Computer Learning Based Grain Measurements.

摘要

Measurement of grain size is an important part of LANL's analysis of metallographic images of plutonium alloys. This has been accomplished in the past by using the Abrams three circle method or manual grain tracing. We have developed a computer based grain recognition program that discovers grain boundaries based on multiple criteria, and segments the image into well defined grains. The boundaries between grains are defined virtually, ie without width, so that morphological measurements do not have to be corrected to account for boundary width. Machine learning techniques are being applied to this grain segmentation process to minimize worker input and maximize correct grain identification.