Citation: Wang, C.; Zhao, J.; Yu, Z.;
Xie, S.; Ji, X.; Wan, Z. Real-Time
Foreign Object and Production Status
Detection of Tobacco Cabinets Based
on Deep Learning. Appl. Sci. 2022, 12,
10347. https://doi.org/10.3390/
app122010347
Academic Editor: Enrico Vezzetti
Received: 14 July 2022
Accepted: 9 October 2022
Published: 14 October 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Real-Time Foreign Object and Production Status Detection of
Tobacco Cabinets Based on Deep Learning
Chengyuan Wang, Junli Zhao
∗
, Zengchen Yu, Shuxuan Xie, Xiaofei Ji and Zhibo Wan
College of Computer Science & Technology, Qingdao University, Qingdao 266071, China
* Correspondence: zhaojl@yeah.net; Tel.: +86-532-85953151
Abstract:
Visual inspection plays an important role in industrial production and can detect product
defects at the production stage to avoid major economic losses. Most factories mainly rely on manual
inspection, resulting in low inspection efficiency, high costs, and potential safety hazards. A real-time
production status and foreign object detection framework for smoke cabinets based on deep learning
is proposed in this paper. Firstly, the tobacco cabinet is tested for foreign objects based on the YOLOX,
and if there is a foreign object, all production activities will be immediately stopped to avoid safety
and quality problems. Secondly, the production status of tobacco cabinet is judged to determine
whether it is in the feeding state by the YOLOX position locating method and canny threshold
method. If it is not in the feeding state, then the three states of empty, full, and material status of
the tobacco cabinet conveyor belt are judged based on the ResNet-18 image classification network.
Ultilizing our proposed method, the accuracy of foreign object detection, feeding state detection
and the conveyor belt of tobacco cabinet state detection are 99.13%, 96.36% and 95.30%, respectively.
The overall detection time was less than 1 s. The experimental results show the effectiveness of our
method. It has important practical significance for the safety, well-being and efficient production of
cigarette factories.
Keywords:
deep learning; foreign object detection; production status detection; tobacco cabinet;
cigarette factory
1. Introduction
In the tobacco production process, it is often necessary to confirm the status of the
tobacco cabinet in real time. Tobacco factory workers need to climb up the tobacco cabinet
for a long time and inhale a large amount of tobacco and dust particles every day, which
seriously endangers their health. The aisles of tobacco cabinets are often multi-layered
and narrow, which also creates huge safety hazards. After the production is completed,
workers need to clean the interior of each tobacco cabinet. If the cleaning tool is forgotten
in the tobacco cabinet, it will affect the quality of the next batch of cut tobacco, and may
even cause the batch of tobacco to be scrapped. In order to reduce the cost of workers and
eliminate the huge potential dangers of climbing the tobacco cabinet, the authors propose
an automatic detection method that can detect the safety status and production status of
tobacco cabinets in real time.
At present, most factories use visual inspection performed manually by workers,
which often requires huge labor costs. Some factories identify and detect tobacco cabinets
based on traditional machine learning methods, which have disadvantages such as low
accuracy and long time consumption. In this paper, considering the powerful feature
extraction ability of deep learning, the authors propose to realize real-time monitoring of
tobacco cabinets based on the deep learning method. Compared with visual monitoring
of workers and traditional machine learning methods, the accuracy and precision will be
greatly improved.
Appl. Sci. 2022, 12, 10347. https://doi.org/10.3390/app122010347 https://www.mdpi.com/journal/applsci
