Quantifying soot nanostructures: Importance of image processing parameters for lattice fringe analysis
Quantifying soot nanostructures: Importance of image processing parameters for lattice fringe analysis
复制标题
DOI：
10.1016/j.combustflame.2019.10.020
复制
发表时间：
2020-01
影响因子：
4.4
通讯作者：
S. A. Pfau;A. Rocca;M. Fay
中科院分区：
文献类型：
--
作者：
S. A. Pfau;A. Rocca;M. Fay
条纹分析是对碳烟纳米结构进行量化的常用方法。然而，所涉及的过滤器的设置及其对结果的影响很少被提及。在这项研究中，实验评估了三个滤波参数以及图像采集的两个方面的影响。为了进行分析，使用了一个炭黑以及一个柴油发动机和一个汽油直喷(GDI)发动机的碳烟样品。高斯低通滤波标准偏差大于1.5只会产生条纹度量的微小差异。2.0和3.0之间的标准差实现了条纹的逼真表示。发现白色顶帽变换盘大小与所有条纹分析指标之间存在线性关系。对于逼真的纳米结构表示，5像素和7像素的磁盘大小最合适。用Otsu方法计算的阈值通常会产生最好的纳米结构表示。任何偏差都会扭曲提取的条纹，并显著减少它们的总数。因此，建议在不改变的情况下一致使用OTSU阈值。由于欠聚焦和过聚焦而偏离中性电子显微镜焦点，导致条纹长度和大津阈值都有明显的下降。借助相应粒子的快速傅里叶变换进行一致聚焦对于获得可靠的结果至关重要。通过重复平均图像降低噪音水平的影响被发现比所使用的相机模型要小得多。感兴趣区域的大小与提取的条纹数呈线性相关，但不影响条纹度。为了统计上可靠的分析，建议至少有4000条条纹。GDI样品显示出最短的条纹和最高的曲折度。对于柴油烟尘和碳黑，可以观察到类似的条纹长度。曲度最大的是GDI碳烟，其次是柴油碳烟和碳黑。
Fringe analysis is a commonly used method to quantify soot nanostructures. However, the settings of the involved filters and their impact on the results are rarely addressed. In this study, the influence of the three filter parameters as well as two aspects of the image acquisition was assessed experimentally. For the analysis, a carbon black as well as one diesel engine and one gasoline direct injection (GDI) engine soot sample were used. Gaussian low-pass filter standard deviations larger 1.5 yielded only minor differences in fringe metrics. Standard deviations between 2.0 and 3.0 enabled realistic representation of fringes. A linear correlation was found between the white top-hat transformation disk size and all fringe analysis metrics. For realistic nanostructure representation, disk sizes of 5 px and 7 px are most suitable. Threshold values as calculated by Otsu's method generally yielded the best nanostructure representation. Any deviation distorted the extracted fringes and noticeably reduced their total number. Thus, consistent use of Otsu thresholds without alterations is advised. Deviating from the neutral electron microscope focus point by under- and over-focusing resulted in distinctive drops in both fringe lengths and Otsu thresholds. Consistent focusing with the help of fast Fourier transformations of the respective particles is vital for reliable results. The effect of reduced noise levels by repeated averaged images was found to be minor beyond the model of the camera used. The region of interest size correlated linearly with the number of extracted fringes, however, it did not affect the fringe metrics. For statistically reliable analysis, a minimum of 4000 fringes is suggested. The GDI sample exhibited the shortest fringes and the highest tortuosity. For diesel soot and carbon black, similar fringe lengths could be observed. The highest tortuosity was found for GDI soot, followed by diesel soot and carbon black.