Model and research of a cigarette box defect recognition system based on deep learning

Abstract

China’s tobacco industry ranks first in the world in terms of output. It can be said that tobacco is one of the pillar industries in China. Cigarette boxes are the main product produced by tobacco companies. In the production process, manually screening defective cigarette boxes is time-consuming and labor-intensive. This system uses YOLOv5 to automatically detect cigarette box defect images and provide defect category warnings. The system mainly includes four parts: the image acquisition unit, image processing unit, processing execution unit, and data acquisition and analysis unit. According to the defect type, the workshop will produce defective cigarette packs of brands such as Fengheshan and Red Diamond, and the defects of each brand will be identified and classified. Recognition accuracy is up to 99.9%. This system achieves the detection of cigarette box surface defects, effectively improving the detection capability of cigarette box defects in the automated production line.