Automating Interpretation of Images and Visual Inspections in Modern Manufacturing and Medical Settings

Abstract

Automating visual inspection is rapidly finding its way into non-invasive industrial and medical processes. In this work, we address detection automation opportunities in the task of interpreting both medical and industrial images, we combine domain knowledge with state-of-the-art, domain-independent methods such as Convolutional Neural Networks (CNN) to address them. Specifically, we address two problems in additive manufacturing and one in medical imaging as described below. We provide a method to improve the resolution of and automate the extraction of local and global porosity features from XCT images of additively manufactured Ti-6Al-4V. We also address the automation issues in characterizing local roughness features, particularly the curvature of microscopic surface pits observed in microscopy images of AM Ti-6Al-4V. Finally, we developed a deep learning model to automate the task of detection and segmentation of different muscles in images obtained from MRI of the lower back.