利用基于深度学习的边缘检测进行原子探针层析成像中的相分割
核实验
2020-01-08 v1 仪器与探测器
摘要
原子探针层析成像（APT）有助于对微观结构特征进行纳米及原子尺度表征与分析。具体而言，APT 非常适用于研究颗粒或异相体系的界面性质。传统上，APT 数据中析出相与基体相之间界面的识别，要么通过基于用户提供的浓度值提取等浓度面，要么通过手动扰动浓度值直至等浓度面在定性上匹配界面来获得。这些方法主观、不可扩展，并且可能因局部成分不均匀性而导致不一致。我们提出了一种基于深度神经网络的数字图像分割方法，该方法将自然图像中学习到的知识迁移以自动将 APT 获得的数据分割为不同相。该方法不仅提供了一种分割数据和提取界面性质的高效途径，而且无需为训练分割模型进行昂贵的界面标注。我们在此考虑一个具有基体中的析出相以及三种不同界面形态——层状、孤立和互连——的系统，这些形态由析出相的不同相对几何结构获得。我们通过界面的定性可视化以及通过与使用更传统方法获得的近邻直方图进行定量比较，展示了我们分割方法的准确性。
引用
@article{arxiv.1904.05432,
title = {Final results for the neutron $\beta$-asymmetry parameter $A_0$ from the UCNA experiment},
author = {B. Plaster and E. Adamek and B. Allgeier and J. Anaya and H. O. Back and Y. Bagdasarova and D. B. Berguno and M. Blatnik and J. G. Boissevain and T. J. Bowles and L. J. Broussard and M. A. -P. Brown and R. Carr and D. J. Clark and S. Clayton and C. Cude-Woods and S. Currie and E. B. Dees and X. Ding and S. Du and B. W. Filippone and A. Garcia and P. Geltenbort and S. Hasan and A. Hawari and K. P. Hickerson and R. Hill and M. Hino and J. Hoagland and S. A. Hoedl and G. E. Hogan and B. Hona and R. Hong and A. T. Holley and T. M. Ito and T. Kawai and K. Kirch and S. Kitagaki and A. Knecht and S. K. Lamoreaux and C. -Y. Liu and J. Liu and M. Makela and R. R. Mammei and J. W. Martin and N. Meier and D. Melconian and M. P. Mendenhall and S. D. Moore and C. L. Morris and R. Mortensen and S. Nepal and N. Nouri and R. W. Pattie and A. Perez Galvan and D. G. Phillips and A. Pichlmaier and R. Picker and M. L. Pitt and J. C. Ramsey and R. Rios and R. Russell and K. Sabourov and A. L. Sallaska and D. J. Salvat and A. Saunders and R. Schmid and S. J. Seestrom and C. Servicky and E. I. Sharapov and S. K. L. Sjue and S. Slutsky and D. Smith and W. E. Sondheim and X. Sun and C. Swank and G. Swift and E. Tatar and W. Teasdale and C. Terai and B. Tipton and M. Utsuro and R. B. Vogelaar and B. VornDick and Z. Wang and B. Wehring and J. Wexler and T. Womack and C. Wrede and Y. P. Xu and H. Yan and A. R. Young and J. Yuan and B. A. Zeck},
journal= {arXiv preprint arXiv:1904.05432},
year = {2020}
}
备注
6 pages, 7 figures, to appear in proceedings of the International Workshop on Particle Physics at Neutron Sources 2018