IEEE Geoscience and Remote Sensing Magazine - March 2021 - 215

The results obtained by Jansen et al. [321] showed that a
Cercospora leaf-spot index obtained from spectral indices
was correlated with the Cercospora disease severity in sugar
beet varieties. Furthermore, the classification accuracy of
tomato diseases was better using a superresolution method (which produced a spatial resolution-enhanced image
using low-resolution images) compared to using the conventional image-scaling methods with RGB images [103].
Behmann et al. [104] employed a miniaturized, handheld
hyperspectral camera to identify powdery mildew at the
canopy scale in barley. In short, these results indicated that
the occurrence of disease and the resistance of crop varieties to infection can be effectively estimated using nearground and UAV-based sensors. The fluorescence sensor
was more sensitive to crop diseases than other instruments.
It has the potential to be used in combination with hyperspectral, multispectral, RGB, and thermal cameras for the
early detection of crop disease severity. Such sensors have
a great potential to be harnessed to estimate the crop disease severity in future crop breeding programs. These nearground and UAV-based sensors will detect the resistance of
crop varieties to disease more effectively when combined
with newly developed image classification methods.
PHENOLOGICAL PARAMETERS
Crop phenology provides important information for simulating and monitoring crop growth and development [322],
estimating crop yields [323], optimizing production management and decision-making [324], and analyzing crops'
response to climate change [289]. Remote sensing-based
observations with short revisit times and large spatial coverage have been employed to successfully retrieve crop phenological parameters using VI time-series data at regional
to global scales [76], [77]. The NDVI and enhanced VI have
been commonly employed to estimate crop phenology. The
most-studied phenological parameters contain the start
of season (SOS), peak of season, and end of season (EOS)
[78]. Currently, various methods have been used to extract
these phenological parameters using VI time series data, including dynamic threshold methods [79], fixed threshold
methods [79], maximum-slope methods [80], functionfitting methods [81], moving average methods [82], and the
valley-detection method [83].
The dynamic threshold method is the most commonly
employed for estimating the crop SOS and EOS because it
needs fewer parameters, has an easier application, and produces high-accuracy results. Sakamoto [84] developed the
shaped model-fitting method with MODIS wide-dynamicrange VI time-series data to evaluate the timing of phenological parameters for 36 crop development stages of major
U.S. food products. However, Verger et al. [85] found that
leaf area index (LAI) is more robust and sensitive than VIs
in considering vegetation from different satellite data. Furthermore, Luo et al. [86] used global land surface satellite
LAI products and optimal filter-based phenology detection
methods to obtain the crop SOS and EOS and produced a
MARCH 2021

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

1-km spatial resolution phenological survey of three staple
crops in China from 2000 to 2015.
PolSAR is very sensitive to crop morphological structure
changes [87]. RADARSAT-2 and TerraSAR-X provide opportunities to explore the potential of these data for crop
SOS and EOS estimation. Lopez-Sanchez et al. [88] used
dual-polarized TerraSAR-X data and simple decision tree
algorithms to evaluate rice phenological parameters. Furthermore, they harnessed the three polarimetric parameters (entropy, anisotropy, and alpha angle) from polarimetric RADARSAT-2 data and hierarchical decision trees to
estimate rice phenological parameters [89]. Wang et al. [87]
applied the multitemporal RADARSAT-2 dataset, SAR polarimetric decompositions, and an RF algorithm to monitor canola, maize, soybean, and wheat phenology based on
crop phenological parameters. With the fast development
of UAVs and sensors, near-real-time crop phenology retrieval is feasible. Yang et al. [90] proposed a near-real-time
deep-learning method for monitoring rice phenological parameters and estimated rice phenology from UAV RGB images with an accuracy rate of 83.9%. In the future, the PhenoCams network will be combined with satellite images to
better estimate different crops' phenological parameters.
UAV images and advanced machine learning algorithms
will improve crop phenological parameter retrieval in near
real time and thus enable optimized crop management.
LODGING
Crop lodging is the permanent deviation of crop plants from
the upright position [325]. Crop lodging not only influences harvest operations but also results in yield losses and
can reduce quality [326]. Therefore, accurate and rapid
evaluation of crop lodging is important for research and
screening, optimizing field crop management, obtaining
better production, and estimating crop yield losses. The
reflectance and backscatter at different wavelengths are
affected by changes in the crop structure [327], biochemical properties [328], and morphology [329] in the lodging condition.
Many researchers have used crop phenotyping platforms to estimate crop lodging [134]-[143]. An early study
by Fitch et al. [330] applied the linear polarization of backscatter from wheat to evaluate its potential in detecting
wheat lodging using ground-based platforms. Ogden et al.
[135] used motor-driven cameras to analyze textural information in images to estimate the extent of rice lodging. Liu
et al. [331] demonstrated that optical hyperspectral data
can be used to detect rice lodging. Recent advances in the
development of UAVs and sensors have been used to detect crop lodging areas and evaluate lodging severity [134],
[136]-[138]. Chapman et al. [136] reported that wheat
lodging areas are successfully detected using thermal images from UAV-mounted thermal cameras. The combination of spectral and texture features and the digital surface model further increased the estimation accuracy of
rice lodging classification using single-feature probability
215

IEEE Geoscience and Remote Sensing Magazine - March 2021

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