Research On Cigarette Verification System For Virtual Warehouse Base On Machine Vision

Special-shaped cigarettes are cigarettes that do not meet the standard specifications of China in terms of appearance and size.Due to the pursuit of fashion and personalized products by smokers in recent years,there has been an increase in the types and consumption of special-shaped cigarettes.Currently,tobacco companies rely on fully-automated facilities to sort standard cigarettes,but the unique characteristics of special-shaped cigarettes differ greatly from standard-shaped cigarette,such as specifications,shape and weight,and there are many types of special-shaped cigarette,so it requires manual sorting through a virtual cigarette warehouse.The sorting process involves workers obtaining order information from work-station screens,retrieving the cigarettes from a temporary warehouse,and placing them in the corresponding bin of the virtual cigarette warehouse.However,this process faces two major problems that severely restrict sorting efficiency.As we all know,the variety of special-shaped cigarettes is quickly updated,and some cigarettes have a high similarity in appearance,making it difficult for workers to pick cigarettes by name.What’s more,workers are prone to fatigue under high-intensity labor,resulting in misreading order information and sorting errors,thereby increasing the workload of unpacking and inspection.Therefore,efficient and accurate verification of special-shaped cigarettes has become an urgent problem to be solved for tobacco companies.This paper focuses on the work steps in the virtual cigarette warehouse and proposes a machine vision technology-based cigarette verification system to address these problems.The main research content is as follows.To meet the demands of tobacco logistics enterprises,a vision-based cigarette identification and verification system is designed.This system includes hardware selection for machine vision,specialized support for virtual cigarette warehouses,and software development for verification.Verification of special-shaped cigarettes can be considered as an application of image classification.After the operator places the special-shaped cigarette in the virtual cigarette bin,the software classifies the cigarette and reports the results,allowing the operator to proceed with the verification task.An automatic cigarette image acquisition system is developed,and a deep neural network is explored.This paper studies the cigarette image classification algorithm,which include K-Nearest Neighbor,Support Vector Machine and Random Forest.Additionally,the construction method for cigarette image datasets is discussed to facilitate training of deep neural networks.The paper also studies the processing of cigarette images in a deep neural network,which includes feature extraction,filtering,nonlinear processing,classification,and output.Additionally,attention mechanism(CBAM)and Earth Mover’s Distance(EMD)are studied.We delves into the cigarette recognition and verification algorithm,and the neural network classification method with limited image templates.Various deep neural networks for image classification and detection are explored,including VGG19,Res Net50,and YOLOv5.Optimization methods for Res Net50 are also studied.A Res Net50-based cigarette verification algorithm is proposed,which can address the difficulty of verifying cigarettes in virtual cigarette warehouses using existing verification systems.The verification system needs to be customer-oriented,where both cigarette image collection and model training are completed by the customer,however,due to various reasons,the number of samples provided by them is insufficient to train a good model.Therefore,this paper proposes a cigarette classification network based on few-shot learning.CBAM and EMD focus on the global and local features of the cigarette images separately,and the feature information is summarized and analyzed to recognize the cigarette type.We discusses the design and implementation of cigarette verification software.To address the inconvenience of existing verification software and meet the needs of operators,a verification display software has been designed and developed.The software retrieves the order information and displays the corresponding cigarette image.It also receives the output of the cigarette recognition model and visually displays it,making it easier for workers to search for and verify the cigarette.The verification display software overcomes the shortcomings of incomplete functions in existing cigarette recognition software and has been tested on-site to verify its various functions.