29. Correlation coefficients of multi-temporal satellite imagery based NDVI, i.e., June 1, 7,
15, and July 17 2015, with the sampled grain protein contents were calculated as 0.77,
0.79, 0.68, and -0.34, respectively.
Conclusion: NDVI of satellite image has good correlation with grain protein contents
before ripening growth stage, whilst the reliability of using satellite image based NDVI
to estimate grain protein contents deteriorates as ripening stage approaches.
To further quantitatively analyze relationship
between NDVI and grain protein contents,
different regression models such as linear,
exponential, power, and second-order polynomial
regressions were constructed, respectively.
Conclusion: second-order polynomial regression
model best fitted the regression model, with the
coefficient of determination value of 0.67 and
RMSE of 0.58.
衛星画像とドローン画像の融合
30. NDVI= (NIR‐RED) ⁄ (NIR+RED)
VDVI= (2G‐B‐RED) ⁄ (2G+B+RED)
Regression analysis showed that second-order
polynomial models best fitted between UAV’s
RGB image based VDVI and satellite image
based NDVI, when compared with other
regression models such as linear, exponential,
and power. Coefficient of determination and
RMSE are calculated as 0.58 and 0.127.
8 UAV flights
(100m AGL, 12.5cm)
2, 10, 19, 25,
June; 2, 10, 16,
24, July 2015.
4 Satellite images
(5m)
1, 7, 15 June,
and 17 July
2015
衛星画像とドローン画像の融合
34. Based on the stepwise regression model, map of estimated grain weight per square meter
(yield map) was generated by extracting each pixel values out of the maps of
accumulative VIs.
Within-field spatial variations of wheat yield could be observed, which could be seen as
the comprehensive presentation of the spatial variations of soil fertility, tiller density,
effective water potential, canopy aeration condition, and so on.
The mean value of grain weight per square meter was calculated as 0.72kg. About 36%
areas had grain weight below the average, and over-flourishing areas (high yield)
coincided with the spots of lodging by visual inspection.
ドローン画像による収量の推定
35. 17 May, 2016@ wheat farmland of
Hokkaido Univ.
Altitude : about 50m AGL
Image size: 35m*25m
Spatial resolution: 18mm
ドローン画像による小麦の茎数推定
38. Natural
Neighborhood
Interpolation
According to the power regression model, stalk densities were estimated by
extracting 125 samples of FCC randomly taken all around the field.
By using spatial interpolation method in GIS software, stalk density map
was generated.
y = 1067.7x1.37
ドローン画像による小麦の茎数推定
39. About 43% acreages have the stalk density under 400 per
square meter, which need high dose of nitrogenous
topdressing desperately for generating enough stalks before
the end of tillering stage.
Most areas reached the stalk density between 401-800 per
square meter, occupying about 45% acreages, which need
moderate or low dose of nitrogenous topdressing.
About 12% acreages in the field
have the stalk density beyond
801 per square meter, which
should be applied with no more
fertilizer, otherwise over-
flourishing canopies would lead
to the occurrence of lodging at
later growth stage.
ドローン画像による小麦の計数推定
40. What is Topographic map?
Topography is the study of
earth surface shapes and
features, and topographic
map refers to a 2D graphic
representation of a terrestrial
or 3D land surface feature
using contour lines,
hypsometric tints, and relief
shading.
地形マッピング
41. Why Topographic map?
In precision agriculture, high accurate topographic maps are essential to such
operations including soil preparation, drainage arrangement, land forming and
high precision land levelling.
1. Surface unevenness of farmlands has been pointed out as a major issue that affected
agricultural drainage efficiency of the current irrigation systems, which have been
under great pressure for producing more with lower water supplies.
2. Crops are vulnerable to stagnant water throughout germination period to early
growth stages, and puddles are potential threat leading to crop drowning and
occurrence of infestation as well as plant diseases due to high humidity.
3. Unleveled rice paddy suffered about 24% yield loss in average when compared to the
effectively leveled ones for the same rice variety and the same fertilizer input
(Rickman 2002).
地形マッピング
42. Solution: laser-assisted and GPS
based land levelling.
Topographic survey is prerequisite for
high precision land levelling.
Land levelling accuracy,
efficiency, as well as energy
consumption is in high
accordance with the delicacy of
each topographic map.
地形マッピング
43. LiDARfixgps hhhH
To conduct topographic
survey for high precision
farmland levelling
operation in a simple and
totally autonomous
manner using a UAS-
LiDAR system.
December 7th 2016 @
Mackey, Australia.
coscos DhLiDAR
ドローンによる地形マッピング
44. PPK GPS altitude measurements changed along
time, ranging from 0.08m to 0.095m due to
usage of different combinations of satellites
over time, and the accuracy could be confirmed
as about 0.015m (1.5cm).
LiDAR distance measurements remained
substantially constant over time mostly
ranging from 1.873-1.881m, and the
measuring accuracy was confirmed
within 0.01m (1cm).
ドローンによる地形マッピング
45. The topographic map could
practically provide reference
for laser-assisted land leveler to
locate laser transmitter, and
also could be used to produce
cut/fill ratio map for GPS
based land leveler.
ドローンによる地形マッピング
47. Blue: RTK GPS
equipped combine
harvester’s track on
21st July 2017;
Topographic map
based on RTKGPS
data of combine
harvester.
Topographic map
based on data
integration of
UAV’s PPK GPS
and LiDAR data.
Extracting height information from the same positions on two
different topographic maps to calculate RSME and make
correlation analysis.
Red: PPK
GPS equipped
UAS-LiDAR
system’s track
on 10th Aug.,
2017.
ドローンによる地形マッピング
48. Red: 35
samples of
UAV-LiDAR’s
PPK GPS, and
35 samples of
combine
harvester’s
RTK GPS. RMSE of UAV-LiDAR’s
elevation and RTK GPS’s
elevation: 0.035m.
ドローンによる地形マッピング
49. Orthomosaic image taken on 7th Sep., 2017, and DEM map generated
based on stereo image techniques during orthomosaic process.
ドローンによる地形マッピング
50. Green lines showed UAV-LiDAR system’s flight track on 7th Sep., 2017,
and topographic map generated using natural neighborhood
interpolation method.
ドローンによる地形マッピング
51. Larger coverage (light weight);
Better spatial resolution (0.048*0.048 m);
High relative precision.
Vs
Absolute accuracy cannot be guaranteed due to
propagated error.
High absolute accuracy;
Vs
Poor details due to large track
intervals (0.5*20m).
Photogrammetry UAS-LiDAR
ドローンによる地形マッピング
55. Sugarcane NDVI map on 13th Oct., 2016@Mackey, Australia
Red: UAS-LiDAR’s flight track before harvesting
Green: UAS-LiDAR’s flight track after harvesting
Based on second flight (after harvesting), farmland
topographic map was generated for precision land
levelling;
Conduct correlation analysis by extracting crop
height and NDVI.
Green-Red= Sugarcane height.
(Lon., Lat., Sugarcane height)
(Lon., Lat., NDVI)
作物生育量の推定
56. Active NDVI: Cropspec (2 laser channel: NIR & Red);
Passive NDVI: multispectral camera (RGB + Red edge+ NIR)
To study effect of changing flight altitude (5m-12m)
on Cropspec’s NDVI.
To conduct correlation analysis by calculating
passive and active NDVI.
作物生育量の推定