Automatic Detection of Threat Objects in X-ray Baggage Inspection Using Mask Region-Based Convolutional Neural Network and Deformable Convolutional Network
Jongchol Kim, Jinmyong Ri, Hyok Jo · Russian Journal of Nondestructive Testing · 2022
Abstract In X-ray baggage inspection, due to the increase in the types of dangerous goods and the composition of the baggage contents, it is not suitable to clearly display various baggage and dangerous goods on the image acquired by the X-ray inspection system. In recent years, many deep learning-based object detection and recognition technologies have appeared, and object detection algorithms are also developing at a rapid pace along with the development of deep learning technologies. Object detection models commonly use convolution networks to process input values, but the convolution network has a problem of using a common filter regardless of the size of the object. This study aims to improve the accuracy of the threat objects detection algorithm in X-ray baggage images by applying deformable convolutional networks that make filters corresponding to objects using offsets. The proposed method was applied to Mask R-CNN(mask region based convolutional neural network) using GDXray and SIXray, which are X-ray security inspection image data sets, and the experimental results show that the proposed model performance can be improved compared to the existing algorithms.