Cropland Extraction Based on OBIA and
Adaptive Scale Pre-estimation
Dongping Ming, Xian Zhang, Min Wang, and Wen Zhou
Abstract
Object-based image analysis (OBIA) provides a solution for
cropland extraction from high spatial resolution remote sensing
images. Currently, scale parameter selection is often depen-
dent on subjective trial-and-error methods or post-evaluation
of multi-segmentation, which directly reduces efficiency of
cropland extraction. This paper proposes a cropland extraction
method combining spatial statistics based adaptive scale
parameter pre-estimation and object-oriented classification.
SPOT5 multi-spectral image in Baishan is used as experimental
data to verify the validity of the methodology. Experimental
results show that the pre-estimated scale parameter can yield a
classification result with both high classification accuracy and
completeness for extracting cropland information. This present-
ed method avoids the time-consuming trial-and-error practice
by accelerating the object-oriented classification procedure. Hi-
erarchical rule set based classifications achieve higher accura-
cies and lower fragmentation than nearest neighbor-supervised
classification. Additionally, this methodology can be rapidly
transplanted into different regions and it is helpful for dynamic
land-use monitoring and precision agriculture.
Introduction
Agriculture production is an important part of sustainable de-
velopment. Monitoring croplands is the key to food security
in the twenty-first century (Thenkabail, 2010), and cropland
distribution maps are basic requirements for cropland man-
agement and precision agriculture (Ma
et al
., 2014). However,
most currently available cropland products suffer from major
limitations such as the absence of precise spatial locations of
cropped areas and coarse resolution with significant uncer-
tainties in areas, locations, etc.(Thenkabail
et al
., 2012, Then-
kabail
et al
., 2009a; Thenkabail
et al
., 2009b; Thenkabail and
Wu, 2012).Traditional pixel-based methods have been suitable
for coarse resolution imagery. With the increase of spatial res-
olution of remote sensing images, pixel-based multi-spectral
image classification in cropland extraction not only leads to
misclassification but also results in broken patches caused by
the salt and pepper effect. A considerable amount of research
has suggested that object-based approaches are superior to
traditional pixel-based methods in the classification of high
spatial resolution data (Qian
et al
., 2015). Thus, object-orient-
ed classification has become a useful alternative for cropland
extraction, especially for very high-resolution (
VHR
) images.
OBIA
can effectively incorporate spatial information and
expert knowledge into the classification, and the classified
image objects are useful foundations for remote sensing and
GIS
integration. Thus,
OBIA
also provides a better solution and
has been used in cropland extraction for high spatial resolu-
tion remote sensing images (Lu
et al
., 2007; Shen
et al
., 2011;
Ma
et al
., 2014).
However, with increasing complexity of the scenes, classes
and features to be extracted, automation of object-oriented
classification needs to be fed with more human experience in
a usable form, which will take a long time. Multi-scale image
segmentation is the foundational procedure of
OBIA
that trans-
forms the digital image from discrete pixels to homogeneous
image object primitives. Scale parameter selection in the
segmentation is especially key to
OBIA
because inappropriate
scales will lead to over-segmentation or insufficient segmen-
tation, which will directly reduce the accuracy and efficiency
of cropland extraction from high spatial resolution remote
sensing images. Of course, for high spatial resolution images,
although there are other types of classification strategies that
do not require a single parameter to be tuned or modification
of a single parameter, such as smoothing-based classification
(Schindler, 2012) and Random Fields-based contextual image
classification (Moser and Serpico, 2013; Zhong
et al
., 2014),
segmentation-based
OBIA
is still becoming increasingly prom-
inent in remote sensing science. There have been studies on
OBIA
-based cropland extraction, and adaptive scale parameter
selection is one of the main difficulties in this method (Shen
et
al
., 2011; Ma
et al
., 2014; Li
et al
., 2015; Aguilar
et al
., 2015).
Currently, the selection of segmentation scale parameters
in
OBIA
-based cropland extraction is mainly dependent on
experience (Ma
et al
., 2014; Sun
et al
., 2014) or post-evalua-
tion of segmentation. According to whether reference poly-
gons or a reference image is involved, remote sensing image
segmentation evaluation can be divided into two classes:
supervised image segmentation evaluation (Zhang
et al
., 2008)
and unsupervised image segmentation evaluation (Espindo-
la
et al
., 2006; Belgiu and Dra
ˇ
gut, 2014). Supervised image
segmentation evaluation usually uses reference polygons or
training objects to compute empirical goodness or segmenta-
tion error to parameterize the segmentation scale (Tian and
Chen, 2007; Zhang
et al
., 2008; Clinton
et al
., 2010; Anders
et
al
. 2011; Liu
et al
., 2012; Belgiu and Dra
ˇ
gut, 2014; Yang
et al
.,
2015a; Zhang
et al
., 2015). Supervised segmentation evalua-
tion is good at identifying the optimal scale of target objects,
however, its dependence on reference data makes it more diffi-
cult to use in operational settings (Belgiu and Dra
ˇ
gut, 2014).
Unsupervised image segmentation evaluation considers sole
intrasegment homogeneity (Baatz and Schäp, 2000; Hay
et al
.,
2005; Kim
et al
., 2008; Dra
ˇ
gut
et al
., 2010; Zhao
et al
. 2012;
Dra
ˇ
gut
et al
., 2014), both intersegment heterogeneity and
intrasegment homogeneity (Espindola
et al
., 2006; Johnson
Dongping Ming, Xian Zhang, and Wen Zhou are with the
School of Information Engineering, China University of
Geosciences (Beijing), 29 Xueyuan Road, Haidian, Beijing,
100083, China (
).
Min Wang is with the Key Laboratory of Virtual Geograph-
ic Environment (Nanjing Normal University), Ministry of
Education, Nanjing, Jiangsu, China, 210023; and also with the
Jiangsu Center for Collaborative Innovation in Geographical
Information Resource Development and Application, Nanjing,
Jiangsu, China, 210023.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 8, August 2016, pp. 635–644.
0099-1112/16/635–644
© 2016 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.8.635
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2016
635
