Anomaly detection from images in pipes using GAN

Abstract In recent years, the number of pipes that have exceeded their service life has increased. For this reason, earthworm-type robots equipped with cameras been developed to perform regularly inspections sewer pipes. However, inspection methods not yet established. This paper proposes a method for anomaly detection from images in using Generative Adversarial Network (GAN). A model combines f-AnoGAN and Lightweight GAN is used detect anomalies by taking difference between input generated images. Since GANs are only trained non-defective images, they able convert an image containing defects into one without them. Subtraction estimate location anomalies. Experiments were conducted actual cast iron confirm effectiveness proposed method. It was also validated sewer-ml, public dataset.