GRover: developing sensors for vineyard usespaa.com.au/pdf/468_SPAA_Nov_EJE.pdf · GRover:...
Transcript of GRover: developing sensors for vineyard usespaa.com.au/pdf/468_SPAA_Nov_EJE.pdf · GRover:...
GRover: developing sensors for vineyard use
CSIRO AGRICULTURE AND FOOD
Everard Edwards and Matt Siebers Mark Thomas, Rob Walker
"Infrared spectrum" by Ibarrac at English Wikipedia. Licensed under CC BY-SA 3.0 via Wikimedia Commons - http://commons.wikimedia.org/wiki/File:Infrared_spectrum.gif#/media/File:Infrared_spectrum.gif
Non-destructive sensing
All objects emit radiation (passive sensing) and will absorb some received radiation (active sensing).
Development of ‘sensors’ since early 19th century:
• Daguerrotypes (1830’s), • bolometer (1880), sensitive to 0.0001°C, • X-ray image (1896), • etc.
Non-classified data from satellite imaging
since 1960: • e.g. infra-red – used
for monitoring cloud cover.
Sensing
Boulevard du Temple - 1838
Landsat 8 (2013) – free satellite data
14th Oct 2016, USGS Earth Explorer, http://earthexplorer.usgs.gov
1 pixel = 100 m x 100 m
We are here:
Remote sensing (e.g. satellite, aerial): • large area sampling, • but limitations in:
• frequency of coverage, • speed of data/analysis provision, • view angles.
Proximal sensing (e.g. tractor mounted):
• potentially higher resolution, • many possible viewing angles, • ‘instant’ data availability, (local
hardware / web-based tools). • but requires local knowledge/skills.
Remote sensing vs. proximal sensing
Multi-spectral image of vineyard Remote Sensing Australia
Greenseeker in use during fertiliser application.
• New technologies (sensors and software) have become pervasive through our lives and society.
• e.g. my phone contains: fingerprint, accelerometer, gyroscope, proximity, barometer, compass, A-GPS + two RGB cameras, one with a ‘colour spectrum’ sensor.
• Field measurements are labour intensive (whether for science or farming) and always benefit from greater ground coverage.
‘Digital viticulture’
Can we utilise these technologies to improve crop management?
• Fast Phenomics: grapevine trait characterisation in the field.
• New non-destructive technologies for simultaneous yield, crop condition and quality estimation.
• New technologies for dynamic canopy and disease management. • Evaluation of new technology and new scion-rootstock
combinations for improved water use efficiency and reduced costs.
CSIRO & ‘Digi Vit’ Wine Australia Projects
Agriculture & Food
Sensors for crop management & phenomics
• Based in the Winegrapes and Horticulture Group at the Waite Campus, Adelaide.
• Need for non-destructive, sensor based, systems to make detailed large scale field measurements for: • Field ‘phenomics’; the assessment of many breeding lines
in-field. • Crop management utilising plant based measurements.
• New and developing technologies will provide non-contact sensors for: • accurate yield forecasts, • fruit composition/ripeness, • canopy management, • disease assessment, • water management, etc.
A mobile vineyard platform (GRover)
• Group has developed a self-propelled (manual steer) platform with HRPPC. • Will take multiple sensors at multiple
positions. • Very large payload weight. • Can view all parts of vine (aboveground). • No regulatory compliance required.
• Currently fitted with LiDAR scanner • biomass components, • canopy properties, • potentially yield estimation.
• Stereo RGB and hyperspectral in process of being added.
Using GRover to measure canopy size
The LiDAR sensor generates a 3D ‘point cloud’.
Point cloud is analysed to provide field measurements.
One example ‘is voxelisation’.
Using GRover to measure canopy size
The LiDAR sensor generates a 3D ‘point cloud’.
Point cloud is analysed to provide field measurements.
One example ‘is voxelisation’.
R² = 0.8906
02468
1012141618
0 200000 400000 600000
Leaf
are
a/pa
nel (
m2)
Number of voxels
Provides an accurate estimate of canopy area.
Using GRover to measure pruning weight
Pruning weight correlations can differ between genotypes and environments
R² = 0.6853 R² = 0.8184
0 2 4 6 8
10 12 14
0 10000 20000 30000 Voxel number
Pruning weight vs. voxel number Pr
unin
g w
eigh
t / p
anel
(kg)
Dual wavelength ‘echidna’ LiDAR, DWEL
Greyscale renders of DWEL data, a) vines with leaves stripped away to expose fruits, b) highlighted using a ratio of the two DWEL wavelengths, and c) vines with leaves in place.
Dual wavelength ‘echidna’ LiDAR, DWEL
• Sensing & data analytics is a rapidly expanding area.
• Likely to see major impact of this technology over next 5-10 years.
• Offers a wealth of new tools for precision agriculture.
• LiDAR is a robust instrument for biomass measurements, but still expensive for commercial vineyard use and optical parts need to be clean.
• New CSIRO projects are examining a range of instruments for a variety of vineyard measurements.
Summary
Demonstration of ‘point cloud’
CSIRO Agriculture and Food Everard Edwards Research Team Leader t +61 8 8303 8649 e [email protected] w www.csiro.au/agriculture
CSIRO AGRICULTURE AND FOOD
Acknowledgements Jose Jimenez-Berni & others at the High Resolution Plant Phenomics Centre, Canberra. Mick Schaefer, CSIRO Land & Water / Auscover.