2019 Pittsburgh AISTech

Sessions Speakers Exhibitors Products Press Releases Show Specials Floor Plan Buzz

Inclusion Classification by Computer Vision and Machine Learning (Room 411)

This work applied computer vision and machine learning methods to characterize non-metallic inclusions in steel by only backscattered electron scanning electron microscope (BSE-SEM) images collected from automated inclusion analysis. In this work, convolutional neural networks (CNNs) were employed to classify observations as inclusions or not inclusions (pores or surface contamination) and furthermore to classify inclusions by composition category (e.g. alumina, spinel, calcium sulfide, etc.). The data for CNN training and testing were collected from industrial samples. The overall objective is to create methods that will lead to identification and characterization (size, shape, chemistry) of inclusions via BSE-SEM images.