Particle Size Analysis using Deep Learning Techniques

Abstract:

Recent times have seen immense progress in deep learning techniques and its ability to solve image-based problems to the point of exceeding human performance. Particle sizing of ore and coal from images has been used in the minerals processing industry for many years and provides an important measurement for crushing and milling operations. This paper shows how modern deep learning models have been utilized to provide a step-change in measurement accuracy. The new technique is shown to be more human-like and much more robust to deterioration in image quality caused by sunlight, shadows and dust. The paper also shows the increase in performance in an industrial application where image analysis is used to detect large boulders on haul trucks in order to avoid crusher damage and consequential downtime.