Design and Implementation of Cigarette Stamp Authenticity Identification System Based on Machine Vision

To swiftly and accurately identify the authenticity of various cigarette types and advance automation in the tobacco industry, a comprehensive cigarette adhesive mark authenticity identification system was designed. This system integrates image acquisition, preprocessing, feature extraction, and authenticity identification. Utilizing hardware platforms such as Raspberry Pi, industrial cameras, and stepper motors, along with custom-developed software for cigarette mark image acquisition, the system automatically captures batch images of cigarette marks to construct a robust image dataset. The transfer learning method was employed to train the authenticity identification model for cigarette rubber marks using the MobileNetV3 network. The model's effectiveness was then validated with independently collected authenticity samples. Subsequently, the trained model was deployed in an embedded system, enabling the complete automation of the cigarette authenticity identification process. The results demonstrated that the MobileNetV3-based cigarette adhesive mark authentication model achieved an accuracy of 94.70%, showcasing strong authentication capabilities with high efficiency and precision. This system provides reliable cigarette authenticity identification services, enhancing the automation and intelligence of the tobacco industry, and significantly supporting the technological advancement of the sector.