AUTHOR=Rao Jianghao , Qin Peng , Zhou Gaofan , Li Meihui , Zhang Jianlin , Bao Qiliang , Peng Zhenming 

TITLE=Combination of deep learning with representation learning in X-ray prohibited item detection

JOURNAL=Frontiers in Physics

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1174220

DOI=10.3389/fphy.2023.1174220

ISSN=2296-424X

ABSTRACT=For X-ray inspection detection, the detector converses the collected X-
rays from the objects into electrical signals, and then transmitted to the
computer for image processing and analysis. From the aspect of digital
image procesing, most of detection tasks focus on data processing and
transformation to get valuable features, which makes the algorithms more
effective.
The consistant requirements for speed and accuracy in X-ray prohibited
items detection are still not fully satisfied, especially these pictures in
special imaging condition. For noisy X-ray images with heavy occlusion,
the direct and suitable approach of representation learning is the optimal
solution.
According to our study, we realized that heterogeneous information
fusion from different extraction approaches can be applied effectively on
this issue. We proposed two innovation algorithms to extract effective
features on X-ray objects which significantly improve the efficiency in
the X-ray prohibited items detection. The brief model we proposed fuses
the representations learned from the noisy X-ray images, and outper-
forms the best model named DOAM-O so far on OPIXray. Furthermore,
the attention module we designed to select information of deep learn-
ing and representation strengthen the model, considering this, the model
still costs shorter time in the processes of both training and inference,
which makes that easier to be trained on a lighter computing device.