Automated Detection of Primary Particles from Transmission Electron Microscope (TEM) Images of Soot Aggregates in Diesel Engine Environments

Abstract

The major challenge of the post-processing of soot aggregates in transmission electron microscope (TEM) images is the detection of soot primary particles that have no clear boundaries, vary in size within the fractal aggregates, and often overlap with each other. In this study, we propose an automated detection code for primary particles implementing the Canny Edge Detection (CED) and Circular Hough Transform (CHT) on pre-processed TEM images for particle edge enhancement using unsharp filtering as well as image inversion and self-subtraction. The particle detection code is tested for soot TEM images obtained at various ambient and injection conditions, and from five different combustion facilities including three constant-volume combustion chambers and two diesel engines. Through a comparison between automatically detected and manually selected primary particles from extensive datasets, five key image-processing parameters of the self-subtraction level, negative Laplacian shape parameter, maximum and minimum diameter of primary particles, and CHT sensitivity are optimised. From the analysis of the size distribution and mean diameter of primary particles, it is found that the automatic method is much more dependent upon the minimum primary particle diameter and CHT sensitivity than the other three parameters. With the optimised set values, the new particle detection code shows a good agreement with the results from the manual method.

External irradiation effect on the growth and evolution of in-flame soot species

“…A summary of the mean values of the soot aggregates at different HABs in NI-flame and Iflame is given in Table 4-2. From the table, the values are observed to range between 1.7 and 1.9 for both flames, which are within the asymptotic values that are typically reported in the literature (i.e., [35,57]). In the NI-flame, the soot aggregates are found to have values of 1.73, 1.76 and 1.84 at 20, 30 and 40 mm HABs, respectively.…”

“…However, the thermophoretic sampling of soot coupled with transmission electron microscope (TEM) imaging remains the only method that permits a direct observation of soot morphological information. The soot morphological information is useful to the combustion research community as it provides important insights to the formation and evolution of soot particulates in different combustion conditions [30,[34][35][36][37].…”

“…The identification and the measurement of the soot particles from the TEM images, however, can be complex as the soot TEM images are typically of low contrast, noisy and have non-uniform background signal level due to the nature of the TEM imaging process [38]. Additionally, the processing of the TEM images, which mainly proceeds through direct manual assessment, can be challenging and time-consuming as the soot samples have some degree of overlapping, in addition to having large variability in shape and size, even when they are collected from the same combustion environment [35,39]. A large number of TEM images would therefore need to be processed manually for statistically meaningful quantities to be generated.…”

“…The operator would also need to have a clear understanding of the object structures and properties, to be able to consistently distinguish the objects from the background when analysing the images [35,39].…”

“…The objective of the first stage of the proposed image processing sequence is to reduce the image noise, whilst maintaining the edge information of the aggregates. This is to be achieved through the use of median filtering and self-subtraction [35]. In the second stage of the proposed image processing sequence, an edge detection method (Canny Edge Detection algorithm) is applied to identify the contours that are present at both the external and internal regions of the soot aggregates.…”