Original scientific paper

https://doi.org/10.17559/TV-20220222034206

Intelligent Detection of Dangerous Goods in Security Inspection Based on Cascade Cross Stage YOLOv3 Model

Jianjun Wu orcid.org/0000-0002-7549-7360 ; College of Information Technology and Communication, Hexi University, No. 846, Huancheng North Road, Zhangye, Gansu Province, China
Shaowen Liao ; College of Information Technology and Communication, Hexi University, No. 846, Huancheng North Road, Zhangye, Gansu Province, China

Abstract

At present, it mainly depends on the human eye to identify the X-ray scanning image, when the security detector is used to detect the dangerous goods in the baggage. It is labor intensive and prone to false or missed detection. This paper proposes an intelligent detection method of dangerous goods in security inspection based on a novel cascaded cross-stage YOLOv3 model (abbreviated to CCS-YOLOv3). Considering the different sizes, disorderly lay or serious overlap of various objects in the scanning image, this method first enhances the scanned image to improve the quality of the data set. After that, the traditional YOLOv3 is improved by cascading cross-stage mode, and the backbone network of YOLOv3 is improved to cascade cross-stage Darknet. And then the backbone network is followed by a spatial pyramid pooling (SPP) module. Following that, the feature pyramid network (FPN) is connected in series with a bottom-up feature pyramid structure to realize the feature fusion. The results of model Ablation experiment and baggage scanning image detection show that the cascade cross-stage YOLOv3 model significantly improves the image detection speed and precision, and the model is effective and feasible.

Keywords

cascade cross stage networks; detection of dangerous goods; feature fusion; intelligent security inspection; YOLOv3 model

Hrčak ID:

275305

URI

https://hrcak.srce.hr/275305

Publication date:

19.4.2022.

Visits: 1.139 *