Microstructure refers to a material's internal structure. It records a material's origin and establishes all of its chemical and physical attributes. Although microstructural characterization is common and well-known, microstructural classification is typically carried out manually by human experts, which introduces subjectivity-related uncertainties. Only a few previous studies exist since it is extremely difficult to automatically classify microstructures because they may be a combination of various phases or constituents with complicated substructures. Previous studies concentrated on expertly created and constructed features and separated the classification of microstructures from the feature extraction stage. By simultaneously learning the features from the input and the classification phase, Deep Learning techniques have recently demonstrated outstanding performance in vision applications. In this work, it is suggested to apply a Deep Learning algorithm to classify low carbon steel's microstructure using instances of certain microstructural elements. This work proposes the classification of the steel alloys based on the features of microstructure,which were extracted through an External Attention Transformer network (EATNet). Combining the external attention mechanism with the transformer can provide better performance when compared to conventional Convolutional Neural Network based approach. The proposed classification models were trained using microstructure images of ferritic-martensitic steels containing 9 to 12 wt% Cr, also referred to as 9% Cr steel.

microstructural classification, Deep Learning techniques, External Attention Transformer network, conventional Convolutional Neural Network, ferritic-martensitic steels