Enhanced Deep Electric Pole Anomaly Detection Using Generative Adversarial Network-based Data Augmentation

Dongkun Lee, Jonghwan Hyeon, Ho-Jin Choi

Abstract: This paper reports the effect of a novel artificial neural network architecture for industrial anomaly detection using generative adversarial network (GAN)-based data augmen-tation. We show that GAN-based data augmentation enhances the performance of end-to-end electric pole anomaly detection. With the convolutional neural network (CNN) hyperparameter search, our method outperforms vanilla CNN and Cutout augmentation by an average of 2.2 % p and 1.6 % p, respectively and has an accuracy of over 88 % for the test dataset.