2025-143-16

Training model for nuclear material cladding oxide segmentation

Vedoucí práce: Ing. Petr Čech, Ph.D.

Konzultant: Mgr. Jaroslav Knotek

Podstata a cíl práce

The goal of this project is automating the measurement process of oxidation layers in nuclear materials from microscopy images using deep learning. Currently, this process is performed manually and repeatedly by researchers. The automation involves training a deep learning model on a dataset, which requires preprocessing. Given the dataset's size, augmentation techniques are used to enhance training data size and diversity before model training. Then, the model needs to be integrated into the existing software Fiji-ImageJ in the same format as the previous manually labeled samples.

Náplň činnosti studenta

The student will obtain a dataset containing measurement files. These measurements need to be preprocessed into labels for the given images. Then, the student needs to choose an optimal deep learning model and train it on the dataset. After the training, the results need to be evaluated. The last part will involve integrating this model into Fiji-ImageJ software and adjusting it for optimal user experience.

Doplňující informace

Místo řešení: Ústav informatiky a chemie (143)

Téma je možné zvolit pro následující specializace