more>>在线期刊
more>>新闻发布
more>>相关下载
友情连接

透射电镜像差测量数据集的构建及优化

杨威威#，郭 惠#，陈桂森，李德响，王翰为，何玉涛*，明文全，陈江华*
  • 摘要
  • 参考文献
  • 相关文章

2025年 第44卷 第5期: 1000-6281（2025）05-531-14     下载地址：全文下载请到同方知网，万方数据库或重庆维普等数据库中下载！

透射电镜像差测量数据集的构建及优化

杨威威1，2#，郭  惠1，2#，陈桂森3，李德响1，2，王翰为1，2，何玉涛1，2*，明文全1，2，陈江华1，2*

(1. 海南大学热带海洋工程材料及评价全国重点实验室，海洋材料表征技术创新研究院，海南 海口 570228；2. 海南大学精密仪器高等研究中心，皮米电子显微学海南省重点实验室，海南 海口 570228；3. 湖南大学材料科学与工程学院高分辨电镜中心，湖南 长沙 410082)

摘　要　　在透射电子显微镜中，准确测量像差参数对波函数重构、高分辨率三维电子断层成像及图像定量模拟至关重要。目前常用的检测像差的方法通常需要在已知球差的前提下，利用几何平均算法估算其他低阶像差参数。卷积神经网络作为一种非线性图像分析工具，理论上可通过监督学习从大量功率谱图中直接检测像差参数（包括球差）。然而，实验获取的二维功率谱图易受衰减和噪声影响，往往无法准确呈现环形特征，且实验中获取大量图像费时繁琐，这使得构建庞大且高质量的数据集成为该方法面临的主要挑战之一。针对上述问题，本文提出了一种二维功率谱图数据集构建程序及可视化用户操作界面。该方法的优势在于：只需输入一张待检测的二维功率谱图，即可自动生成所需的数据集，并自动确定各项像差参数的范围、删除重复项以及排除采样伪像的影响。

关键词　　透射电子显微镜；衬度传递函数；像差

中图分类号：TN911.73; TP391.41; O437.2      文献标识码：A Doi:10.3969/j.issn.1000-6281.2025.05.003

 

Construction and optimization of a transmission electron microscopy aberration measurement dataset

YANG Weiwei1, 2#, GUO Hui1, 2#, CHEN Guisen3, LI Dexiang1, 2, WANG Hanwei2, HE Yutao1, 2*, MING Wenquan1, 2, CHEN Jianghua1, 2*

(1. Innovation Institute for Ocean Materials Characterization Technology, State Key Lab of Tropic Ocean Engineering Materials and Materials Evaluation, Hainan University, Haikou Hainan 570228; 2. Key Laboratory of Pico Electron Microscopy of Hainan Province, Center for Advanced Studies in Precision Instruments, Haikou Hainan 570228; 3. Center for High-Resolution Electron Microscopy, College of Materials Science & Engineering, Hunan University, Changsha Hunan 410082, China)

Abstract　　Accurate measurement of aberration parameters in transmission electron microscopy is critical for wavefunction reconstruction, high-resolution three-dimensional electron tomography, and quantitative image simulation. Existing methods for aberration detection typically require known spherical aberration as a prerequisite and use geometric mean algorithms to estimate other low-order aberrations. As nonlinear image analysis tools, Convolutional neural networks theoretically offer the capability to directly detect aberration parameters (including spherical aberration) through supervised learning on large datasets of power spectra. However, experimentally obtained two-dimensional power spectra are often affected by attenuation and noise, making it difficult to accurately represent ring features. Moreover, obtaining a large number of experimental images is time-consuming and labor-intensive, posing a significant barrier to constructing large-scale, high-quality datasets for model training. To overcome these limitations, this paper presents a two-dimensional power spectrum dataset construction method along with a visual user interface. The proposed method enables automatic generation of datasets from a single input power spectrum image, while automatically determining aberration parameter ranges, removing duplicates, and eliminating sampling artifacts.

Keywords　　transmission electron microscopy; contrast transfer function; aberration correction