1. Fabienne Maupas - IIRB Congress, Deauville - 05/06/2018
1
High-throughput phenotyping of
sugar beet crop using multispectral
drone imagery
2. 2
Background of the study
• A study in the framework of the Aker’s project
• Development of a drone remote sensing method for growth monitoring
To have non-destructive and timely monitoring of crop development and
biochemical traits
For high throughput
3. 3
Objectives of the study
• The key feature of drone imagery is the fine spatial resolution and therefore
the potential to separate soil and vegetation pixels.
How does this fine spatial resolution help us to better
estimate green fraction (GF), green area index (GAI) and
leaf chlorophyll content (Cab) ?
4. 4
Materials and methods
Year Trial Soil
Nitrogen fertilization level
(kg N/ha)
Number of
varieties
2016
Barenton 1 Sandy loam 0; 100; 150 4
Barenton 2 Sandy loam 0; 100; 150 9
St Memmie 1 Calcareous loam 40; 80; 120 4
St Memmie 2 Calcareous loam 40; 80; 120 9
2017
St Memmie 3 Calcareous loam 0; 40; 80 1
Charmont Chalk 0; 70; 110; 150 1
Nizy Clay loam 0; 40; 80 4
274 microplots available
5. • Airphen multispectral camera
http://www.hiphen-plant.com/plant-phenotyping/airphen_41.html
• Hexacopter
5
Materials and methods
Flight altitude : 35 m
Spatial resolution : 1.6 cm
850nm
450nm
530nm 675nm
570nm 730nm
Spatial resolution : 1.6 cm
6. 6
Materials and methods
• Range of targeted variables
Variable Unit Min - Max Mean Standard deviation
Green fraction - 0.18 - 0.97 0.61 0.25
Green area Index m²/m² 0.13 - 4.57 1.50 1.0
Chlorophyll µg/cm² 21.2 - 51.1 33.7 7.0
7. 7
Estimation of targeted variables
Vegetation index References VI formulation used in this study
𝐕𝐀𝐑𝐈 Gitelson et al. (2002)
𝑅530 − 𝑅675
𝑅530 + 𝑅675 − 𝑅450
𝐍𝐃𝐕𝐈 Rouse et al. (1973)
𝑅850 − 𝑅675
𝑅850 + 𝑅675
𝐂𝐈 𝐠𝐫𝐞𝐞𝐧 Gitelson et al. (2006a, 2005, 2003)
𝑅850
𝑅530
− 1
𝐂𝐈 𝐫𝐞 Gitelson et al. (2006a, 2005, 2003)
𝑅850
𝑅730
− 1
𝐌𝐓𝐂𝐈 Dash and Curran (2004)
𝑅850 − 𝑅730
𝑅730 − 𝑅675
𝐦𝐍𝐃 𝐛𝐥𝐮𝐞 Jay et al. (2017a)
𝑅450 − 𝑅730
𝑅450 + 𝑅850
𝑅800 − 𝑅670
𝑅800 + 𝑅670
8. 8
Comparison of different approaches
Reflectance image
of microplot
Average reflectance
of microplot
GF, GAI, Cab
estimates
(Approach 1)
Average over soil and
vegetation pixels
Low spatial
resolution
Average over all
pixels
Vegetation Indices (VI)
calculated for each pixel
VI image
GF, GAI, Cab estimates
(Approach 2)
Thresholding image to
select green pixels
VI image of green
pixels
Average VI over all, dark
and bright green pixels
VI image of green
pixels
High spatial
resolution
9. 9
Methods to separate pixels
Pixels with VARI>0.14
Separation according to
value in the NIR band
10. 10
Results : Green fraction estimation
• Comparison between approach 1 and the best option from approach 2
Approach 1 with VARI Approach 2 with VARI and all pixels
The estimation error
decreases by 20%
11. 11
Results : Green area index estimation
• Comparison between approach 1 and the best option from approach 2
Approach 1 with VARI Approach 2 with VARI and all pixels
The estimation error
decreases by 5%
12. 12
Results : Chlorophyll estimation
• Comparison between approach 1 and the best option from approach 2
Approach 1 with MTCI Approach 2 with mNDblue on the
darkest vegetation pixels
The estimation error
decreases by 26%
13. 13
Key points of this study
• The cm resolution allows us to focus on a subset of pixels
• Standard approaches are outperformed by the proposed method
• A slight improvement for Green area index, much better for Green
fraction and Chlorophyll
• For leaf biochemistry, the possibility to remove soil pixels is likely to
improve the estimations
14. 14
Key points of this study
Best vegetation indices :
• Canopy structure variables are best estimated using 𝑉𝐴𝑅𝐼
• Chlorophyll is best estimated using mNDblue applied on dark pixels
Prospects :
• Better caracterisation of genotypes
• Combining sensor observations with models to improve their performances
• Monitoring of plant growth over time and reaction to stress
15. 15
Acknowledgments
• Sylvain Jay, Ghislain Malatesta and Stéphanie Heno from ITB
• Frédéric Baret and Marie Weiss from INRA
• Alexis Comar and Dan Dutartre from Hiphen