Anomaly Detection in Airport based on Generative Adversarial Network for Intelligent Transportation System

Abstract

Deep learning is used in various application, and there are many outstanding performances in many fields recently. Generative Adversarial Networks (GAN) is one of the deep learning models proposed based on the zero-sum game theory and has become a new research hotspot. One of the important issues in intelligent transportation system is airport anomaly detection. However, there is a crical problem that the abnormal data set is relatively insufficient, and even the normal data. The insufficient data set makes difficult to adopt deep learning technique for training a good model. Therefore, this paper adopts the Generative Adversarial Network to obtain data distribution through unsupervised learning and generate more information. In addition, GAN is less limited by the amount of data, so we use GAN-based model for implementation of anomaly detection in airport. The latent vector space of all nomal images is learned by GAN, and abnormal images can be distinguished through particular nerual network structure of GAN. The proposed method has significant performance as seen in the part of experiment.