A Deep Learning-based Dangerous Goods Detection And Tracking Algorithm From X-ray Images

As a contactless security inspection technology,X-ray security inspection equipment is widely used in the detection of dangerous goods in all kinds of densely populated public places to ensure the safety.However,X-ray security inspections rely heavily on huge labor efforts.Thus,manual inspections are inefficient.The automatic identification and tracking of dangerous goods in X-ray images has gradually become one of hot research topics in this field.The recent works have conducted extensive and in-depth research work in the field of X-ray dangerous goods identification.Specifically,how to improve the accuracy and efficiency of detection of dangerous goods is one of the key research questions.Existing works normally use features extracted for the detection of dangerous goods,and many tests are performed to determine whether there are dangerous goods in testing images.However,the exploration of these tasks with the cooperation of X-ray image detection and real-time tracking needs to be strengthened.In order to enable the automatic detection and tracking,this thesis proposes a detection and tracking method for an X-ray image dataset based on deep learning.The main research contents are as follows:Firstly,through the analysis of the process of the widely used X-ray security inspection equipment,the requirements of the security inspection equipment for the detection of dangerous goods in the positioning accuracy,classification accuracy,and real-time performance are collected.Meanwhile,we conduct the data collection and preprocessing to generate a real-world X-ray dangerous goods image dataset.Secondly,we propose a deep learning-based X-ray image dangerous goods detection and tracking algorithm from the X-ray image dataset.The proposed methods has two major steps.On the one hand,the deep learning detection network based on the Improved Signal Shot Multibox Detector(SSD)method is designed to improve the detection accuracy.On the other hand,the dangerous goods' tracker based on the detection results is implemented and tracked.As a result,managers and detectors could work together to achieve the nearly real-time detection tracking.Thirdly,the proposed method is compared with the other popular detection methods(such as Faster-RCNN,YOLO-v2,SSD300,SSD512,etc.),based on two evaluate metrics(average accuracy and speed).Experimental results show that the proposed detection and tracking algorithm can achieve the real-time detection and tracking while ensuring the high accuracy.In another word,the proposed algorithm outperforms baseline algorithms.The X-ray dangerous goods detection and tracking algorithm proposed in this paper has theoretical and practical value for auxiliary security inspection.It could assist security personnel to identify those common dangerous goods while improving the security inspection efficiency.