Sreerupa Das, Lockheed Martin; Chris Holland, Lockheed Martin
We’ll discuss convolutional neural networks (CNNs), which have recently become the preferred tool for many visual detection tasks including object classification, localization, detection, and segmentation. CNNs are specialized neural networks composed of many layers and specifically designed to analyze grid-like data like images. One of their key features is the ability to automatically detect important features within an image such as edges, patterns, and shapes. Previously, these features had to be manually engineered by subject matter experts. Inspired by the significant achievements CNNs have experienced in computer vision, we’ll examine U-Net, a specific CNN architecture, which is suited for visual defect detection. We’ll identify situations for the use of this architecture for external defect detection on aircrafts, and experimentally discuss its performance across a dataset of common visual defects.