The inner structure of a material is called microstructure. It stores thegenesis of a material and determines all its physical and chemical properties.While microstructural characterization is widely spread and well known, themicrostructural classification is mostly done manually by human experts, whichopens doors for huge uncertainties. Since the microstructure could be acombination of different phases with complex substructures its automaticclassification is very challenging and just a little work in this field hasbeen carried out. Prior related works apply mostly designed and engineeredfeatures by experts and classify microstructure separately from featureextraction step. Recently Deep Learning methods have shown surprisingly goodperformance in vision applications by learning the features from data togetherwith the classification step. In this work, we propose a deep learning methodfor microstructure classification in the examples of certain microstructuralconstituents of low carbon steel. This novel method employs pixel-wisesegmentation via Fully Convolutional Neural Networks (FCNN) accompanied bymax-voting scheme. Our system achieves 93.94% classification accuracy,drastically outperforming the state-of-the-art method of 48.89% accuracy,indicating the effectiveness of pixel-wise approaches. Beyond the successpresented in this paper, this line of research offers a more robust and firstof all objective way for the difficult task of steel quality appreciation.
展开▼
