Identification and Characterization of Particle Shapes from Images of Sand Assemblies Using Pattern Recognition

Abstract

Pattern recognition was used on images of contacting three-dimensional sand particle assemblies to detect foreground sand grains that exhibit full-particle projections. Once identified, the roundness and sphericity of those particles could be determined using previously developed computational geometry image analysis methods. The pattern recognition used cascade and sliding window techniques from the literature with modifications to account for the unique features of sand grains, including variable sizes and a range of elongations and orientations of long axes in an image. A library of over 85,000 images was created to train the pattern recognition program. An example problem illustrated the capabilities of the method to generate particle roundness and sphericity curves.