Application of Deep Learning Workflow for Autonomous Grain Size Analysis.
Alexandre Bordas1, Jingchao Zhang2, Juan C Nino1
1Department of Materials Science and Engineering, University of Florida, Gainesville, FL 32612, USA.
Molecules (Basel, Switzerland)
|August 12, 2022
Summary
This study introduces an automated workflow for measuring grain size in microscopy images using computer vision (CV). The new method significantly reduces processing time and maintains accuracy compared to traditional manual techniques.
Area of Science:
- Materials Science
- Computer Vision
- Image Analysis
Background:
- Traditional grain size determination relies on manual analysis of microscopy images, which is time-consuming and requires expertise.
- Computer vision (CV) offers automated solutions for microstructural characterization, simplifying processes like grain size measurement.
- Existing CV methods show promise but often require further refinement for accuracy and efficiency.
Purpose of the Study:
- To develop an end-to-end, automated workflow for accurate grain size determination from microscopy images.
- To compare the performance of a pre-trained holistically nested edge detection (HED) model against traditional methods and Canny edge detection.
- To evaluate the impact of morphological operations on the accuracy of CV-based grain size measurements.
Main Methods:
- An automated workflow using a pre-trained holistically nested edge detection (HED) model for edge detection in microscopy images.
- Comparison of HED model performance against the Canny edge detection method and traditional ASTM standards (Heyn's and Saltykov's methods).
- Post-processing using open-source image processing packages to extract grain size data and analysis of morphological operation effects.
Main Results:
- The HED model significantly outperforms the Canny edge detection method in accuracy for optical microscope images.
- The automated workflow reduces image processing time by one to two orders of magnitude compared to traditional manual methods.
- The proposed convolutional neural network (CNN)-based workflow achieves accuracy comparable to manual methods while drastically cutting down processing time.
Conclusions:
- The developed end-to-end CNN-based workflow provides an efficient and accurate automated solution for grain size determination.
- This approach minimizes manual intervention, making microstructural characterization faster and more accessible.
- The HED model integrated into this workflow represents a significant advancement over conventional edge detection techniques for this application.
