Accurate Detection of Built-Up Areas from High-
Resolution Remote Sensing Imagery Using a Fully
Convolutional Network
Yihua Tan, Shengzhou Xiong, Zhi Li, Jinwen Tian, and Yansheng Li
Abstract
The analysis of built-up areas has always been a popular
research topic for remote sensing applications. However,
automatic extraction of built-up areas from a wide range
of regions remains challenging. In this article, a fully con-
volutional network (
FCN
)–based strategy is proposed to
address built-up area extraction. The proposed algorithm
can be divided into two main steps. First, divide the remote
sensing image into blocks and extract their deep features
by a lightweight multi-branch convolutional neural net-
work (
LMB-CNN
). Second, rearrange the deep features into
feature maps that are fed into a well-designed
FCN
for im-
age segmentation. Our
FCN
is integrated with multi-branch
blocks and outputs multi-channel segmentation masks
that are utilized to balance the false alarm and missing
alarm. Experiments demonstrate that the overall classi-
fication accuracy of the proposed algorithm can achieve
98.75% in the test data set and that it has a faster process-
ing compared with the existing state-of-the-art algorithms.
Introduction
The built-up area is represented as the aggregation area of
buildings in remote sensing images, which is the main area of
human activity. Different from individual buildings, built-up
areas are a notion of closed-shape regions that include primar-
ily buildings but also lawns, parking lots, and plants. The ac-
curate regional scope of built-up areas is of great significance
in many fields, such as disaster preven
lization of land and resources, and urb
manual extraction of built-up areas req
and labor costs. Thus, it is crucial to automatically extract the
built-up areas from remote sensing images via pattern recog-
nition technology. The automatic extraction of built-up areas
also poses many challenges because built-up areas appear
with various types and styles in which spectral variation and
confusion, illumination changes, environmental diversity,
and the extent of plant covering are difficult to describe in
uniform features. In addition, the number of remote sensing
images that need to be processed is usually large, so the speed
of the algorithm must be considered.
There are many remote sensing data sources available,
which makes the automatic extraction of built-up areas
feasible. Li
et al.
(2014) chose synthetic aperture radar (
SAR
)
images as their research data, while others used polarimetric
SAR
(
PolSAR
) imagery (Xiang
et al.
2016). Multispectral images
are the basis of various spectral indexes, and their applications
are derived from the differences of spectral characteristics
between ground objects (Varshney and Rajesh 2014; Bouzekri
et al.
2015; Kaimaris and Patias 2016). Liu
et al.
(2014) further
extended spectral indexes to hyperspectral images. In addi-
tion, most people have chosen panchromatic images as experi-
mental data (Benediktsson
et al.
2003; Zhong and Wang 2007;
Sirmacek and Unsalan 2009), and pseudocolor images with
multisource fusion have also been favored (Pesaresi
et al.
2011;
Liu
et al.
2013; Chen
et al.
2018). Furthermore, some of the
less frequently used data have also attracted researchers’ atten-
tion in recent years, hoping for a breakthrough, such as road
network data (Zhou 2016; Zhou and Guo 2018), stereo imagery
that incorporates height information (Peng
et al.
2017), night-
time light data from the Visible Infrared Imaging Radiometer
Suite and the Defense Meteorological Satellite Program-Oper-
ational Linescan System (Dou
et al.
2017; Yang
et al.
2017; Ma
et al.
2017), and thermal infrared remote sensing considering
the temperature difference between different regions (Zhang
et al.
2017a; Tarawally
et al.
2018; Wang
et al.
2018). Panchro-
matic images have the characteristics of low acquisition cost,
high resolution, and good image quality. In this article, we
design a built-up area extraction algorithm and conduct our
experiments on a 1-m-resolution panchromatic image.
Through decades of research, a large number of extraction
schemes for built-up areas have been proposed in the litera-
ture, and various classification and segmentation algorithms
this field. Among the existing extraction
ort vector machine type classifiers have
i and Narayanan 2004; Tao
et al.
2012; Li
et al.
2015a; Li
et al.
2017a), and graph-based segmentation
algorithms have also been popular among researchers ((Sir-
macek and Unsalan 2009; Tao
et al.
2012; Li
et al.
2015b; Liu
et al.
2017). In addition, there have also been many methods
based on conditional random fields (Zhong and Wang 2007),
Markov random fields (Smits and Annoni 1999), and thresh-
old-based methods (Kovács and Szirányi 2013). A good clas-
sifier or segmentation algorithm can provide more accurate
and robust results, which is common in computer vision and
pattern recognition tasks. However, for built-up area extrac-
tion, the choice of features is more crucial. In recent studies,
artificial features are still the mainstream despite their limited
performance. The spectral indexes based on multispectral
data are popular due to their simple and fast calculations
(Varshney and Rajesh 2014; Bouzekri
et al.
2015; Vecchi
et
al.
2015; Sun
et al.
2017). However, they are greatly affected
Yihua Tan, Shengzhou Xiong, Zhi Li, and Jinwen Tian are
with the National Key Laboratory of Science and Technology
on Multi-Spectral Information Processing, School of Artificial
Intelligence and Automation, Huazhong University of Science
and Technology, Wuhan 430074, China.
Yansheng Li is with the School of Remote Sensing and
Information Engineering, Wuhan University, Wuhan
430079, China (Corresponding author: Yansheng Li; E-mail:
)
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 10, October 2019, pp. 737–752.
0099-1112/19/737–752
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.10.737
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2019
737
