A Comparative Study Towards Particle Identification Employing Semi-Automated Image Processing in Experimental SEM Images

Manually characterising particles in scanning electron microscopy (SEM) images is very time consuming and costly, often requiring an expert. In previous work we developed a method which has the potential to automate the particle analysis process with limited user interaction. In this study we compare the previously developed method with other automated machine learning (ML) approaches on experimentally obtained SEM images. We randomly selected four images each from two different datasets to test and compare the algorithms. The presented image processing methodology generally outperformed the selected ML method on the selected experimental SEM images. It was found that the presented method struggled to distinguish varying and grainy background from actual particles, leading to occasional overcounting, while the ML method consistently undercounted the particles.