Research on multi-source microstructure image recognition of foam ceramics using convolutional network combine with...

Area of Science:

• Materials Science
• Artificial Intelligence
• Image Analysis

Background:

• Foam ceramics possess valuable industrial properties like high porosity and temperature resistance.
• Their complex microstructures pose challenges for traditional image analysis and machine learning methods.
• Existing methods struggle to capture both global and local microstructural features effectively.

Purpose of the Study:

• To develop a novel artificial intelligence recognition model for foam ceramic microstructure image analysis.
• To enhance the accuracy and detail of microstructural feature identification in foam ceramics.
• To address limitations of traditional machine learning in capturing complex feature dependencies.

Main Methods:

• Proposed a novel artificial intelligence recognition model, FD-Conv, integrating Transformers for global information and Convolutional Neural Networks (CNNs) for local features.
• Incorporated a frequency domain block detail enhancement mechanism to improve recognition.
• Utilized multi-source microscopic image data for training and validation.

Main Results:

• The FD-Conv model achieved a minimum accuracy improvement of 7.6% over state-of-the-art methods.
• Successfully identified foam ceramics with diverse compositions and formulations.
• Quantified microstructural phase characteristics with improved precision.

Conclusions:

• The FD-Conv model offers a significant advancement in foam ceramic microstructure image analysis.
• This artificial intelligence approach enhances recognition accuracy, particularly for multi-source microscopic image feature learning.
• The model's ability to analyze diverse ceramic types and quantify microstructural phases opens new avenues for material characterization.