XRDI: A Database of X-Ray Dangerous Items
Yu Wang, Li Ming Yang, Zhuo Qing · 2021
In order to provide a data set and basic detection indicators for the identification and detection of X-ray dangerous items, the study proposed a method to generate X-ray dangerous items data, which is to collect more than ten kinds of dangerous items and over 100 kinds of non-dangerous items. The original data under different photographing directions is processed by segmenting and denoising, random rotating and zooming, and gray-scale image overlay synthesis based on the principle of X-ray imaging. Finally, a data set containing 10,000 positive samples and 100,000 negative samples is generated, which contains three difficulty levels divided by photographing directions. The research uses several existing mainstream target detection algorithms to conduct algorithm experiments on different data sets, gives the mAP and Recall results of different kinds of dangerous items under different difficulty test sets, organizes and compares the results under the conditions of different training sets and test sets, explores the influence of the number of training samples on the results. These results verified the research significance of the data set in the field of dangerous items identification and detection, and laid the foundation for subsequent research.