IEEE Geoscience and Remote Sensing Magazine - March 2021 - 208

lower leaves of the canopy, which can reflect solar radiation
back to the upper section of the leaf cover [168]. Therefore,
the leaf thickness and growth status as well as the canopy
architecture primarily influence the reflectance pattern in
this spectral region [27]. The reflectance is gradually decreased with increasing wavelengths of up to 2,500 nm because the absorption of solar radiation is increased by the
water in crop leaves [59].
Multispectral cameras commonly include several spectral bands in the visible and infrared spectral regions [34].
Spectral bands are generally sensitive to leaf pigments and
the leaf and canopy structure [45], [67]. Compared with
their multispectral counterparts, hyperspectral cameras
have a higher spectral resolution, with continuous or discrete spectral bands in the visible and infrared spectral regions [17], [264]. Hyperspectral bands are more sensitive to
small changes in leaf pigments, such as carotenoids, chlorophyll a and b, and xanthophylls as well as leaf and canopy
structures [265]. Multispectral or hyperspectral bands are
used to obtain VIs through band calculations. More advanced methods based on radiative transfer model inversion techniques can also be employed [266].
To reduce the effects of illumination conditions on multispectral and hyperspectral cameras, radiometric image
calibration should be carried out at each band [266]. First,
several calibration targets (1 × 1 m) that have different reflectance values are placed in the UAV flight area. Second,
the digital number (DN) of each target is collected from
UAV multispectral and hyperspectral images, and the corresponding actual reflectance of each target is also measured
using, for example, an ASD spectroradiometer [166]. Third,
the regression relationship between the reflectance and the
DN value is established at each band image through empirical linear or nonlinear methods, and the DN value in
the images is converted to the normalized reflectance using
established regression relationship equations [71].
Figure 1(b) and (c) gives examples of maize and wheat
images acquired with MicaSense RedEdge-M and Cubert
UHD185 cameras, respectively. Multispectral and hyperspectral cameras can export images that include DN values and texture, reflectance, and color information [45].
Crop phenotyping trait estimations from multispectral and
hyperspectral cameras are presented in Tables 1 and 2, including GAI, plant density, ear density, FAPAR, canopy coverage, staygreen/senescence, crop dynamics monitoring,
phenology, plant diseases and pests, and so forth, across
the whole crop growing season. The models, parameters,
phenotype estimating traits, imaging environments, imaging techniques, and limitations of multispectral and hyperspectral cameras are detailed in Table 2.
The advantage of multispectral and hyperspectral cameras is that they have spectral information (except texture
and color information) that could be better used to estimate crop phenotyping traits through fusion algorithms
(this is an image process of integrating related information
from two or more images into a single image and could be
208

applied to acquire more useful information from the input
images) [59], [115]. The application of multispectral and hyperspectral cameras in the field of crop phenotyping traits
has attracted the attention of breeding programs [22], [103],
[104], [267]. The main limitations concern the very large volume of data associated with spectral imagery and the generally more complex operation of hyperspectral cameras compared to multispectral and RGB cameras. These shortcomings
impede their application and adaptation [36], [266].
In summary, multispectral cameras are mainly used
for VI-based traits, due to the limited number of available
bands, while hyperspectral cameras enable the calculation
of more advanced crop phenotyping traits, such as photosynthetic status and fluorescence [265], [266]. With the
rapid development of multispectral, hyperspectral, and
computer image-processing technologies, cheaper multispectral and hyperspectral cameras will be developed and
combined with high-performance graphics and computerclustering technologies.
THERMAL CAMERAS
Thermal cameras are used to measure infrared radiation in
the thermal spectral infrared regions as a water stress indicator [115]. Thermal cameras operate in the spectral range
from 3 to 14 μm, and their most-used spectral wavelengths
are 3-5 μm and 7-14 μm, respectively [71], [266], [268]. The
infrared radiation atmospheric transmission is very close to
its maximum value in these two spectral ranges [121]. The
thermal sensitivity of the spectral range from 3 to 5 μm is
associated with higher energy levels than the spectral range
from 7 to 14 μm, but using wavelengths of 7-14 μm has
advantages for certain applications [269]. The wavelength
ranges between 8 and 14 μm can reduce errors from the
atmospheric absorption of infrared radiation for objects
through longer atmospheric paths [269].
Thermal cameras should be calibrated to maintain the
measurement accuracy and stability of their images. First,
a thermal camera is preheated for roughly 20 min in the
field before a UAV flight to reduce the temperature drift,
and the automated nonuniformity correction is enabled
during the data measurement [270]. Second, according to a
thermal camera's preheating guidelines, one JPEG image of
a blackbody radiometric background scene is taken, which
is immediately obtained from one 14-b TIFF-format image
of the same scene [235]. Third, the blackbody radiometric
JPEG is loaded into FLIR Tools software, which provides
options for imaging target-related parameters and environmental conditions, to execute a radiometric conversion
[271]. At the same time, the air humidity and temperature
are added from a weather station, along with the target distance (flight height) and object emissivity to obtain temperature values that are output in a text file [272]. The bias
between the blackbody and the image temperature is determined. Finally, each pixel's radiometric value in the 14-b
TIFF image is correlated to the corresponding value in the
text file, with a linear model being used as a radiometric
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2021

IEEE Geoscience and Remote Sensing Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2021