Quality Control of Cigarettes Packaging using Convolutional Neural Network

1st International Conference on Advanced Engineering and Technology

工业技术（总论）

Nazar, A.^1 ; Nurwiyadi, M.N.P.^1 ; Syai'In, M.^1 ; Khumaidi, A.^1 ; Adhitya, R.Y.^1 ; Rinanto, N.^1 ; Hasin, M.K.^1 ; Widodo, H.A.^1

Automation Engineering Study Program, Department of Marine Electrical Engineering, Shipbuilding Institute of Polytechnic Surabaya, Indonesia^1

关键词: Cigarette packaging;    Convolution neural network;    Convolutional neural network;    Defective products;    Essential point;    Packaging system;    Sorting system;    Tobacco industry;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/462/1/012002/pdf
DOI  :  10.1088/1757-899X/462/1/012002

学科分类：工业工程学

来源: IOP

【 摘 要 】

The tobacco industry is a source of incomes for our country. Currently, cigarette consumers in Indonesia are already very high. Quality is considered as an essential point for every industry to satisfy its customers. However, some problems often occur in the Industry. Including the decline in cigarettes quality. One of the reasons for the decline in the quality of cigarettes is the lack of selection of the sorting system on cigarette packaging. The lack of quality of the packaging system makes the quality of cigarettes decreased. In a tobacco company in East Java, the fetching of defective products is still done manually. Based on the problem, the author takes that study case as the background of this research. In this system, the packs of cigarettes will be detected and selected using image processing. On the image processing will be implemented as a method of convolution neural network (CNN) this algorithm has a function to detect texture from cigarette packaging. If there is non-conforming or defective cigarette packaging, the rejector will be active and discard the cigarette packaging and it will not enter into the next packing process. The result of training data from this study had a success rate of 96.06%. When using real-time data, the system is able to classify 148 data from 154 existing data correctly. The real-time system success rate is 94,59%.