Subido por nejawif398

UAV corn

Autonomous vehicles are leading a sensing revolution in
precision agriculture. Unmanned ground vehicles reduce
agricultural labor needs and im- prove the precision with
which the fields are managed. Highly mobile unmanned
aerial vehicles (UAVs) have been used to produce highaltitude, on-demand aerial imagery of fields, in a variety of
light wavelengths, allowing producers to quickly detect
pest infestations, weeds, and other harm-ful environmental
conditions. We believe UAVs, such as the one in Figure 1,
will use improved sensors and algorithms to go from highaltitude operation far above the crops, to low- altitude
autonomous operation within 1 m of crops in a field. Lowaltitude operation improves the spatial resolution of sensor
data, enables the use of shorter range, cheaper, and lighter
sensors (Solari, Shanahan, Ferguson, Schepers, & Gitelson,
2008), and makes physical interaction with the en-vironment
possible. These changes will allow smaller, lower cost, and
easier-to-operate UAVs to measure crop properties in
unstructured agricultural settings. Safe operation of these
next-generation UAVs requires precise localization that is
not possible with current systems’ reliance on low precision
GPS sensors.
UAVs’ high mobility and ability to fly above the crops
make them an excellent tool for collecting data from fields.
However, their limited payload capacity and a need for pre- cise
localization in field environments present challenges tousing
UAVs for field research. In this work, we developa system
and algorithms that use a laser scanner as botha means of
precisely localizing a UAV in a field and as a mechanism for
collecting scientific data. The localization is more accurate
than standard GPS systems, and has lower cost than precise
real-time kinematic (RTK) GPS solutions. Furthermore, our
system does not require a predefined fieldmap, which is not
always available, and dynamically reactsto field conditions in
flight. The laser-based system robustly copes with the chaotic
field environment and unpredictablelighting conditions that
camera-based visual odometry so-lutions struggle with.
A specific example of the utility of UAVs that motivates our work is corn phenotyping trials, such as those
shown in Figure 2. A field will contain many different genetic varieties. The field is divided into dozens or hundredsof
subplots, consisting of two rows, spaced 0.762 m (30 in)
apart, for a short distance (10–15 m). Each field may con- tain
the same variety planted multiple times, to increase the
population size for statistical analysis. Additionally, the
subplots have different treatments, such as different irrigation or fertilizer application rates, to analyze the response of
the different varieties to these test conditions. Manual
measurements and ground vehicles measure the plants’ response to environmental stimuli throughout the growing
UAV operating in corn field.
season (Busemeyer et al., 2013; Montes, Technow, Dhillon,
Mauch, & Melchinger, 2011; Reis, 2013). Researchers mea- sure
a wide variety of plant traits in these experiments, ranging
from the root structure of corn plants to the plant height and
biomass in a field at different stages of the plants’ lives.
Collecting field data via manual measurements or ground
vehicles has severe drawbacks. Collecting measure- ments via
manual measurements is time consuming for researchers.
Repeated field measurements by researchers risk damaging
the plants as the researchers walk through their dense canopy.
Automated test equipment mounted on ground vehicles is faster
than manual measurements, but also risks damaging the
plants, may compact the soil, and thus change the growing
conditions, and ground vehicles cannot traverse the field in all
environmental conditions. In particular, collecting data from
mature fields is especially challenging, as the dense plant
growth is difficult for field researchers and robots to move
through. The combination of these factors severely limits the
spatiotemporal resolution of phenotyping trials in large
environments. In practice, the low spatiotemporal data
resolution makes it difficult or impossible to accurately
measure the impact of short-termenvironmental stimuli on the
plants. Instead, large studies at the end of the growing season
measure the combined impact of all the stimuli that the plants
experienced over