Classification of Microstructural Steel Images Using an Attention-Aided Transfer Learning Network

Steels serve as the most widely-used structural metallic materials in industrial practice. Steels have a wide variety of microstructures and mechanical properties. The microstructure of a metal and its physical properties are highly correlated. Manual Microstructure analysis and characterization tasks require deep expertise in this field, and it is a time-consuming process. Contemporary machine vision and machine learning techniques facilitate autonomous microstructure recognition with a high degree of accuracy. To this end, the main objective of this research is to develop a Deep Convolutional Neural Network (DCNN) model based on inductive transfer learning with an attention module for steel microstructural image classification. This attention network, which has been learned in a supervised manner, precisely selects a subset of feature maps with meaningful and discriminative information. In this paper, several regularization schemes have also been investigated with this model. Experimental results show that the proposed model outperforms the baselines by a significant margin.