Related Experiment Video
Updated: Jun 15, 2025
09:32
Electrospun Nanofiber Scaffolds with Gradations in Fiber Organization
Published on: April 19, 2015
9.8K
Geometric feature extraction in nanofiber membrane image based on convolution neural network for surface roughness
Dong Hee Kang1,2, Na Kyong Kim3, Wonoh Lee1
1Department of Mechanical Engineering, Chonnam National University, 77 Youngbong-ro, Buk-Gu, Gwangju, 61186, Republic of Korea.
Heliyon
|August 22, 2024
Summary
This study introduces a novel artificial intelligence method using convolution neural networks (CNNs) to analyze membrane surface roughness from images. The approach accurately extracts geometric features, offering a more comprehensive surface characterization than traditional techniques.
Area of Science:
- Materials Science
- Artificial Intelligence
- Image Analysis
Background:
- Traditional surface roughness measurement methods (e.g., atomic force microscopy, optical profiler) have limitations in analyzing porous membranes due to restricted analysis areas and depth resolution.
- Scanning electron microscopy (SEM) images, when combined with advanced feature extraction, offer potential for broader surface roughness analysis across various resolutions.
Purpose of the Study:
- To develop and validate an artificial intelligence-based method for extracting average surface roughness from membrane images.
- To leverage convolution neural network (CNN) models for analyzing geometric characteristics and improving surface roughness measurement accuracy.
Main Methods:
- Image preprocessing techniques were applied to enhance geometric patterns by amplifying pixel intensity disparities.
- A feature map-based CNN model was employed for feature extraction from membrane images.
- Statistical analysis and magnitude spectrum were used to classify surface roughness and predict logarithmic average surface roughness values.
Main Results:
- The CNN model achieved a Mean Absolute Percentage Error (MAPE) of 4.80% on the test dataset for predicting logarithmic average surface roughness.
- Geometric patterns and magnitude spectrum analysis confirmed the classification of surface roughness.
- The method demonstrated effective indirect surface measurement by analyzing geometric patterns.
Conclusions:
- The proposed CNN-based feature extraction combined with statistical analysis provides a valuable indirect method for measuring surface roughness.
- This approach effectively reveals hidden physical characteristics in surface geometries from irregular pixel patterns.
- The technique reduces randomness error in structural characteristic analysis and offers a more comprehensive understanding of surface topography.
