Microstructural characterization allows knowing the components of a microstructure in order to determine the influence on mechanical properties, such as the maximum load that a body can support before breaking out. In almost all real solutions, microstructures are characterized by human experts, and its automatic identification is still a challenge. In fact, a microstructure typically is a combination of different constituents, also called phases, which produce complex substructures that store information related to origin and formation mode of a material defining all its physical and chemical properties. Convolutional neural networks (CNNs) are a category of deep artificial neural networks that show great success in computer vision applications, such as image and video recognition. In this work we explore and compare four outstanding CNNs architectures with increasing depth to analyze their capability of classifying correctly microstructural images into seven classes. Experiments are done referring to ultrahigh carbon steel microstructural images. As the main result, this paper provides a point-of-view to choose CNN architectures for microstructural image identification considering accuracy, training time, and the number of multiply and accumulate operations performed by convolutional layers. The comparison demonstrates that the addition of two convolutional layers in the LeNet network leads to a higher accuracy without considerably lengthening the training.

Automatic Microstructural Classification with Convolutional Neural Network

Stefania Perri;Pasquale Corsonello;
2019

Abstract

Microstructural characterization allows knowing the components of a microstructure in order to determine the influence on mechanical properties, such as the maximum load that a body can support before breaking out. In almost all real solutions, microstructures are characterized by human experts, and its automatic identification is still a challenge. In fact, a microstructure typically is a combination of different constituents, also called phases, which produce complex substructures that store information related to origin and formation mode of a material defining all its physical and chemical properties. Convolutional neural networks (CNNs) are a category of deep artificial neural networks that show great success in computer vision applications, such as image and video recognition. In this work we explore and compare four outstanding CNNs architectures with increasing depth to analyze their capability of classifying correctly microstructural images into seven classes. Experiments are done referring to ultrahigh carbon steel microstructural images. As the main result, this paper provides a point-of-view to choose CNN architectures for microstructural image identification considering accuracy, training time, and the number of multiply and accumulate operations performed by convolutional layers. The comparison demonstrates that the addition of two convolutional layers in the LeNet network leads to a higher accuracy without considerably lengthening the training.
978-3-030-02827-5
978-3-030-02828-2
CNN; Image processing; Microstructure characterization
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/20.500.11770/299394
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? 3
social impact