Skip to the main content

Transactions of FAMENA, Vol. 48 No. 4, 2024.

Original scientific paper

https://doi.org/10.21278/TOF.484060123

Quantitative Automated Detection of Voids, Pores, Cracks, and Fibre Orientation in Scanning Electron Microscopy Images Utilising Mask Convolutional Neural Networks (M-CNN) for Natural Fibre Composite Characterisation

Ganesan Elizabeth Rani orcid id orcid.org/0000-0002-4513-2109 ; Department of Artificial Intelligence and Data Science, KIT-Kalaignarkarunanidhi Institute of Technology, Coimbatore, Tamil Nadu, India *
Marimuthu Sakthimohan ; Department of Electronics and Communication Engineering , KIT-Kalaignarkarunanidhi Institute of Technology, Coimbatore, Tamil Nadu, India
Arockiasamy Felix Sahayaraj ; Department of Mechanical Engineering, KIT-Kalaignarkarunanidhi Institute of Technology, Coimbatore, Tamil Nadu, India
Mariappan Sornalakshmi ; Department of Computer Science PG, Arulmigu Kalasalingam College of Arts and Science, Krishnan Kovil, Tamil Nadu, India

* Corresponding author.

Full text: english pdf 5.699 Kb

page 45-64

downloads: 35

cite

Abstract

In recent years, image processing in various scientific domains has gained prominence, particularly in materials science. This study leverages advanced deep learning techniques, specifically the Mask Convolutional Neural Network (M-CNN), for quantitative analysis of Scanning Electron Microscopy (SEM) images. Our approach involves the rigorous classification of SEM images in RGB format, utilising M-CNN to intelligently scrutinise edges and quantify cracks, pores/voids, and fibre orientation. Notably, M-CNN plays a pivotal role in accurately predicting material strength by categorising angles into ductile (45°) and brittle (90°) ones. The proposed software achieves a remarkable 99% accuracy in detecting and quantifying structural elements within SEM images, marking a significant advancement in materials science and demonstrating the potential of advanced image processing techniques for material analysis.

Keywords

Deep Learning; Scanning Electron microscopy; Mask Convolutional Neural Networks (M- CNN)

Hrčak ID:

321642

URI

https://hrcak.srce.hr/321642

Publication date:

20.10.2024.

Visits: 194 *