Semantic segmentation in crystal growth process using fake micrograph machine learning

Area of Science:

• Materials Science
• Artificial Intelligence
• Machine Learning

Background:

• Microscopic evaluation is crucial for materials research, but high-quality images are needed for AI analysis.
• Analyzing microscopic images, especially distinguishing crystal and amorphous states in materials like Ge thin films, presents challenges.

Purpose of the Study:

• To develop a machine learning method for generating "fake micrographs" to enhance AI analysis of low-resolution microscopic images.
• To improve the identification of crystalline and amorphous states in materials where contrast is difficult.

Main Methods:

• Machine learning was used to generate synthetic micrographs mimicking a crystal growth process.
• The generated fake micrographs were used to train AI models.
• The performance of different machine learning models, including ResUNet++ , was evaluated.

Main Results:

• The machine learning model successfully learned from generated fake micrographs.
• The trained model accurately identified low-resolution real micrographs as crystalline or amorphous.
• ResUNet++ achieved over 90% accuracy in image analysis.

Conclusions:

• The developed technology enables automatic and rapid analysis of low-resolution microscopic images.
• Generating fake micrographs is an effective strategy to overcome image quality limitations in materials research.
• This approach has broad applicability in various material science investigations.