observation. Two
UAVs
with 2.0 and 2.8 m wing-span, fuel engine,
and payload of 2.27 kg were used for their high stability. They
were equipped with a video system with a small overlay board.
RGB
,
NIR
, and thermal mosaicked images were analyzed
in Jensen
et al.
(2012) for determining the temperature in
streams, once the land is separated out in the image mosaics.
They used an
UAV
(AggieAir, 2015) for this application. This
fixed-wing platform was designed for riparian and wetlands
applications (Jensen
et al.
, 2011) and was also used in Zaman
et al.
(2011) to quantify the spread of invasive grass species,
Phragmites australis
, in a large and important wetland in
northern Utah.
Sediments, oil spills, or other pollutants can be also de-
tected and tracked in aquatic environments. This task is car-
ried out in Zang
et al.
(2012) with a fixed-wing
UAV
with the
following specifications: airframe length of 1.8 m, wingspan
of 2.8 m, maximum
TOW
of 15 kg, and payload 3 -3.55 kg. The
UAV
is equipped with a digital
RGB
commercial camera and a
multispectral device in the spectral range of
RGB
and
NIR
.
Monitoring of swamps was carried out in Lechner
et al.
(2012); the purpose is to protect species sensitive to changes in
hydrological conditions. Object-based image analysis methods
were applied to characterize swamp land-cover on the Newnes
Plateau in the Blue Mountains near Sydney, Australia. A fixed-
wing
UAV
with 2 m wingspan weighing approximately 3.9 kg
was the platform used. It is equipped with a commercial cam-
era with the spectral range coverage modified by a
NIR
filter.
Monitoring of
Eriophorum vaginatum
at Mer Bleue peat-
land was carried out by Kalacska
et al.
(2013) because of its
relevance in methane exchanges in large areas. A rotorcraft
1.2 m long with a main rotor diameter of 1.3 m was used.
Video images were recorded in
RGB
to generate georeferenced
mosaics. The content of each mosaic was classified to identify
E. vaginatum tussock cover using a supervised classification
approach based on multi-distance cluster analysis.
Ouédraogo
et al.
(2014) determined micro-topography
changes in watersheds with high agricultural activity. This
study was based on the analysis of agricultural structures
(crops, furrows, ridges) affecting the topography through the
generation of high precision
DEMs
.
Biodiversity Analysis
From the point of view of biodiversity analysis, Getzin
et al.
(2012) determined the floristic diversity in the forest under-
storey. This application is environmental monitoring where
gaps in the canopy are analyzed with the goal of preserving
structural diversity and niche differences within habitats for
stabilizing species coexistence, which was achieved utilizing
the high-resolution
RGB
images captured with a fixed-wing
UAV
. Later, Getzin
et al.
(2014) have studied gap distributions
in forest canopies to determine the regeneration of trees based
on statistical analysis, which was achieved using high spatial
resolutions (7 cm/pixel) based on
RGB
ortho-rectified images
acquired from a
UAV
with weight of 6 kg and wing span of 2 m.
Also regarding biodiversity analysis, land-use change
monitoring is an important application to control greenhouse
gas emissions and biodiversity loss (Wich and Koh, 2012).
Rural Roads and Geological Infrastructures
Unpaved roads in rural environments were monitored with
the aim of identification and surface inspection (Zhang, 2014).
The platform used was a helicopter with payload of 680 g
equipped with a digital commercial camera together with a
navigation system (
GPS
/
INS
). Also, rural roads were monitored
in Zhang (2014), where stereo-based techniques are applied to
obtain 3
D
orientation with a
RGB
camera onboard a helicopter
as an
UAV
. Geological hazard analysis is considered in Qian
et
al.
(2012) to early detection of anomalies and to prevent disas-
ters on infrastructure such as roads or bridges.
Vegetation: Classification, Coverage, and Development
Vegetation analysis, including classification and identification
(Ishihama
et al.
, 2012), coverage, and development are also an
application area where
UAVs
can be used successfully. Even
at some point, kites and balloons, equipped with commercial
cameras, have been used for monitoring vegetation in perigla-
cial areas in Alaska (Boike and Yoshikawa, 2003). Ecology is
an area highly benefited by the use of
UAVs
where the versatil-
ity and maneuverability of such vehicles, equipped with dif-
ferent sensors, offer researchers and end users and excellent
opportunity (Anderson and Gaston, 2013).
Vegetation analysis in agriculture, forestry, and forest map-
ping is excluded here, as it was considered in the Agriculture
and Forestry Section.
Reid
et al.
(2011) and Bryson
et al.
(2010) applied texture
descriptors to classify vegetation in natural and farmland
environments from images captured with a machine vision
system installed on a fixed-wing
UAV
.
Rangeland ecosystems cover large areas for different uses
including natural habitat, recreational opportunities, or cattle
forage.
UAVs
were early identified for applications in this con-
text (Hardin and Jackson, 2005). McGwire
et al.
(2013) used
NDVI for comparing spatial variability in green leaf cover
of semi-arid rangeland areas based on images captured with
UAVs
and Landsat Thematic Mapper with 2 cm and 30 m reso-
lution, respectively. The
UAV
was a 1.4 m length with a main
rotor span of 1.58 m and tail rotor span of 27 cm equipped
with a
CMOS
imaging chip, capturing images in wavelengths
ranges of green, red, and
NIR
comparable to
CIR
images and
TM
bands 2, 3, and 4.
From the point of view of the classification paradigm, La-
liberte
et al.
(2011) proposed imagery to obtain orthorectified
mosaics with radiometric calibration for rangeland vegetation
classification based on rule-based approaches. A fixed-wing
aircraft with wingspan of 1.8 kg and 10 kg is the platform
used, which was equipped with three sensors: a forward look-
ing color video camera, a digital
RGB
camera installed on the
wing, and a multispectral (700 g) sensor on the nose, consist-
ing of six individual
CMOS
digital cameras, arranged in a 2 × 3
array and using filters with center wavelengths from 450 nm
- 850 nm. Rango
et al.
(2009), Laliberte and Rango (2009 and
2011), and Laliberte
et al.
(2010) proposed an image texture-
based method for determining the coverage of rangelands with
different textures by applying several scales in ortho-rectified
mosaics. The platform was also the one used in Laliberte
et
al.
(2011), and the classification approach used was: object-
based, rule-based, and textured-based (homogeneity, con-
trast, dissimilarity, entropy, angular second moment, mean,
standard deviation, correlation, entropy). Later, Laliberte
et al.
(2011) analyzed different geometric errors of image mosaics
and classification accuracies at different levels of detail in
rangelands with the platform and sensors described above.
Different types of vegetation were classified in Arnold
et al.
(2012 and 2013), through vegetation indices, based on an im-
agery multi-spectral system, consisting of three visible bands
(400 nm - 590 nm, 500 nm - 590 nm, 590 nm - 670 nm) and
two infrared bands (670 nm - 850 nm, 850 nm - 1000 nm). The
images were acquired simultaneously with multiple
CCD
arrays
where the incoming light is projected with specially designed
dichroic-coated prisms to single monochrome
CCDs
. The multi-
spectral system was mounted on an unmanned helicopter.
Kelcey and Lucieer (2013) applied different texture de-
scriptors for describing both vegetation and non-vegetation
areas based on the co-occurrence descriptor and the random
forest machine-learning technique in ortho-mosaics. Knoth
et al.
(2013) classified four different types of vegetation using
NIR
imagery for monitoring restoration in cutover bogs.
Regarding the coverage analysis, Breckenridge
et al.
(2011)
and Breckenridge and Dakins (2011) used a camera based
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
April 2015
303
