Detection of Hidden Defects inside Polymer Tubes Using Anomaly Detection with Generative Adversarial Neural Network Based on Terahertz Scanning Images

Abstract

Defects hidden inside a perfluoroalkoxy (PFA) polymer tube were detected using a generative adversarial network (GAN) for terahertz (THz) imaging based on anomaly detection (AnoGAN). The THz signals were analyzed with respect to the size and angle of the defects in the PFA sheets and tubes. The anomaly score distribution was derived based on the degree of deviation. THz images with defects can be classified within a 1% error based on the anomaly score distribution of the THz imaging data. Additionally, the AnoGAN model successfully distinguished the outlier regions of the normal images from the defective images.