Detection of Prohibited Articles based on Lightweight Convolution Neural Network. Issue 1 (August 2021)

• Record Type: Journal Article
• Authors: Lin, Xintao; Zhang, Yanxi; Zhang, Nanfeng; Gao, Xiangdong; Ruan, Jieshan
• Is Part Of: Journal of physics. Volume 1986:Issue 1(2021)
• Publication Date: 2021-08
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1986/1/012051
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Prohibited articles detection is an important way to ensure public safety. In order to improve the flexibility of security inspections in complex public places, this paper uses three different convolutional neural networks, YOLOv3, MobileNet-SSD, and YOLOv3-Tiny, to detect the prohibited articles on the security inspection image data. Two performance indicators, mean Average Precision and Frames Per Second, were used to test the effect of the propose model. Finally, this trained model was deployed to the Raspberry Pi mobile device. Experimental results show that using YOLOv3-Tiny model for prohibited articles detection can improve the detection speed by about 10 times and the high detection precision is also maintained. This experiment provides a research idea for the fast detection of prohibited articles on mobile devices.