DetectNano: deep learning detection in TEM images for high-throughput nanostructure characterization.

Area of Science:

• Polymer Science
• Materials Science
• Artificial Intelligence

Background:

• Characterizing self-assembled polymeric vesicles in transmission electron microscopy (TEM) images is crucial but challenging.
• Current methods for analyzing polymer nanostructures are often slow and lack objectivity.

Purpose of the Study:

• To develop an automated, rapid, and unbiased framework for identifying and measuring polymer nanostructures in TEM images.
• To enhance the efficiency and reproducibility of nano-object characterization in polymer self-assembly research.

Main Methods:

• Utilized a deep learning framework based on YOLOv8 for object detection.
• Integrated Weighted Box Fusion to improve detection accuracy.
• Trained the model on a diverse dataset including multiple polymer vesicle morphologies for robust performance.

Main Results:

• Achieved accurate detection and size estimation of polymer nanostructures.
• Demonstrated a detection time of under 2 seconds per image, significantly faster than traditional software.
• Showcased robust performance on unseen TEM images, handling various morphologies.

Conclusions:

• The AI-powered framework enables highly efficient and reproducible nano-object characterization.
• This automation accelerates research in polymer self-assembly.
• Paves the way for general AI-driven automation in materials science.