Instrumentation & Measurement Magazine 25-2 - 33

Fig. 4. Prospects and challenges.
(Fig. 4). Solving the problems in smart manufacturing by combining
state-of-the-art technologies, such as 5G and cloud
computing, is the way forward. The specific challenges and
trends can be grouped into the following three areas:
◗ The intelligence of robots needs to be improved due to the
expansion of the scale of the production line. In the future,
DL and robot vision will be combined with multiple
sensors to significantly enhance the perception of robots.
As the number of sensors equipped with robots increases,
efficient data transmission also becomes important. Thus,
5G communication technology has significant implications
for expanding the smart manufacturing line. Its
introduction can solve the problem of data transmission
in each line, implement the sharing of visual data
of intelligent robots, realize the cooperative operation of
multiple robots by using intelligent planning technology,
and help build a large-scale smart manufacturing production
line.
◗ The real-time performance of smart manufacturing needs
to be further improved. Although advanced models such
as Faster RCNN and CenterNet have achieved high accuracy
of detection, they are difficult to apply to mobile
hardware platforms that have limited computing power,
such as ARM, FPGA, and ASIC. Some electronic systems
reduce computation and improve real-time performance
by optimizing deep learning models. For example, the
authors of [14] uses some model compression techniques
to improve the detection speed of the smart robots for
pharmaceutical production. In addition, some electronic
systems use computation offloading to improve computing
efficiency. Li designs an inspection system based
on fog computing, which can offload the computation
burden to the fog nodes, significantly decrease the overload
of the central servers and improve the calculation
efficiency [39]. However, the speed of data transmission
still limits the real-time performance of computation
offloading. In recent years, the development of big data
April 2022
and 5G has made cloud computing applicable to industrial
production. The introduction of these advanced
technologies can help overcome the limitation of the
computational capacity of hardware chips, improve the
real-time performance of the electronic system, and
promote the integration of robotics and AI technology.
◗ Data acquisition is an important issue for DL. For
complex neural networks, it is often necessary to collect
a large dataset to solve the problem of overfitting, especially
when training the robot for grasping tasks. The
development of a big data platform can solve the problem
of insufficient data. By uploading images collected by
vision sensors, such a system can achieve data interoperability
across geographical areas. In addition, the real-time
acquisition of the working status of operations also helps
predict and adjust the production line.
Summary
Labor shortage and low productivity pose daunting challenges
to the development of manufacturing in China.
DL-based robot vision is an efficient tool to solve these challenges.
In this paper, we introduced the industrial demand for
robot vision and the advantages of DL and then reviewed applications
of DL and robot vision in industrial production lines
from four perspectives, with a focus on contributions made by
Chinese researchers and companies. Finally, we analyzed the
trends in current research of combining DL and robot vision
with high-end technologies.
Acknowledgment
This work was supported in part by: the National Key RD Program
of China under Grant 2018YFB1308200, the National
Natural Science Foundation of China under Grants 61971071,
6202780012, the Hunan Science Fund for Distinguished Young
Scholars under Grant 2021JJ10025, the Changsha Science and
Technology Major Project under Grant kh2003026, the Joint
Open Foundation of state Key Laboratory of Robotics under
IEEE Instrumentation & Measurement Magazine
33
Instrumentation & Measurement Magazine 25-2

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-2
