About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Integration of Simulation, Crowdsourcing, and Complexity Scoring for Ambiguous Feature Evaluations in Materials Imaging |
| Author(s) |
Gabriella Bruno, Ryan Jacobs, Matthew J. Lynch, Dane Morgan, Kevin Field |
| On-Site Speaker (Planned) |
Kevin Field |
| Abstract Scope |
We present a novel ambiguity analysis pipeline for materials images that leverages simulated TEM datasets with controllable feature size, density, and background parameters to probe feature-image ambiguity systematically. By generating diverse, synthetic TEM images and applying crowdsourcing for feature annotation, we quantify per-feature ambiguity as the fraction of labelers detecting each placed feature. These granular ambiguity scores are directly compared with global image complexity metrics—such as entropy, clutter, and symmetry—and local complexity scores derived from feature neighborhoods to evaluate how image characteristics impact human annotation accuracy rigorously. Our results illuminate the interplay between both global and localized sources of complexity and the detectability of radiation-induced cavities, key for assessing radiation tolerance in advanced nuclear materials. This framework aids in preemptively identifying images with high ambiguity, ultimately enhancing the reliability of human-based and automated cavity detection in highly complex or noisy TEM images central to materials development. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Nuclear Materials, Characterization, Other |