用无监督学习自动识别多晶衍射图中的晶粒对应斑点
Application of machine learning in grain-related clustering of Laue spots in a polycrystalline energy dispersive Laue pattern
- 结合层次聚类与K均值算法，将相似斑点聚类以还原晶粒结构
- 在模拟与镍丝实验数据上均实现准确识别，聚类数由肘部法确定
- 适合材料表征中快速分析多晶衍射图像的研究者使用
我们通过将能量色散劳埃衍射（EDLD）实验中晶粒对应劳埃斑点的识别问题建模为可由无监督机器学习解决的聚类问题，提出了一种新方法。为实现对劳埃图中晶粒的可靠高效识别，采用层次聚类（HC）与K均值算法相结合的方式，对相似劳埃斑点进行分组，从而揭示衍射图中的潜在晶粒结构。同时，利用肘部法确定最优聚类数量，以确保结果准确性。为评估该方法性能，我们在模拟数据和实际镍丝测量数据上进行了实验。模拟数据旨在模拟真实EDLD实验特征，实验数据来自实际测量。
原文摘要 · Abstract (English)
We address the identification of grain-corresponding Laue reflections in energy dispersive Laue diffraction (EDLD) experiments by formulating it as a clustering problem solvable through unsupervised machine learning (ML). To achieve reliable and efficient identification of grains in a Laue pattern, we employ a combination of clustering algorithms, namely hierarchical clustering (HC) and K-means. These algorithms allow us to group together similar Laue reflections, revealing the underlying grain structure in the diffraction pattern. Additionally, we utilise the elbow method to determine the optimal number of clusters, ensuring accurate results. To evaluate the performance of our proposed method, we conducted experiments using both simulated and experimental datasets obtained from nickel wires. The simulated datasets were generated to mimic the characteristics of real-world EDLD experiments, while the experimental datasets were obtained from actual measurements.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。