Detect occluded items in X-ray baggage inspection.

X-ray baggage inspection automatically determines whether there are prohibited items in passenger luggage and has recently achieved good progress due to the development of deep convolutional neural networks (DCNNs). However, the performance of X-ray baggage inspection degrades with heavily occluded and cluttered baggage owing to the interaction of the imaged objects. Inspired by the idea that discriminant cues may still exist in multiscale perceptions, discrepancies among different classes, and associations among different tasks, we propose a novel approach that exploits pseudo semantic masks to enhance the discriminant ability of feature representations. Moreover, the features at neighboring scales interact to further explore context information. We verify our method on the publicly available SIXray, OPIXray, and HIXray datasets. The results show that our method outperforms other state-of-the-art X-ray baggage inspection methods by 1.02% and 0.85% in terms of mean average precision (mAP). [ABSTRACT FROM AUTHOR]