Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2024.0429000
Urban Interior Boundaries Delimitation
Respecting to Human Mobility
YU WEI1, XI ZHAO2
1College of Computer Science and Technology (College of Data Science), Taiyuan University of Technology, Taiyuan, 030024, China (e-mail:
weiyu123112@163.com)
2School of Management, Xi’an Jiaotong University, The Key Lab of the Ministry of Education for Process Control & Efficiency Engineering, Xi’an, 710049,
China (e-mail: 1029198725@qq.com)
Corresponding author: Xi Zhao (e-mail: 1029198725@qq.com).
This work was supported in part by the Shanxi Province Basic Research Program under Grant 202203021222103.
ABSTRACT Administrative divisions are regional divisions of the state for the purpose of hierarchical
administration. In recent years, the process of urbanization has greatly promoted the urban development.
This development is not only reflected in the expansion of urban areas but also in economic and social
patterns. All these changes affect the way the urban operates. Then, a concern arising from the changing
urban dynamics is that whether current administrative division accords with urban development? Existing
studies conceptualize the urban space as the environment created by human activities, and elaborate the
importance of urban boundaries respecting to human activities in urban management. Following this concept,
we delineate the urban interior boundaries formed by human activities. Specifically, taking Xi’an in Shaanxi
Province of China as an example, this study first explores the region-based human crowd mobility patterns
to verify that human mobility can establish a stable correlation between regions, or capture the objective
correlations between regions. Then, the above human crowd patterns have been found to be applicable for
mining unusual urban regions from the perspective of anomaly detection, and empirical evidence has found
that these regions are of great significance for understanding the urban spatial structure. Finally, we employ
the community detection technology to naturally delimit the urban interior boundaries formed by human
mobility, and make a comparison with the official urban boundaries. Some unexpected communities that are
closely linked due to human activities appear from the results, and these findings help the urban planners
re-examine the administrative division.
INDEX TERMS Anomaly regions, community detection, human mobility, urban interior boundaries.
I. INTRODUCTION
THE world is in the process of urbanization. The world’s
population is shifting from rural to urban areas. In 1960,
33.61% of the world’s population live in urban areas, to-
day the percentage is 55% (Figure 1) and this percentage
is expected to increase to 68% by 20501. The population
migration leads to the expansion of urban areas. Except for
these obvious changes, the urbanization has also led to eco-
nomic and social changes [1], [2]. All these changes have
impacts on the arrangement of spatial regions, which is known
as the administrative division. The administrative division is
committed to providing efficient management for the society,
assisting in the overall development of the urban. Thus, facing
the process of urbanization, it is necessary to re-examine the
administrative division.
1https://data.worldbank.org/topic/urban-development
In reality, there have been cases in which the government
tries to adjust administrative divisions, but the proposal has
been opposed by citizens. This may be due to the subjective
nature of official adjustments [3], which do not align with
the spatial cognition of urban residents. Spatial cognition is
human understanding and perception of geographic space [4],
[5]. For an individual, his spatial cognition is determined
by his activities in the urban space [6], [7], and this spatial
cognition dilutes the human impression of official bound-
aries. Respecting the boundaries of human activities is of
great significance for urban management [8], [9]. Administra-
tive divisions are the division of regions (small geographical
spaces compared to cities) together. This is not a matter of
individual circumstances, but needs to be in line with the
spatial cognition of the relevant regional human crowd, which
is influenced by the activity patterns of the regional human
VOLUME 11, 2023 1
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
crowd. From a perspective consistent with human spatial cog-
nition, urban space can be conceptualized as the environment
created by human activity [10], [11]. Following this concept,
we delineate urban boundaries formed by human activities
from a data-driven perspective.
FIGURE 1: Change in the proportion of urban and rural
population worldwide from 1960 to 2019.
To achieve this, a complete and reasonable process is to
first find a good proxy to characterize the human activity
across spatial regions, and then explore the corresponding
activity pattern of regional crowd, and finally try to divide the
urban space according to the activity pattern. The region here
is a small geographical space, which is the basic unit of urban
space division. However, after finding the proxy for the hu-
man activity, existing studies [12]–[16] directly implements
community detection methods on the spatially-embedded net-
works deriving from human activity across regions to delin-
eate the urban boundaries. These studies lack an examination
on the activity patterns of regional crowd, they do not explain
what kind of regional crowd’s activity pattern they follow
and how their methods respect to the activity pattern, all of
which lead no basis for the practice of their methods and
no the interpretations for their results. In fact, the proxies of
existing studies are diverse, involving communication, trade,
social contact and so on, and all correspond to the complex
and varied spatial interactions. Their lack of explanation may
lead to a misunderstanding, that is, any activity can be used to
establish a network to carry out community detection, and the
results can be taken as the result of urban interior boundaries,
which is obviously unreasonable. Administrative divisions
should follow the activity pattern of regional human crowd,
with the aim of grouping together regions that are relevant
to the activities of regional human crowd. However, human
activities are diverse. An activity may vary greatly among
individuals, so that the activities of a regional crowd may
present an unpredictable situation. For example, the activities
of a regional crowd may make the region establish a wide
range of associations with other regions at random. At this
situation, whether the corresponding human activity actu-
ally capture the objective spatial correlation between regions
remains to be investigated, so the rationality of the results
obtained by community detection technology is questionable.
Even if community detection techniques are used to obtain
partitioning results, the reliability and interpretability of the
results are relatively poor.
This study employs the human mobility to characterize hu-
man activity across urban regions, and conducts a case study
to redraw the urban boundaries of Xi’an, China. At this time,
the urban interior boundaries respecting to human activities
should conform to the mobility patterns of regional crowd.
Thus, we first explore the region-based human crowd mobil-
ity patterns to reveal the mechanism of spatial interaction. We
characterize the region-based human crowd mobility patterns
through three features: mobility entropy [17], [18], displace-
ment [19], and human turnover, and find that the whereabouts
of region-based human crowd present a relatively certain
trend, which indicates that the human mobility establishes
reliable correlations between regions. Then, we analyze the
spatial interaction generated by human mobility from the
perspective of network and anomaly detection respectively,
and mine the key regions in the urban structure. These key
regions contribute to a better understanding of human spatial
interactions in urban. With these mobility patterns, we utilize
the community detection technology in a logic way to redraw
the interior boundaries of Xi’an. Compared with the official
boundaries, several unexpected communities have emerged
which really present closer internal relations. These findings
capture the latest urban dynamics and help to objectively
examine the urban interior boundaries.
The major contributions of this paper include the follow-
ing:
•Our study sheds light on the mobility patterns of region-
based human crowd, and concludes that the human
crowd moves within the urban in a relatively certain
trend. These findings provide the evidence that human
mobility can capture the objective correlation between
urban regions and provide a reasonable basis for spatial
division based on human mobility.
•In contrast to the network attribute-based approach com-
monly used in the literature, our study finds that anoma-
lous mobility patterns of region-based human crowd can
be used to explore key urban regions, which can help to
reveal the spatial structure of urban.
•Based on the real human mobility data in a whole ur-
ban, we redraw the interior boundaries of Xi’an, China.
Compared with the official urban boundaries, the results
provide reasonable insights for urban planning.
We have organized this paper according to the following
sections. Section II describes the work related to the deter-
mination of urban boundaries. In Section III, we introduce
the process and main methods of this study. Specifically, in
Section III-A, we first introduce the study area and research
data, then we briefly describe the process of processing hu-
man mobility data into the spatial interaction network, which
characterizes the regional interrelations. Then, in Section
2VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
III-B, we explore the region-based human crowd mobility
patterns, and in Section III-C, we describe the community
detection method for automatically dividing urban regions. At
last, in Section IV, we divide regions together in an automatic
way to find the boundaries formed by human activities, and
make a comparison with the official boundaries.
II. BACKGROUND AND RELATED WORK
The administrative division is determined by government
agencies to serve politics and administration. However, there
may be several subjective operations in the process of deter-
mining the boundary [20], which may result in some improper
divisions. For urban citizens, their perception of urban space
depends on their activities in the urban [21], which weakens
the impression for official boundaries. And for human ac-
tivities in urban, modern research holds that socio-economic
factors play an increasingly important role in driving human
activities [22], [23]. Taking mobility in human activities as an
example, its patterns may be changed by the socio-economic
factors such as wage imbalance. Human activities are related
to the social economy of urban, and can objectively reflect
the dynamics of the urban to some extent. There have been
studies exploiting human activities across the urban space
to assess the effectiveness of urban growth boundaries [24].
Following these, Jiang et al. [20] propose the ’natural city’
which refers to the environment formed by human activities.
Under these circumstances, existing studies have sought
to employ a good proxy to characterize the human activities
across urban regions, and then resort to the network-based
approach to detect objective urban boundaries. Chen et al.
[12] utilize the activity trajectories of Nanjing citizens riding
shared bicycles to construct a spatial interaction network,
and then implement a community detection algorithm on
this network to delineate urban activity areas, and ultimately
compare the boundaries of the areas with the boundaries of of-
ficial administrative districts. Jin et al. [13] utilize the human
travel data obtained from Baidu Huiyan Platform to conduct
the spatial interaction analysis, and then directly apply the
community detection algorithms to delineate the borders of
activity spaces in the city. Shen et al. [14] apply a multi-
level modularity optimisation algorithm to detect community
structures in the London Metropolitan Area, where the multi-
level is designed for the commuting flows of different groups
of people, so the method is essentially a hierarchical use of
community detection algorithms directly on spatial interac-
tion networks. Blondel et al. [25] employ mobile commu-
nications to examine the border in Belgium, and detect the
underlying linguistic border. Sobolevsky et al. [15] employ
human communication activities to characterize the spatial
interaction within the country, such as France, and redraw the
geographical maps according to human activities. Zhong et
al. [16] use the smart card data of three years to respectively
determine the boundaries of Singapore in these three years,
and identify the evolution of urban structure brought about
by development.
All these studies build spatial interactions through human
activities first, and then directly apply the community detec-
tion technology to aggregate related regions. However, few
studies have truly understood the patterns of these human
activities, let alone how they follow the human activity pat-
terns. Although human activities may follow the distance
decay effect, that is, the intensity of interaction between two
regions decreases with the increase of geographical distance,
the effect does not explicitly express the correlation between
regions, and only this effect is insufficient to explain how
human activities shape the boundaries of ’natural city’. The
patterns of human activities are not clear, so the rationality
of using community detection technology directly is ques-
tionable, and the interpretability of corresponding results is
poor. Therefore, this study first explore the mechanism of
spatial interaction produced by human mobility, making sure
that this spatial interaction do establish reliable correlations
between regions, and then we delineate the urban boundaries
respecting to human activities.
III. MATERIALS AND METHODS
The framework of this study mainly contains three parts
(Figure 2). The first part is the data preparation stage. We
map individual raw cellular data into urban spatial activity
trajectories, and then aggregate crowd trajectories to form
spatial interaction network. The second part is the analysis of
region-based human crowd mobility patterns. At this stage,
three mobility-related features mobility entropy, displace-
ment, and human turnover are proposed. We analyze the basic
human crowd mobility patterns through univariate analysis
and correlation analysis, and then conduct anomaly pattern
analysis to explore the abnormal regions. The third part is
urban interior boundaries delimitation. Community detection
algorithm is applied to automatically delineate the internal
boundaries of urban, and the obtained boundaries are used
for comparison with the actual official boundaries.
A. STUDY AREA AND DATA PROCESSING
1) Xi’an districts
As a national central city in western China, Xi’an is an
important scientific research, education and industrial base
in China. As of 2023, Xi’an covers an area of 10752 square
kilometers, and comprises 11 districts and 2 counties.
In history, the administrative division of Xi’an has been
adjusted many times. Administrative division aims to serve
the development of urban, regional economy and society. In
recent years, Xi’an has experienced rapid development. The
urban population has expanded from 8.156 million in 2015 to
9.8687 million in 2018; the built-up area of Xi’an has grown
from 548.60 square kilometers in 2015 to 724 square kilome-
ters in 2018; and the urban GDP has increased from 58.12
billion in 2015 to 83.4986 billion in 20182. In order to satisfy
the development needs of national central cities and better
undertake the task of leading regional economic development,
it is necessary to examine the existing administrative divisions
2http://tjj.xa.gov.cn/tjnj/2019/zk/indexch.htm
VOLUME 11, 2023 3
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
FIGURE 2: Framework of the methodology.
of Xi’an. The study area covers all the districts of Xi’an, as
shown in Figure 3.
2) Cellular data
Our research data, coming from the 3G cellular network of
Xi’an, records the mobility information of 24,770,715 mobile
phone users in Xi’an from October 1, 2015, to October 31,
2015. A mobile phone in normal operation will be associated
with the nearest base station to record its current location,
leaving the original location data.
We process each user’s above daily original data into stay
point sequence which represents this user’s space activity
trajectory of that day referring to studies [26]. In our study,
a stay point refers to a geographical region within 500 meters
in which a user stays in this region for more than 30 minutes.
Stay points indicate the meaningful activity regions for users.
Then, a user i’s space activity trajectory in one day is rep-
resented as a stay point sequence SPi={spi1,spi2,···,spin},
where each point spij ∈SPicontains the longitude, latitude
and arrival and departure time.
3) Mobility based spatial interaction network
Based on the stay point sequence in section III-A2, we obtain
the spatial interaction information generated by human mo-
bility. Specifically, taking a user’s stay point sequence as an
example, the spatial activities of this user result in the spatial
interactions between corresponding regions of the adjacent
stay points. Considering the spatial activities of all users, we
can obtain the representation of spatial interaction among
the regions within the urban caused by human mobility. We
define the interaction network as an undirected weighted
graph G= (N,E,W), where Ncontains all the regions in the
study area, E={(i,j) : i,j∈N}contains all edges which
directly connect the regions iand j, and W={wij :i,j∈N}
are the travel volume between the regions iand j.
According to the data processing in section III-A2, adjacent
stay points may relate to the same geographic region. Our
focus is on the interaction between regions. Therefore, con-
sidering the large number of all users’ stay points, we divide
the urban into grids, and take grids as the basic region to study
the spatial interaction.
To determine the grid size, with the above user trajectory
data, we calculate each user’s radius of gyration which is
a metric to distinguish users’ mobility patterns, and chose
the distinct geographic distance which separates total users
equally into two main groups as the grid size [27], [28].
Figure 4 depicts the cumulative distribution function of the
radius of gyration, in which users with the radius of gyration
less than 1094 meters account for 50.2% of the total. Finally,
we set each grid as a square with 1000 meters width. Mapping
the above user stay points into grids, we calculate the travel
volume between any two grids.
Then, we obtain the interaction network generated by hu-
man mobility. Descriptive statistics for the basic topological
properties of the interaction network are provided in Table 1.
TABLE 1: The basic topological properties of the interaction
network.
Attribute Value
Number of nodes 3325
Number of edges 1276913
Average degree 768.068
Average trip volume by weighted edges 72323.67
Average shortest path length by edges 1.783
B. REGION-BASED HUMAN CROWD MOBILITY PATTERNS
Although we have obtained the urban spatial interaction net-
work, the regional crowd mobility pattern that affects the
spatial division is still unclear. If the mobility of region-based
human crowd presents an irregular situation, it is impossible
to delineate the reliable boundaries respecting to human activ-
ities. Therefore, it is imperative to explore the human crowd
mobility patterns to better understand the spatial interaction
caused by human mobility.
1) Basic patterns analysis
In this section, we first explore the entropy and displacement
of region-based human crowd. Then coupling with the human
4VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
(a) Changes in Xi’an (b) Xi’an districts
FIGURE 3: The study area.
CDF
0.0 2.5 5.0 7.5 10.0
0.00
0.25
0.50
0.75
1.00
(1.094, 0.52)
Radius of gyration (km)
FIGURE 4: The CDF of the radius of gyration.
turnover which is measured by the associated travel volumes,
we comprehensively understand the urban spatial interaction.
In the study of individual mobility patterns, [17] measure
the entropy of each individual’s trajectory, and find the po-
tential predictability of human mobility. Here, the entropy is
a measure of uncertainty [29], [30]. In our study, the spatial
interaction is related to the human mobility at collective level.
Therefore, we take the region-based human crowd as the basic
unit, and treat the spatial interactions between regions as the
representation of human crowd mobility, and then employ the
entropy to measure the certainty of human crowd mobility.
Taking region ias an example, regions interacting with region
iare considered as the locations visited by the human crowd
in region i, and the corresponding travel volumes measure the
interaction frequencies with region i.
We employ two types of entropy to measure the region-
based human crowd mobility pattern. The first random en-
tropy is calculated as:
H1 = log2ni,(1)
where niis the number of regions interacting with region
ithrough human mobility, indicating the probability of the
human crowd in region irandomly interacting with other
regions. The second region-based human crowd mobility en-
tropy is calculated as:
H2 = −
ni
X
j=1
pi(j) log2pi(j),(2)
where pi(j)is the probability of the human crowd in region i
interacting with region j, characterizing the uncertainty of the
human crowd interaction in region i.
To characterize the uncertainty of human interaction across
the whole regions, we calculate H1and H2for human crowd
in each region, and the distributions P(H1) and P(H2) are
shown in Figure 5. There is a obvious left shift of P(H2)
compared with P(H1), which indicates that the mobility of
region-based human crowd is far from random. P(H1) peaks
when H1is approximately equal to 9.5, which indicates that
the human crowd in one region may on average interact with
2H1≈724.08 regions, in the case of random movement. In
contrast, P(H2) peaks when H2is approximately equal to
5.2, which indicates that the real uncertainty of the human
crowd mobility in a typical region is not 724.08 but 25.2=
36.76. These results demonstrate that region-based human
crowd tends to interact with several certain regions, that is
to say, there is a relatively directional trend in the mobility of
human crowd.
0 2 4 6 8 10 12 14
Entropy
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Probability
H1
H2
FIGURE 5: The probability distribution of two mobility en-
tropy. The H1is the random entropy, and the H2is the region-
based human crowd mobility entropy. The obvious left shift
of P(H2) compared with P(H1) indicates that the mobility
of region-based human crowd is far from random.
VOLUME 11, 2023 5
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
Except for the entropy, we explore the displacement of
region-based human crowd, as it measures how far the human
crowd typically moves. For the human crowd in region i, the
displacement is calculated as:
di=1
Wi
ni
X
j=1
wij ×dij,(3)
where niis the number of regions interacting with region i
through human mobility, wij is the the travel volume between
the regions iand j,Wi=Pni
j=1 wij, and dij is the distance
between region iand region j.
Figure 6 shows the distribution of displacement. The distri-
bution peaks when the displacement is approximately equal to
3162.68 meters. Thus, the movement range of human crowd
in a typical region is within 3162.68 meters, which seems to
indicate that there is an implicit boundary of crowd activity.
10210310 4105
Log for displacement
0.00000
0.00005
0.00010
0.00015
0.00020
Probability
FIGURE 6: The probability distribution of displacement.
The above statistical results of the univariate analysis is a
mixture of all regions with different situations. Taking into
consideration the region human turnover which is measured
by the associated travel volumes, we characterize each region
with three features: entropy, displacement and region human
turnover. To better understand the spatial interaction within
the urban, we conduct correlation analysis on these features
to reveal more general conclusions next.
Figure 7a depicts a scatter plot of the region-based human
crowd mobility entropy and the log amount of region human
turnover. The plot overlays a linear trend, which presents a
positive correlation between the entropy and region human
turnover. This situation indicates that with the enhancement
of region human turnover, the uncertainty of the human
crowd’s whereabouts also increases correspondingly. There
are differences between individual mobility, and it is the
diversity of a large number of individuals that leads to in-
creased uncertainty at the collective level. While Figure 7b
presents a negative correlation between the log of displace-
ment and the log amount of region human turnover, which
indicates that human crowd in the regions with high human
turnover are more likely to interact with nearby spaces in
close proximity. Combining these two results, we find that
although the uncertainty of human crowd mobility is high in
the regions with large human turnover, these crowds interact
more evenly with the nearby regions around them; and in the
regions with low mobility uncertainty, these crowds are more
inclined to interact with some specific regions far away. These
conclusions are further confirmed by Figure 7c.
In conclusion, although individual mobility presents di-
versity, at the collective level, region-based human crowd
mobility exhibits a degree of certainty that is far from random,
and the movement range seems to be limited by an implicit
boundary. The region-based human crowd either interacts
with neighboring regions, exhibiting localized activity, or are
more inclined to interact with particular regions. All of these
indicate that some regions are indeed more closely linked by
human activities, and that this relationship is reliable, which
reflects the objective interrelations between urban regions.
2) Anomaly pattern analysis
The above results represent the major patterns of urban spatial
interaction. Although these patterns are generic to most re-
gions, we still notice that minority regions present anomalous
patterns, which are presented as the outliers in Figure 7, such
as the region with high region human turnover and small
entropy in Figure 7a. Anomalies present the special patterns
which are different from normal instances [31], [32]. In an
urban environment, anomalies are sometimes critical, as they
may undertake special functions or indicate the particular
events in the urban [33]–[35]. The anomalies may play an
important role in understanding urban interaction. Existing
studies usually detect the key elements of urban structure,
such as urban centers, from a network perspective to under-
stand the urban interaction [16], [36], [37]. Here, from an
anomaly detection perspective, we discriminate the outliers,
and compare them with the key elements of urban structure
mentioned in existing studies to better understand their role
in urban spatial interaction.
Specifically, we use three features, entropy, displacement,
and human turnover, to characterize the crowd mobility pat-
terns in each region, and apply the Isolation Forest method
to separate the anomalous instances from the rest of the
instances [38]. This method explicitly isolates the anomalies
based on the features, and ensure the feature-values of the sep-
arated anomalies are very different from the normal instance.
Considering the complexity of human mobility, the spatial
interactions in regions with low human turnover have a certain
contingency, thus this study focuses on the anomalous regions
with high region human turnover. Here, we present several
anomalous regions and their corresponding features in Table
2. In these regions, the region human turnovers are relatively
high, while the human crowd mobility entropy are small, and
the displacements are large. This indicates that the large-
scale human crowd in these regions tend to interact more
directionally with several remote regions, which is contrary
to most other regions.
Different from the above anomalies, existing studies mine
the key elements of urban structure such as hubs from a
network perspective to better understand urban interactions.
6VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
(a) (b) (c)
FIGURE 7: Relationship between mobility entropy and displacement of region-based human crowd, and region human turnover.
(a) The plot of mobility entropy and the region human turnover. (b) The plot of displacement and region human turnover. (c)
The plot of displacement and entropy.
TABLE 2: The features of several anomalous regions.
Region Related travel volume(log) Mobility entropy Displacement(log)
1 13.97 1.52 16.27
2 13.32 2.77 16.35
3 10.98 1.16 11.81
4 9.46 1.32 16.39
These key elements are more central than others, and are often
associated with the large degree property in the urban related
network [39]. Next, we explore the relationship between the
abnormal regions mined from our anomaly detection perspec-
tive and the key regions mined from the network perspective.
Referring to the study [36], we employ two centrality indices
betweenness centrality and PageRank to determine the hubs
and centers in the urban, and compare them to our outliers.
According to the urban spatial interaction network, we
calculate the two central indicators of each region, and select
the first 1% of the two indicators to map to the geograph-
ical space. As shown in Figure 8, the hubs and centers are
usually concentrated within the urban, and sometimes they
are the same or very close to each other, while our outliers
are located more on the outskirts of the urban. We further
map the spatial interactions of some anomalous regions into
geographical space as shown in Figure 9. We find that the
large numbers of human in these anomalous regions are more
likely to interact with the distant regions within the urban,
such as the urban center regions. These outliers are more
like the gateway for external people to enter the urban. For
example, in Figure 9a, there is a bus station near the abnormal
region, and in Figure 9b, there is a highway station entrance
near the abnormal region. They may be the key elements for
human commuting between the urban and the outside, and
are important for understanding urban interactions. Together
with the results mined from the network perspective, these
elements mined from the human crowd mobility patterns help
us better understand the urban structure.
FIGURE 8: The hubs(blue), centers(red) and anomalous re-
gions(green). The hubs and centers are mined through the
two centrality indices betweenness centrality and PageRank
from a network perspective, and the anomalous regions are
distinguished according to the region-based human crowd
mobility patterns.
C. COMMUNITY DETECTION
There are various community detection algorithms, and the
most popular is the modularity optimization approach [40],
[41]. The modularity is defined as the number of edges falling
within groups minus the expected number in an equivalent
network with edges placed at random [40], and it is a com-
monly used metric to measure the strength of network com-
munity structure. For a particular division of the network, the
modularity is expressed as
Q=1
2mX
C∈PX
i,j∈CAij −kikj
2m(4)
VOLUME 11, 2023 7
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
(a) Region1 (b) Region2
(c) Region3 (d) Region4
FIGURE 9: The spatial interactions of four selected anomalous regions.
where Pis the community set of a particular network division,
node iand node jbelong to the same community C,Aij is the
number of edges between node iand node j,m=Pi,jAij/2,
kiand kjare the degrees of node iand node jin the network,
respectively. The modularity with a higher value indicates a
stronger community structure of the network division, that is,
the better the division quality is.
IV. RESULTS AND DISCUSSION
A. URBAN INTERIOR BOUNDARIES DELIMITATION
According to the conclusions in Section III-B, human crowd
in a typical region is more inclined to interact with several
specific regions, and tends to move within a certain range.
The interactions between regions are mutual, thus human
crowds interact more purposefully within a certain range,
which presents that the intra-interaction of the collection of
regions within this range are much higher than the inter-
interaction between this collection and other geographic re-
gions. As a result, several regions are more closely linked
through human activity, and it seems that these regions appear
to be naturally isolated from other regions, just like there is a
geospatial boundary formed by human activities. Therefore,
we can delineate the boundaries formed by human activities
through dividing closely related regions together. We hope to
achieve this in an automatic way, and make a comparison with
the official boundaries.
In our study, the mobility based spatial interaction network
is the representation of human interacting across the urban
regions, in which the nodes represent the regions and the
weights represents the interaction intensities between regions.
According to the patterns of human activity described above,
dividing closely related regions together is consistent with
the paradigm of community detection. Thus, based on the
derived spatial interaction network, we utilize the community
detection algorithm to determine the boundaries formed by
human activities.
In this study, we adopt the Louvain method [42] to divide
our spatial interaction network, which optimizes the mod-
ularity to obtain the optimal number of communities and
the network division automatically. Specifically, this method
employs a repeated two-stage process to divide the nodes
into groups. At the first stage, each node is assigned to a
community. At the second stage, for each node in the network,
this method traverses all neighbor nodes of this node, and
measures each gain of modularity brought by removing this
node from its own community to the community where its one
neighbor node belongs to. Then this node is assigned to the
community which achieves the maximum modularity. This
process traverses all nodes and is repeated until no higher
modularity can be achieved. In fact, we have also tried another
two commonly used community detection algorithm Infomap
[43] and walktrap [44], however, by considering the results
of network division and modularity, we finally choose the
Louvain method with better comprehensive performance. All
8VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
these three methods are conducted using the i-graph package
on the R platform, and their average number of communities
and modularity after 100 times of execution are shown in
Table 3.
TABLE 3: The division results of three community detection
methods.
Method Walktrap Infomap Louvain
Average number of communities 233 78 10
Average modularity 0.45 0.49 0.51
To evaluate the quality of the network division, we calcu-
late the internal interaction ratio for each community, i.e. the
percentage of the travel volumes human move within the same
community. As shown in Table 4, the internal interaction in
each community achieves a high percentage, with 88.52% at
the most and 50.74% at the least, which fulfills our purpose
of dividing closely interacting regions together.
TABLE 4: The internal interaction ratio of each community.
Community
number
Internal interaction
ratio
Community
number
Internal interaction
ratio
1 82.13% 6 81.31%
2 65.33% 7 52.01%
3 88.52% 8 54.54%
4 87.38% 9 51.03%
5 87.40% 10 50.74%
B. COMPARE WITH OFFICIAL URBAN BOUNDARIES
In this section, we make a comparison between the urban
boundaries generated by human activities and the official
urban boundaries. The boundaries of communities generated
by human activities are shown in red in Figure 10, in which the
blue boundaries are the official urban boundaries. Our main
findings are as follows:
FIGURE 10: The 10 communities generated by human activ-
ities.
The urban space of Xi’an is divided into 10 regional sets,
and all communities are made up of geographically adjacent
regions. The community detection algorithm neither specifies
the number of communities nor adds the restriction on the
contiguity of regions. Only driven by optimizing the modu-
larity, a measure of community quality, the method automati-
cally obtains the community structure in which the members
in the same community are closely connected, while the links
between the communities are relatively less. According to
the above human crowd mobility patterns, the region-based
human crowds interact within a certain range, and some
regions within the range may be naturally isolated from other
regions due to the frequent human interactions. Therefore, it
is reasonable for adjacent regions to be divided together, as
shown by the community detection results.
Compared with the official 13 districts, the community
1,2,4 and 5 generally follow the official districts, while other
communities differ greatly from the official districts. Among
the four communities which are the most similar to the official
districts, the boundary of community 4 is basically consistent
with that of the official LanTian County, and the boundary
of community 1 extends to the left and right, occupying part
of the official administrative regions corresponding to com-
munity 2 and community 5, especially the northeast corner of
community 1. Although the boundaries of community 1, 2,
and 5 are somewhat offset from their official counterparts, the
average internal interaction ratio of these three communities
is 78.29%, which is close to the official 79.56%, based on Ta-
ble 4 and Table 5,and the internal interaction ratio of commu-
nity 1 is even higher than that of official Huyi district. Human
space activities will naturally link some regions more closely
to form so-called communities, and the connections between
communities are relatively less, just like there are barriers.
These barriers determine the boundaries of human activities,
and these boundaries may be a more reasonable division of
urban space. The consistency of some community boundaries
with official boundaries suggests that human mobility in these
regions follows official boundaries.
Then, we analyze the results that are clearly inconsistent
with official boundaries. In the northeast corner of Xi’an,
the three official districts Yanliang district, Lintong district
and Gaoling district are merged into two communities: com-
munity 3 and 6. Specifically, the south of Lintong district
alone forms community 6, while the rest of the three districts
are merged to form community 3. According to Table 4 and
Table 5, the average internal interaction ratio of community 3
and 6 is 84.92%, which is higher than 81.59% of the three
corresponding official districts. Among them, the internal
interaction ratio of community 3 is as high as 88.52%, which
is not decreased due to regional merger, but higher than that
of a single administrative district, especially the 74.29% in
Gaoling district. In the case of these three official districts, the
more interconnected regions are divided together by merging
and regrouping. Our division organizes regions in a way that
is more consistent with human activities, which is likely to
reduce the urban management problems caused by a large
number of cross-district activities.
For the remaining six official districts, our method divides
them into four communities: community 7,8,9 and 10. Al-
VOLUME 11, 2023 9
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
TABLE 5: The internal interaction ratio of each official region.
Official
district
Internal interaction
ratio
Official
district
Internal interaction
ratio
Huyi District 81.95% Weiyang District 64.38%
Xincheng District 38.05% Chang’an District 66.33%
Lianhu District 39.30% Gaoling District 74.29%
Baqiao District 56.93% Yanliang District 87.80%
Zhouzhi Country 90.42% Lintong district 82.67%
Beilin District 33.33% LanTian County 84.93%
Yanta District 55.32%
TABLE 6: Interaction of the communities in the core region.
Community
number
Internal interaction
ratio
Interaction ratio with
other 3 communities
Interaction ratio outside
the 4 communities
7 52.01% 32.95% 15.03%
8 54.54% 34.97% 10.49%
9 51.03% 34.73% 14.23%
10 50.74% 39.59% 9.66%
though the internal interaction ratios of these four communi-
ties are lower than that of community 1-6, their average inter-
nal interaction ratio is 52.08%, which is higher than 47.89%
of the six corresponding official districts, according to Table 4
and Table 5. We further analyze the human interaction in these
four communities. Here, we calculate the percentage of the
interaction between each community and the other three com-
munities, and the percentage of the interaction between each
community and community 1-6. As shown in Table 6, except
for the interaction with itself, the interaction between each
community and the remaining three communities accounts
for the vast majority. Taking the community 7 as an example,
52.01% of the interactions occur within itself, followed by
32.95% of interactions with the other three communities, and
only 15.03% of interactions with community 1-6. The average
internal interaction ratio is higher than that of the official dis-
tricts, which means that our division is more consistent with
human activities. As for the relatively low internal interaction
ratio compared to community 1-6, it is just because of the
more interaction between these four communities. As the core
regions of Xi’an, these regions are highly integrated, which
results in a large interaction between these regions and makes
it difficult to divide the communities.
In summary, several official boundaries really respect the
human activities, while in some regions, the human cross-
border activities are more frequent. The objective division
of urban space in this study better reflects the interaction
of human beings in urban space, provides a perspective for
re-examining the official administrative division, and also
provides an opportunity for managers to think about urban
social and economic problems. For example, look for the
driving factors of human cross-border activities in official
regions corresponding to community 3 and community 6 in
Figure 10, and then find the hidden urban problems.
V. CONCLUSION
In this paper, we have used the cellular data to generate
the mobility based spatial interaction network of Xi’an,
China, and objectively delineate the urban interior bound-
aries formed by human activities to re-examine the official
administrative division. To ensure capturing the actually re-
gional relevance, we first explore the region-based human
crowd mobility patterns at the collective level. The analysis
of patterns indicates that region-based human crowd mobil-
ity presents some certainty, which is far from random. This
conclusion confirms that there is indeed a reliable connec-
tion between urban regions due to human mobility. Also,
the discovered patterns are used to distinguish anomalous
regions, revealing the special structure in the urban. All these
patterns and anomalous regions provide the novel perspective
to understand the urban dynamics. Then, based on the reliable
connection between regions, we objectively divide the closely
related regions together. Compared with the official bound-
aries, several unexpected communities have emerged. These
communities really present a closer correlation, respecting
to the objective internal interaction ratio, and reveal several
naturally differentiated regions within the urban. The ob-
tained divisions offer a perspective to re-examine the official
administrative division.
ACKNOWLEDGMENT
This work was supported by the Shanxi Province Basic Re-
search Program, grant number 202203021222103.
REFERENCES
[1] M. Dociu and A. Dunarintu, ‘‘The socio-economic impact of urbaniza-
tion,’’ International Journal of Academic Research in Accounting Financ,
vol. 2, no. 5900, pp. 47–52, 2012.
[2] Y. Li, Z. Xie, B. Li, and M. Mohiuddin, ‘‘The impacts of in situ urban-
ization on housing, mobility and employment of local residents in china,’’
Sustainability, vol. 14, no. 15, p. 9058, 2022.
[3] P. Martí, L. Serrano-Estrada, A. Nolasco-Cirugeda, and J. L. Baeza, ‘‘Re-
visiting the spatial definition of neighborhood boundaries: functional clus-
ters versus administrative neighborhoods,’’ Journal of Urban Technology,
vol. 29, no. 3, pp. 73–94, 2022.
[4] G. L. Allen, Applied spatial cognition: From research to cognitive technol-
ogy. Psychology Press, 2020.
[5] Y. Xu, J. Rollo, D. S. Jones, Y. Esteban, H. Tong, and Q. Mu, ‘‘Towards
sustainable heritage tourism: a space syntax-based analysis method to
improve tourists’ spatial cognition in chinese historic districts,’’ Buildings,
vol. 10, no. 2, p. 29, 2020.
10 VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
[6] X. Ye and C. Andris, ‘‘Spatial social networks in geographic information
science,’’ pp. 2375–2379, 2021.
[7] D. R. Montello, ‘‘Cognition of geographic information,’’ in A research
agenda for geographic information science. CRC Press, 2004, pp. 61–91.
[8] S. Chang, Z. Wang, D. Mao, F. Liu, L. Lai, and H. Yu, ‘‘Identifying urban
functional areas in china’s changchun city from sentinel-2 images and
social sensing data,’’ Remote Sensing, vol. 13, no. 22, p. 4512, 2021.
[9] Y. Chen and A. Deng, ‘‘Using poi data and baidu migration big data to
modify nighttime light data to identify urban and rural area,’’ IEEE Access,
vol. 10, pp. 93 513–93 524, 2022.
[10] M. Gandy, ‘‘Urban political ecology: A critical reconfiguration,’’ Progress
in Human Geography, vol. 46, no. 1, pp. 21–43, 2022.
[11] C. Schliephake and U. Schliephake, Urban Ecologies: City Space, Material
Agency, and Environmental Politics in Contemporary Culture, 01 2015.
[12] W. Chen, X. Chen, L. Cheng, X. Liu, and J. Chen, ‘‘Delineating borders
of urban activity zones with free-floating bike sharing spatial interaction
network,’’ Journal of Transport Geography, vol. 104, p. 103442, 2022.
[13] M. Jin, L. Gong, Y. Cao, P. Zhang, Y. Gong, and Y. Liu, ‘‘Identifying
borders of activity spaces and quantifying border effects on intra-urban
travel through spatial interaction network,’’ Computers, Environment and
Urban Systems, vol. 87, p. 101625, 2021.
[14] Y. Shen and M. Batty, ‘‘Delineating the perceived functional regions of
london from commuting flows,’’ Environment and Planning A: Economy
and Space, vol. 51, no. 3, pp. 547–550, 2019.
[15] S. Sobolevsky, M. Szell, R. Campari, T. Couronne, Z. Smoreda, and
C. Ratti, ‘‘Delineating geographical regions with networks of human in-
teractions in an extensive set of countries,’’ PLOS ONE, vol. 8, no. 12,
2013.
[16] C. Zhong, S. M. Arisona, X. Huang, M. Batty, and G. Schmitt, ‘‘Detecting
the dynamics of urban structure through spatial network analysis,’’ Inter-
national Journal of Geographical Information Science, vol. 28, no. 11, pp.
2178–2199, 2014.
[17] C. Song, Z. Qu, N. Blumm, and A.-L. Barabási, ‘‘Limits of predictability
in human mobility,’’ Science, vol. 327, no. 5968, pp. 1018–1021, 2010.
[18] M. Schläpfer, L. Dong, K. O’Keeffe, P. Santi, M. Szell, H. Salat, S. Ankle-
saria, M. Vazifeh, C. Ratti, and G. B. West, ‘‘The universal visitation law
of human mobility,’’ Nature, vol. 593, no. 7860, pp. 522–527, 2021.
[19] L. Alessandretti, U. Aslak, and S. Lehmann, ‘‘The scales of human mobil-
ity,’’ Nature, vol. 587, no. 7834, pp. 402–407, 2020.
[20] B. Jiang and Y. Miao, ‘‘The evolution of natural cities from the perspective
of location-based social media,’’ Professional Geographer, vol. 67, no. 2,
pp. 295–306, 2015.
[21] K. Lynch, K. Lynch, K. M. Lynch, and K. Lynch, ‘‘The image of the city,’’
Cambridge Mass, vol. 11, no. 1, pp. 46–68, 1960.
[22] H. Barbosa-Filho, M. Barthelemy, G. Ghoshal, C. R. James, M. Lenor-
mand, T. Louail, R. Menezes, J. J. Ramasco, F. Simini, and M. Tomasini,
‘‘Human mobility: Models and applications,’’ Physics Reports, vol. 734,
2017.
[23] S. Des Roches, K. I. Brans, M. R. Lambert, L. R. Rivkin, A. M. Savage,
C. J. Schell, C. Correa, L. De Meester, S. E. Diamond, N. B. Grimm et al.,
‘‘Socio-eco-evolutionary dynamics in cities,’’ Evolutionary Applications,
vol. 14, no. 1, pp. 248–267, 2021.
[24] Y. Long, H. Han, Y. Tu, and X. Shu, ‘‘Evaluating the effectiveness of urban
growth boundaries using human mobility and activity records,’’ Cities,
vol. 46, no. 46, pp. 76–84, 2015.
[25] V. Blondel, G. Krings, and I. Thomas, ‘‘Regions and borders of mobile
telephony in belgium and in the brussels metropolitan zone,’’ Brussels
Studies, vol. 42, no. 4, pp. 1–12, 2010.
[26] Y. Wei, X. Zhao, J. Zou, and E. Herrera-Viedma, ‘‘A complementing
preference based method for location recommendation with cellular data,’’
Knowledge-Based Systems, vol. 183, p. 104889, 2019.
[27] J. Yin, A. Soliman, D. Yin, and S. Wang, ‘‘Depicting urban boundaries
from a mobility network of spatial interactions: a case study of great britain
with geo-located twitter data,’’ International Journal of Geographical
Information Science, vol. 31, pp. 1293 – 1313, 2017.
[28] Y. Wei, M. Mao, X. Zhao, J. Zou, and P. An, ‘‘City metro network
expansion with reinforcement learning,’’ in Proceedings of the 26th ACM
SIGKDD international conference on knowledge discovery & data mining,
2020, pp. 2646–2656.
[29] F. Buono and M. Longobardi, ‘‘A dual measure of uncertainty: The deng
extropy,’’ Entropy, vol. 22, no. 5, p. 582, 2020.
[30] H. Wang, S. Zeng, Y. Li, and D. Jin, ‘‘Predictability and prediction of
human mobility based on application-collected location data,’’ IEEE Trans-
actions on Mobile Computing, vol. 20, no. 7, pp. 2457–2472, 2020.
[31] R. Foorthuis, ‘‘On the nature and types of anomalies: a review of deviations
in data,’’ International journal of data science and analytics, vol. 12, no. 4,
pp. 297–331, 2021.
[32] S. Thudumu, P. Branch, J. Jin, and J. Singh, ‘‘A comprehensive survey of
anomaly detection techniques for high dimensional big data,’’ Journal of
Big Data, vol. 7, pp. 1–30, 2020.
[33] M. Zhang, T. Li, Y. Yu, Y. Li, P. Hui, and Y. Zheng, ‘‘Urban anomaly
analytics: Description, detection, and prediction,’’ IEEE Transactions on
Big Data, vol. 8, no. 3, pp. 809–826, 2020.
[34] K. Kalair and C. Connaughton, ‘‘Anomaly detection and classification
in traffic flow data from fluctuations in the flow–density relationship,’’
Transportation Research Part C: Emerging Technologies, vol. 127, p.
103178, 2021.
[35] M. Zhang, T. Li, H. Shi, Y. Li, P. Hui et al., ‘‘A decomposition approach
for urban anomaly detection across spatiotemporal data.’’ in IJCAI, 2019,
pp. 6043–6049.
[36] E. C. Hui, X. Li, T. Chen, and W. Lang, ‘‘Deciphering the spatial structure
of china’s megacity region: A new bay area—the guangdong-hong kong-
macao greater bay area in the making,’’ Cities, vol. 105, p. 102168, 2020.
[37] M. Roosta, M. Javadpoor, and M. Ebadi, ‘‘A study on street network
resilience in urban areas by urban network analysis: comparative study
of old, new and middle fabrics in shiraz,’’ International journal of urban
sciences, vol. 26, no. 2, pp. 309–331, 2022.
[38] F. T. Liu, M. T. Kai, and Z. H. Zhou, ‘‘Isolation forest,’’ in Data Mining,
2008. ICDM ’08. Eighth IEEE International Conference on, 2009.
[39] M. Curado, L. Tortosa, J. F. Vicent, and G. Yeghikyan, ‘‘Analysis and
comparison of centrality measures applied to urban networks with data,’’
Journal of Computational Science, vol. 43, p. 101127, 2020.
[40] M. E. J. Newman, ‘‘Modularity and community structure in networks,’’
Proceedings of the National Academy of Sciences of the United States of
America, vol. 103, no. 23, pp. 8577–8582, 2006.
[41] S. Fortunato, ‘‘Community detection in graphs,’’ Physics Reports, vol. 486,
no. 3-5, pp. 75–174, 2009.
[42] V. D. Blondel, J. Guillaume, R. Lambiotte, and E. Lefebvre, ‘‘Fast unfold-
ing of communities in large networks,’’ Journal of Statistical Mechanics:
Theory and Experiment, vol. 2008, no. 10, p. 10008, 2008.
[43] C. T. Bergstrom, ‘‘Maps of random walks on complex networks reveal
community structure,’’ Proceedings of the National Academy of Sciences
of the United States of America, vol. 105, no. 4, pp. 1118–1123, 2008.
[44] M. Latapy, ‘‘Computing communities in large networks using random
walks,’’ J.of Graph Alg. & App.bf, vol. 10, no. 2, pp. 191–218, 2006.
YU WEI was born in 1990. Since January 2021, he has been a University
Teacher working with Taiyuan University of Technology. His research inter-
ests include urban planning and machine learning.
XI ZHAO received the B.S. degree in Information Engineering from Xi’an
Jiaotong University in 2003 and the M.S. degree in Systems Engineering
from Xi’an Jiaotong University in 2007, and a PhD in Information and Data
from the University of Central Technology in France in 2010. He went to the
Department of Computer Science at the University of Houston in the United
States for postdoctoral research and was promoted to Research Assistant
Professor. In 2015, he was selected as a special researcher for the Top notch
Talent Program of Xi’an Jiaotong University. In 2018, he was selected as a
professor for the Top notch Talent Program of Xi’an Jiaotong University.
VOLUME 11, 2023 11
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2024.3487921
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
