TD U-Net for Shell Segmentation and Thickness Evaluation in Core-Shell TiO₂ TEM Images

Area of Science:

• Materials Science
• Nanotechnology
• Artificial Intelligence

Background:

• Titanium dioxide (TiO2) is a crucial industrial material, with performance heavily influenced by its core-shell microstructure.
• Current quality control for TiO2 coatings relies on subjective manual inspection, lacking quantitative accuracy.

Purpose of the Study:

• To develop an automated, quantitative method for evaluating the microstructural characteristics of core-shell TiO2 particles.
• To introduce a deep learning model for precise segmentation and shell thickness measurement in TEM images of TiO2.

Main Methods:

• A dataset of 1479 transmission electron microscopy (TEM) images of lab-prepared TiO2 core-shell particles was curated.
• A novel deep learning model, Titanium Dioxide U-Net (TD U-Net), was developed using an encoder-decoder architecture.
• The model underwent a six-step workflow including annotation, mask generation, and augmentation for robust performance.

Main Results:

• TD U-Net achieved a high segmentation accuracy with a Dice coefficient of 0.967.
• The model demonstrated excellent shell thickness measurement, with errors controlled within 5%.
• Performance significantly surpassed existing image-processing models in accuracy and reliability.

Conclusions:

• TD U-Net provides an efficient and reliable automated tool for TiO2 quality assessment.
• The developed intelligent analysis system is applicable for optimizing coating processes and ensuring product quality.
• This deep learning approach addresses limitations of manual inspection in the TiO2 industry.