Chapter 6
Deep Learning and RFID System
Physical Anti-Collision
Radio Frequency Identification (RFID) technology has significant traits such as
batch reading, non-line-of-sight communication, and teletransmission, which is a
non-contact identification technology automatically. As its peculiarities, RFID has a
steady increase in warehouse inventory, electronic payment, object tracking, target
detection, smart city, automatic driving, and so on [1–3]. What is more, the major
superiority of RFID is the simultaneous identification and location of multi-target.
In the field of RFID, the reading performance of tags is an important performance
indicator for measuring tag. Related studies have shown that the tags’ geometrical
distribution has an important influence on the tags’ reading performance. But when
multiple tags are in the reader’s recognition range and the reader responds to them
at the same time, the reader will not recognize the tag correctly. This phenomenon
is called tag collision. At present, we mainly solve the collision problem of multiple
tags in the same channel through some algorithms such as the ALOHA algorithm and
binary tree algorithm. However, these algorithms solve the conflicts in the communication protocol layer, which cannot deal with the impact of external interference
on air interface communication.
In order to further analyze the influence of tag distribution on the overall performance of the tags, we designed the RFID tag distribution optimization system based
on machine vision inspection. Firstly, we use CCD cameras to capture the images of
the tags. Secondly, due to the degradation in the acquired images, the deep learning
method is used to restore the degraded images during image acquisition. Thirdly,
we use an image matching algorithm to obtain the location information of the RFID
tag network. Fourthly, by using the RFID dynamic detection system, the reading
distance of the tag network is measured. Finally, aiming at the nonlinear relationship
between tag network distribution and corresponding reading distance, deep learning
is used to model the nonlinear relationship to optimize the tag distribution.
In 2006, Hinton and his collaborators put forward the concept of deep learning
[4, 5]. Deep learning methods including CNN, DBN, AE, RNN, etc., have been
widely used in many fields [6–8]. At the same time, deep convolution neural network
(CNN) has drawn increasing attention in machine learning [9, 10]. Due to the
© Science Press 2021
X. Yu et al., Physical Anti-Collision in RFID Systems,
https://doi.org/10.1007/978-981-16-0835-3_6
201
Deep Learning and RFID System
Physical Anti-Collision
Radio Frequency Identification (RFID) technology has significant traits such as
batch reading, non-line-of-sight communication, and teletransmission, which is a
non-contact identification technology automatically. As its peculiarities, RFID has a
steady increase in warehouse inventory, electronic payment, object tracking, target
detection, smart city, automatic driving, and so on [1–3]. What is more, the major
superiority of RFID is the simultaneous identification and location of multi-target.
In the field of RFID, the reading performance of tags is an important performance
indicator for measuring tag. Related studies have shown that the tags’ geometrical
distribution has an important influence on the tags’ reading performance. But when
multiple tags are in the reader’s recognition range and the reader responds to them
at the same time, the reader will not recognize the tag correctly. This phenomenon
is called tag collision. At present, we mainly solve the collision problem of multiple
tags in the same channel through some algorithms such as the ALOHA algorithm and
binary tree algorithm. However, these algorithms solve the conflicts in the communication protocol layer, which cannot deal with the impact of external interference
on air interface communication.
In order to further analyze the influence of tag distribution on the overall performance of the tags, we designed the RFID tag distribution optimization system based
on machine vision inspection. Firstly, we use CCD cameras to capture the images of
the tags. Secondly, due to the degradation in the acquired images, the deep learning
method is used to restore the degraded images during image acquisition. Thirdly,
we use an image matching algorithm to obtain the location information of the RFID
tag network. Fourthly, by using the RFID dynamic detection system, the reading
distance of the tag network is measured. Finally, aiming at the nonlinear relationship
between tag network distribution and corresponding reading distance, deep learning
is used to model the nonlinear relationship to optimize the tag distribution.
In 2006, Hinton and his collaborators put forward the concept of deep learning
[4, 5]. Deep learning methods including CNN, DBN, AE, RNN, etc., have been
widely used in many fields [6–8]. At the same time, deep convolution neural network
(CNN) has drawn increasing attention in machine learning [9, 10]. Due to the
© Science Press 2021
X. Yu et al., Physical Anti-Collision in RFID Systems,
https://doi.org/10.1007/978-981-16-0835-3_6
201