UAV-based remote sensing as a monitoring tool for smallholder cropping systems

Adolfo Posadas2, Dieudonné Harahagazwe2, Arnold Bett1, Elijah Cheruiyot1,2

University of Nairobi

International Potato Center (CIP)

UAV Seminar, Geography DepartmentKenyatta University, 8 March 2016

2

Introduction

Background/history

Project objectives

Community of Practice / Training

Adolfo

Light-plant interaction

Credit: Centre for Remote Imaging, Sensing and Processing (CRISP)

Application examples: Characterization of crops – stress detection

Infection Water deficiency

Crop discrimination Soil characterization

3

Remote Sensing in Agriculture

Why UAV? Low-cost Detailed information No clouds effect

2002

Balloons:• Hot air.• Hellium.

2008

Model planes:• Combustion.• Electric. 2009

Helicopter:

•Combustion.

•Electric. 2012

Multirotor:• Quadcopter.• Octocopter.

ARSP at CIP - History

4

The aim of this “proof of concept” project is to develop and validate a low-cost UAV-basedremote sensing tool for crop area determination (ARSIS) using sweet potato as a pilot crop

Specific objectives include:

• Acquire and test digital commercial, multi- and hyper- spectral cameras, design,acquire, assemble, and test micro-sensors

• Implement methods to remove noise and enhance relevant information fromextremely high resolution imageries obtained by Drone

• Introduce Multi scaling algorithms, based on non-linear approaches, capable ofmeasuring at one scale while probabilistically making inference at other spatial scales

• An out scaling plan that describes a path forward for the validated UAV-ARSIS, as alogical next step of the “Proof of Concept” project.

5

Project Objectives

Key Highlights:

– Costs, accessibility, and user-friendliness

– Involving local institution at different stages is a must

– Stepwise – From simple to complex tools

– Complementarity with satellite imageries

– Multiple crops

– Yield assessment?

– Is it feasible to discriminate varieties?

6

32 Participants:

– National, Regional and International institutions

– 5 CGIAR Centers (CIP, ICRAF, CIAT, IITA and ILRI), and ICIPE

Community of PracticeInception Workshop – October 2014, Nairobi

Unmanned Aerial Vehicle (UAV) – platform

Sensors

Support system – power supply, communication, data storage

Arnold

7

Hardware

8

Commercial UAVs

UAV Platforms

Octocopter assembled in NairobiEstimated cost: $ 5000

Quadcopter assembled in NairobiEstimated cost: $ 4000

MikroKopter’s Electronic System$740

9

Ardupilot’s Electronic System$250

Quadcopter Open Source

UAV Platforms

Quadcopter Open Source designed and assembled locally based on ArdupilotEstimated cost: $ 500

10

Reducing Camera Cost While Improving Image Quality

NDVI with TETRACAM ADC Micro NDVI with CIP Built CameraRGB Original Image – Canon EOS

Sensors

11

Active Sensors Under Construction

Sensors

Fluorescence sensor

Photochemical Reflectance Index (PRI) sensor under construction

12

Locally Assembled Multispectral Acquisition System

Acquisition Multispectral System Assembled

1. Power Supply System

2. Mini PC (pcDuino)

3. Cameras CCD (Red & NIR)

4. LiPo Battery Holder 1

2

3

4

Data Acquisition System

UAV assembled in collaboration with ICRAF and UoN

13

UAV Assembly and Testing

UAV used in the first field mission in Mwanza – Tanzania in April 2015

14

Capacity to Repair Broken Down UAVs - Demonstrated

UAV Repair and Maintenance

FixedCrashed Octocopter

Fixed3D-printed Quadcopter crashed

15

Training Capacity

UAV Demonstration at the 6th

SPHI Annual Conference - Kigali

Training at Ministry of Agriculture - Rwanda

Spectral data acquisition

Image mosaicking

Texture-aided supervised classification

Elijah

16

Software/Processing

17

SpectraCIP Software

USB2000+VIS-NIR-ES Spectrometer

Spectral Data Acquisition

Commercial Version: SpectraSuite

SpectraCIP: Advantages• Open-source• Easy to store data – handy

in the field

18

Open Source Image Stitching Software: ISAM-CIP

Mosaic producedSoftware mosaicks two or more images by

identifying identical points in the overlap area.

Mosaicking

19

Texture-based Maximum Likelihood Classification of RapidEye Satellite Image

Cropping areas identified from satellite images at 6.5 m resolution

Cropping Area Identification at District Level

• Additional texture information significantly improves discrimination results

• Classified satellite image aids identification of UAV data collection sites

20

Classification Results and Error Matrix

Cropping Area Identification at District Level

Error Matrix:• Overall Classification Accuracy =

(4202/4964) 84.6495%• Kappa Coefficient = 0.8106

Class Feature Colour Number of

sample points

Number of

feature points

Percentage

cover

Area

(Hectares)

Unclassified [Black] 32,562,557 27.203 146706.04

Maize [Red] 33 1,244,964 1.040 5609.01

Trees [Sea Green] 36 9,014,849 7.531 40615.14

Shrubs [Green] 42 1,998,269 1.669 9002.92

Road_murram [Magenta] 40 7,631,747 6.376 34383.77

Road_tarmac [Purple] 36 2,209,397 1.846 9954.13

Clouds [Cyan] 66 12,263,029 10.245 55249.36

Tilled_land_soil [Maroon] 42 780,023 0.652 3514.28

Buildings [Sienna] 31 51,869,227 43.332 233689.54

Bananas [Blue] 11 111,303 0.093 501.46

Sweetpotato [Yellow] 12 15,339 0.013 69.11

Reference

Cla

ssif

icat

ion

Class Maize Trees Shrubs Murra

m

Tarma

c

Clouds Tilled

land

Buildin

gs

Banan

as

Sweet

potato

Total

Maize 93 0 0 0 0 0 0 0 1 0 94

Trees 0 1304 0 0 0 0 0 0 6 0 1310

Shrubs 2 0 689 0 0 0 0 0 0 1 692

Murram 29 0 0 238 0 0 1 0 0 0 268

Tarmac 0 0 0 0 292 0 0 0 0 0 292

Clouds 0 0 0 0 2 1255 0 9 0 0 1266

Tilled land 0 0 0 0 0 0 89 0 0 0 89

Buildings 15 0 38 299 19 314 5 218 1 3 912

Bananas 6 2 4 0 0 0 0 0 12 0 24

Sweetpot

ato

0 0 5 0 0 0 0 0 0 12 17

Total 145 1306 736 537 313 1569 95 227 20 16 4964

21

Texture-based Classification of UAV Image

Better classification results could be obtained with multispectral imageries, but this require special equipment and knowledge

Crops identified at farm level from images taken with <5 cm resolution using a regular digital camera

Crop Discrimination at Farm Level

Multifractal analysis

Spatial scaling

Adolfo

22

Non-linear Processing

23

Non-linear Processing Methods

14 Sweetpotato varieties at LZARDI experiment, Mwanza Optical spectra

Multifractal spectra

Processing: Variety Discrimination

24

Way Forward

• Develop a UAV Regional Hub in Nairobi:

– Training on UAV hardware and software development

– Training on UAV applications

• Engage in advocacy to improve an enabling environment for the use of UAVs for development in the region

• Etc.

Videos:

• AgroTV–16 – Spanish;

• ARSIS Video – English;

Blog Posts:

• Invasion of the Potato Drones – Lima;

• Community of Practice – SSA;

• UAV Assembling – SSA;

• Crop Discrimination;

• Airborne Agriculture

Photos:

• Testing the Unmanned Aerial Vehicle – Flickr;

Join UAV Community on Online Platforms:

• UAV4Ag on DGroups;

• @UAV4Ag on Twitter.

25

Learn More!

Thank you!

Contacts:Roberto Quiroz, PhD

R.Quiroz@cgiar.org

International Potato Center (CIP)

Q/A

Contributing authors:Adolfo Posadas

Arnold BettElijah Cheruiyot

Dieudonné HarahagazweHildo Loayza

Susan PalaciosMario Balcázar

Luis SilvaRoberto Quiroz