations in conducting this task to determine visual features pertinent for assessing ore slurry quality. Subsequently we extract potentially useful image features and use them to train a regressor to learn the relationship between the visual features and the ore quality. Specifically, an input image is first divided into three layers representing different materials in the slurry through image segmentation. Image features in the bottom two layers are then extracted. We create three types of features - grayscale features, Haralick features and the power spectrum - to evaluate their ability to predict the processability number. For this purpose, a regressor model is trained using Adaboost with one or more types of the visual features as input. Experimental results show that the Haralick features provide the best estimate of the ore quality in terms of the procesability number, and that it is possible to design an automated system for assessing the quality of an industrial product through image processing techniques.