Automated cement fragment image segmentation and distribution estimation via a holistically-nested convolutional network and morphological analysis

Abstract

The distance between two rollers in a ball grinding mill is determined by the size of cement fragments. As such, automated detection of fragment size distribution is of great importance to the cement production industry. Therefore, we propose a holistically-nested convolutional network (HCN) and corresponding morphological analysis to estimate cement fragment distributions. This procedure can be divided into three stages. First, a cement fragment image training set was input to the HCN, helping the algorithm distinguish between fragments and the background. A series of morphological operations including morphological clean and opening operation were then employed to improve segmentation performance and the accuracy of fragment size calculation. Finally, the ‘open’ operation was employed to obtain the fragment distribution from a segmented image. Experimental results demonstrated that the segmentation and size calculation achieved using our algorithm were superior to those using comparable conventional techniques. The precision,F1-measure, PRI and VI using proposed method are improved at least 1.3%, 3.3%, 9.5% and 36.5% respectively.