DefectTrack: a deep learning-based multi-object tracking algorithm for quantitative defect analysis of in-situ TEM...

Area of Science:

• Materials Science
• Nuclear Engineering
• Data Science

Background:

• In-situ irradiation transmission electron microscopy (TEM) is crucial for studying irradiation damage mechanisms in nuclear materials.
• Analyzing millisecond-timescale post-cascade processes, like defect cluster dynamics, is a significant bottleneck.
• Current methods struggle to extract quantitative data on defect cluster properties from complex TEM videos.

Purpose of the Study:

• To develop an automated solution for tracking defect clusters in in-situ TEM videos.
• To enable real-time analysis of defect cluster dynamics, including lifetime and thermal stability.
• To improve the mechanistic understanding of irradiation damage in nuclear materials.

Main Methods:

• Introduction of DefectTrack, a deep learning-based one-shot multi-object tracking (MOT) model.
• Application of DefectTrack to track cascade-induced defect clusters in in-situ TEM videos.
• Evaluation of DefectTrack's performance using MOT metrics (MOTA, MT) and comparison with human expert analysis.

Main Results:

• DefectTrack achieved a Multi-Object Tracking Accuracy (MOTA) of 66.43% and Mostly Tracked (MT) of 67.81%.
• Performance is comparable to state-of-the-art MOT algorithms.
• Statistical analysis showed DefectTrack outperforms human experts in accuracy and speed for quantifying defect cluster lifetime distributions.

Conclusions:

• DefectTrack is the first dedicated MOT model for real-time tracking of defect clusters in in-situ TEM videos.
• The model significantly advances the analysis of irradiation damage by automating defect cluster tracking.
• DefectTrack offers a more accurate and efficient method for understanding defect dynamics in nuclear materials.