In this paper, an approach to predicting randomly-shaped particle volume based on its two-Dimensional (2-D) digital image is explored. Conversion of gray-scale image of the particles to its binary coun-terpart is first performed using backlighting technique. The silhouette of particle is thus obtained, and conse-quently, informative features such as particle area, centroid and shape-related descriptors are collected. Several dimensionless parameters are defined, and used as regressor variables in a multiple linear regression model to predict particle volume. Regressor coefficients are found by fitting to a randomly selected sample of 501 parti-cles ranging in size from 4.75mm to 25mm. The model testing experiment is conducted against a different ag-gregate sample of the similar statistical properties, the errors of the model-predicted volume of the batch is within ±2%.
展开▼
