Machine learning shadowgraph for particle size and shape characterization

Abstract

Conventional image processing for a particle shadow image is usually time-consuming and suffers degraded image segmentation when dealing with images consisting of complex-shaped and clustered particles with varying backgrounds. In this paper, we introduce a robust learning-based method using a single convolution neural network for analyzing particle shadow images. Our approach employs a two-channel-output U-net model to generate a binary particle image and a particle centroid image. The binary particle image is subsequently segmented through a marker-controlled watershed approach with the particle centroid image as the marker image. The assessment of this method on both synthetic and experimental bubble images has exhibited a better performance compared to the state-of-art non-machine-learning method. The proposed machine learning shadow image processing approach provides a promising tool for real-time particle image analysis in industrial applications.

A deep learning-based framework for automatic analysis of the nanoparticle morphology in SEM/TEM images

“…For example, Jiaxu Duan et al 21 proposed a light-weight network based on a U-shaped encoder–decoder structure to extract the binarized masks of particles. Moreover, Zhengyu Liang et al 22 and Jiaqi Li et al 23 proposed light-weight nanoparticle segmentation networks, respectively, to improve the framework. These light-weight nanoparticle segmentation networks only segment specific types of particles in relatively complex SEM/TEM images while performing poorly in processing SEM/TEM images with high background noise, extremely small nanoparticles and dense nanoparticles.…”

Advanced Deep Learning-Based Bubbly Flow Image Generator under Different Superficial Gas Velocities

“…20 Li et al trained a dual-channel U-net model with synthetic images, and it showed high robustness in the test system. 21 In this paper, a gas−water two-phase bubbly flow image generator based on the StyleGan2 network model is established on 15 000 images obtained at 10 different superficial gas velocities. 16 Meanwhile, a bubble detector based on the Yolov3 deep learning network is trained, based on which quantitative comparisons between the synthetic and the original input samples are implemented to validate the reliability and ability of the proposed scheme.…”

“…Li et al. trained a dual-channel U-net model with synthetic images, and it showed high robustness in the test system …”

Bubble Identification from Images with Machine Learning Methods

“…Such pixel-topixel approaches have become very popular for detecting and segmenting cells and nuclei in biomedical microscopic images. For bubbly flows, Li et al, (2021) used a UNet to distinguish between foreground and background pixels as well as to generate centroid approximations. Kim and Park, (2021) used a slightly customized Mask-RCNN version that directly provides a segmentation mask as a result.…”