Deep-layers-assisted machine learning for accurate image segmentation of complex materials

Davydzenka, Tsimur; Sinclair, Daniel; Chawla, Nikhilesh; Tahmasebi, Pejman

doi:10.1016/j.matchar.2022.112175

Cited by 11 publications

(5 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…U-Nets can achieve acceptable performance with less volumes of data. The validity of U-Net is verified in various applications, including the segmentation of 3D tomography images [19,20] and the determination of the thickness of 2D van der Waals heterostructures from optical microscopy images [21]. As shown in figure 1, the algorithm consists of a training phase and a prediction phase.…”

Section: Introductionmentioning

confidence: 91%

A U-Net-based self-stitching method for generating periodic grain structures

Ji,

Koeppe,

Altschuh

et al. 2024

Phys. Scr.

View full text Add to dashboard Cite

When modeling microstructures, the computational resource requirements increase rapidly as the simulation domain becomes larger. As a result, simulating a small representative fraction under periodic boundary conditions is often a necessary simplification. However, the truncated structures leave nonphysical boundaries, which are detrimental to numerical modeling. Here, we propose a self-stitching algorithm for generating periodic structures, demonstrated in a grain structure. The main idea of our algorithm is to artificially add structural information between mismatched boundary pairs, using the hierarchical spatial predictions of the U-Net. The model is trained with 20,000 grain sample pairs simulated from multiphase field simulations, resulting in the successful generation of periodic grain structures as expected. Furthermore, we employ an energy-based metric, the local curvature, to highlight the quality of the generated samples. Through this metric, we determine that the optimum value of the width of the mask is 1/16 of the sample width. The algorithm provides an automatic and unbiased way to obtain periodic boundaries in grain structures and can be applied to porous microstructures in a similar way.

show abstract

Section: Introductionmentioning

confidence: 91%

A U-Net-based self-stitching method for generating periodic grain structures

Ji,

Koeppe,

Altschuh

et al. 2024

Phys. Scr.

View full text Add to dashboard Cite

show abstract

“…Therefore, the segmentation task aims to accurately identify the voxels belonging to highly complex sample structures or partial areas of interest from reconstructed 3D voxels, which would aid in the subsequent analysis process. Numerous efforts have been made, including the use of CNN-based models, 97 UNet, 98 , 99 and Mask R-CNN. 100 Here, we focus on the UNet-based approach and the highly accurate Mask R-CNN approach.…”

Section: Scientific Application-oriented Data Processing On Reconstru...mentioning

confidence: 99%

“…In 2022, Davydzenka et al. 98 introduced a typical UNet architecture for highly accurate segmentation of complex Mg-based alloys among reconstructed voxels. In addition to the segmentation advantage of the UNet architecture itself, segmentation results were further improved using a data augmentation method suitable for the tested samples.…”

Section: Scientific Application-oriented Data Processing On Reconstru...mentioning

confidence: 99%

Towards full-stack deep learning-empowered data processing pipeline for synchrotron tomography experiments

Zhang,

Li,

Wang

et al. 2024

The Innovation

View full text Add to dashboard Cite

“…The results show that the proposed algorithm obtained similar porosity values as were determined by the experts using traditional image processing methods. Davydzenka et al [17] used a deep-layer-assisted machine learning technique for segmentation of complex Mg-based alloy. They have shown that using a larger initial training set an increase in average accuracy of segmentation from 81.1% to 90.0% is possible.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Currently, on the other hand, some of most powerful tools for image segmentation are deep convolutional neural networks, which have shown promising results in recent years [15][16][17]. However, they are supervised methods that require training.…”

Section: Introductionmentioning

confidence: 99%