CH-Net: Deep adversarial autoencoders for semantic segmentation in X-ray images of cabin baggage screening at airports

Abstract

Billions of suitcases and other belongings are checked every year in the X-ray systems in airports around the world. This process is of great importance because it involves the detection of possible dangerous objects such as weapons and explosives. However, the work done by airport surveillance personnel is not free from errors, usually due to tiredness or distractions. This is a security problem that can always be reduced with the help of automatic intelligent tools. This paper attempts to make a contribution to the field of object recognition in X-ray testing for luggage control by proposing a deep learning system that combines a deep convolutional network with an adversarial autoencoder acting as a powerful feature extractor mechanism. The system is developed to separate transmission X-ray images into potentially overlapping regions, separating the X-ray image into organic and inorganic images, taking into consideration the overlapping between the same and different types of materials. To show the superiority of our proposed system, a comparative analysis was carried out including the state-of-the-art deep learning semantic segmentation systems. The proposed method demonstrated highly promising results, achieving the best performance in global accuracy, mean boundary F1 and mean IoU, with a percentage of 80.17%, 76.28% and 76.84%, respectively.