Novel enhanced particle size analysis algorithm using enhanced circular Hough transformation

Abstract text (incl. figure legends and references)

Nanoparticles (NP) have a wide range of applications in diverse research fields, such as lithium ion batteries, catalysis, medical engineering, etc., due to their specific characteristics [1]. The surface effects become predominant as the dimension of the particles is reduced. Therefore, physical and chemical properties of NPs are closely intertwined with their particle size [2]. Due to this critical parameter of particle size, only a slight deviation from the target size is usually intended to achieve uniform characteristics for a specific application. Hence, the reliable analysis of the particle size distribution is an important task for successful implementation of advanced nanotechnological applications.

NPs size distribution can be analyzed by measuring individual particle size from a large number of images of NPs obtained by electron microscopy. The automatic analysis tools are essential, as a large number of images are required for obtaining reliable statistics for the size distribution [3]. However, currently available automatic procedures are limited to provide accurate results if the NPs are severely overlapping. In order to overcome these limitations, we developed a novel approach for the detection of spherical NPs based on the iterative particle analysis using the enhanced circular Hough transformation (CHT) algorithm. This approach solves the issue of overlapping particle clusters by iteratively detecting large particles first, removing them from the image and then find the smaller particles from the remainder of the cluster. The main loop of our iterative routine splits a singular particle size interval into multiple smaller ones and passes through them one by one. This approach keeps consistent detection quality over very large intervals. The CHT algorithm involves finding the best fit based on only contour detection of particles. However, utilizing this approach leads to critical problems, such as the halo error, void error and particle group error, as the CHT algorithm includes no local intensity analysis as explained in detail in Figure 1. In order to avoid the aforementioned problems, additional local intensity criteria are included.

We investigated synthesized FePt NPs using high angle annular dark field contrast (HAADF) imaging via a double-abberation corrected JEOL NeoARM 200F. From the single HAADF image of the FePt NPs at the magnification of roughly 2 million, 630 particles were automatically detected after manual removal of badly detected 19 particles using our algorithm (Figure 2). The measured mean particle size was 2.563 nm with a standard deviation of 0.485 nm. Although the size of NPs varies significantly with severely overlapping particles forming complex clusters in the selected system, overall our novel iterative particle analysis algorithm provided promising results in particle size distribution, showing very minor problems which can be corrected with available manual correction.

References:
[1] Görke, M. and Garnweitner, G., Crystal engineering of nanomaterials: current insights and prospects , CrystEngComm, 2021
[2] Issa, B. et al., Magnetic Nanoparticles: Surface Effects and Properties Related to Biomedicine Applications, International Journal of Molecular Science, 2009
[3] Meng, Y. et al., Automatic detection of particle size distribution by image analysis based on local adaptive canny edge detection and modified circular Hough transform, Micron, 2018