ESTIMATION CROP PHYSICAL PARAMETERS FROM UAV RGB IMAGERY AND DEEP LEARNING

Maria Polinova, Keren Salinas, Anna Brook

Spectroscopy and Remote Sensing Laboratory, The Department of Geography and Environmental Studies, University of Haifa, Israel

Department of Geography and Environmental Studies

Irrigation management

Inputs

Crop

Field parameters

Weather

Irrigation schedule

Field feedback (crop parameters)

Crop growth guides

Study Area

Data collectionOceanOptics USB4000-VIS-NIR

DJI Phantom 4 Professional Flight planning Image processing Orthophotomosaic

Point measurements (20-50 per plant) Spectra 400-1000nm (accuracy 1nm)

Representative crops

Hight patches (“health”) Mixed patches Low patches (“stressed”)

Neural Network. Stage1 – Full Spectral Resolution

Neural Network. Stage2 – Resampled Spec 2 RGB

Neural Network. Results1– estimation accuracy

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10 11 12 13

L2 Cotton 1 – Full Spec

H L

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10 11 12 13

L2 Cotton 1 – Sprc2RGB

H_spec_RGB L_Spec_RGB

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10 11 12 13

L2 Cotton 1 – RGB image

H_image_RGB L_image_RGB

Crop1 – 1.30 m

Crop2 – 1.41m

Crop3 – 1.36m

Crop4 – 1.36m

Crop5 – 1.64 m

Crop6 – 1.59m

Crop7 – 1.46m

Crop8 – 0.90m

Crop9 – 1.47 m

Crop10 – 1.39m

Crop11 – 1.51m

Crop12 – 1.43m

Crop13 – 1.56m

Pre-processed RGB image

Neural Network. Results2– field scaleCrop class H

5 probability levels0.94–0.950.95–0.960.96–0.970.97–0.98

0.98–1

max prob 0.9889

Crop class L

7 probability levels0.6–0.80.8–0.90.9–0.950.95–0.960.96–0.970.97–0.98

0.98–1

max prob 0.992

Perspective of application

Biomass (g)60 180

Total chlorophyll content (mg g-1)

0.13 3.21

No water stress

Water stress

Leaf area index (m2 /m2)

0.10 2.37

Cotton field with disease

Our projects

Past Present Future

Topic 2.1.3: Irrigationtechnologies andpractices

Low-cost irrigationsystem web-DSS basedon agronomist practice(joint Italian-Israeli R&Dprojects)

Solution for high-throughputand low-costphenotyping of wholeplants in realisticagricultural settings(national R&D projects)

Optimization irrigationschedule by weathershort-term weatherforecasting and fieldspectroscopy (fundedby Israeli WaterAuthority)

Thank you for attention!

For more info please contact:

Maria Polinova - polinovamaria88@gmail.com

Keren Salinas - kerensalinas2@gmail.comAnna Brook - abrook@geo.haifa.ac.il

Department of Geography and Environmental Studies