An Accurate Deep Learning Threat Image Detection Algorithm for X-Ray Baggage Dataset
Saif Sarmad Al-Shamari, Hilal Abbood Al-Libawy · 2022
Recently, the use of x-ray imaging has become widespread in public places like stadiums and airports to reduce the risk of terrorist attacks. The importance of baggage inspection using X-ray imaging comes from its ability to detect any possible threat objects like (explosives and guns) by classifying the images of the objects into positive or negative. Different approaches have been followed to improve the accuracy level of threat detection systems, especially in deep learning areas such as the convolution neural networks (CNNs) approach. Researchers obtained overall good detection performance results using publicly available datasets. However, accuracy improvement remains a vital concern in the security sector. In this paper, a deep learning algorithm has been proposed to enhance the performance of threat image projection systems. The core of this algorithm is based on the deep learning network (DenseNet), as well as the test time augmentation method trained on the GDXray dataset. As far as is known the authors, this is the first time that a DenseNet121 has been used to classify GDXray dataset. Moreover, the logistic regression model was added as a final stage in order to get the final prediction. The obtained results indicate the ability of the proposed algorithm to detect threat images with an accuracy of (97.68%), which outperforms the existing approaches using the same dataset.