Segment SEM images
The result's identifiers
Result code in IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27740%2F24%3A10256099" target="_blank" >RIV/61989100:27740/24:10256099 - isvavai.cz</a>
Result on the web
<a href="https://code.it4i.cz/SEM-Image/segment_sem_images_hctpm.git" target="_blank" >https://code.it4i.cz/SEM-Image/segment_sem_images_hctpm.git</a>
DOI - Digital Object Identifier
—
Alternative languages
Result language
angličtina
Original language name
Segment SEM images
Original language description
The study that is supported by the provided software demonstrated that the deep learning model can accurately segment SEM (scanning electron microscope) images, achieving high dice scores for coating layers, pores, and precipitates. The result highlights the significance of the Blender-generated synthetic data in training the deep learning model. The use of blend files for creating realistic synthetic SEM images was pivotal in enhancing segmentation accuracy, particularly for limited real data, showcasing Blender's value in data augmentation for scientific imaging tasks. This approach uniquely combines synthetic data augmentation and HPC resources to address challenges in SEM image segmentation with limited real-world data.
Czech name
—
Czech description
—
Classification
Type
R - Software
CEP classification
—
OECD FORD branch
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Result continuities
Project
—
Continuities
—
Others
Publication year
2024
Confidentiality
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Data specific for result type
Internal product ID
023/13-12-2024_SW
Technical parameters
The software performs segmentation of SEM images, identifying coating layers (e.g., Fe2Al5, FeAl), pores, and chromium precipitates using a U-Net-based deep learning model. The Blender component facilitates synthetic data augmentation via blend files, crucial for enhancing model training. The synthetic data was rendered on Karolina cluster. Large amount of synthetic data can be generated by rendering the blend files using the computational resources on HPC systems.
Economical parameters
Neaplikovatelné
Owner IČO
61989100
Owner name
VŠB - Technická univerzita Ostrava