Title: Pattern recognition techniques for sand particles

Abstract

The pressing need to recognize and track individual sand particles in fundamental research on geomechnics has promoted the rapid development of particle tracking techniques in recent years. This talk presents the latest development of a few innovative pattern recognition techniques for identifying and atching intact and crushed sand particles. These techniques include particle volume-based tracking (PV-track), particle radius -based track (PR-track), spherical harmonics-based tracking (SH-track) and point cloud -based tracking (PL-track). Specifically, PV-track and PR-track are suitable for tracking particles within a neighborhood area but the tracking accuracy and reliability decreases with the increasing deformation of the sand specimen. SH-track is a much more powerful and robust technique which makes use of the SH invariant describing the multiscale morphological features of sand particles. However, the common limitation of PV-track, PR-track and SH-track is that they can only be applied to intact particles with solid structures (i.e., non-porous structure). In contrast, PL-track can deal with both intact and crushed sand particles and has been successfully used to match a group of crushed quartz particles. More importantly, PL-track can be integrated with machine learning techniques to achieve intelligent recognition and tracking, and has been successfully used to identify a group of highly porous carbonate sand particles. The implementation of all these particle tracking techniques is based on the X-ray microtomography scanning of a miniature specimen of sands, which provides the source data for the pattern recognition exercise.