final project overview contact: mzude@atb-potsdamict-agri project “3d-mosaic” final project...

3DMOSAIC intro plattform vision system fruit sensor field trials GIS PRECISION FRUTICULTURE

intro plattform vision system fruit sensor field trials GIS PRECISION FRUTICULTURE

ADVANCED MONITORING OF TREE CROPS FOR OPTIMIZED MANAGEMENT ICT-AGRI PROJECT “3D-MOSAIC”

Final project overview

Contact: [email protected]

3DMOSAIC



Kick off meeting: June 14-15, 2011 Potsdam/Berlin, ATB

Partners 3DMOSAIC



• Transnational Project ICT-AGRI ERANET, FP7

• Coordinator: Leibniz Institute for Agricultural Engineering Potsdam-Bornim (ATB)

• Funding: Ministries and Agencies from participating partner countries (1.1 Mio Euro)

• Duration: 05/2011 - 04/2013

• Partners: 11 partners from 7 countries (EU member and AC)

www.atb-potsdam.de/3D-MOSAIC

Conditions

http://www.atb-potsdam.de/3D-MOSAIC







4

3D-Mosaic targets …

… the monitoring of spatial variability in tree crops and delineation of management zones considering soil and plant data by means of ICT and robotic solutions for approaching the automation of horticultural processes

Concept



Inst

rum

en

t im

ple

me

n-

tati

on

& T

est

ing

Dat

a m

anag

em

ent

&

Ap

plic

atio

n

Project structure



WP 1 – Autonomous Platform

Partners: University of Hohenheim (DE), Universität Kassel (DE), Zürcher Hochschule für angewandte Wissenschaften (CH), Aristotle University of Thessaloniki (GR)

Contact: Claes L. Dühring Jæger University of Hohenheim [email protected]

Competencies Precision Farming Robotics and Automation Navigation Data analysis and Simulation Real Time Applications



WP 1 – Autonomous Platform Armadillo Scout: Electrical powered robot for agricultural applications like scouting and weeding. It has a modular design which makes it easy to adjust height and width.

• Battery Type: LiFeYPO4 • Battery Capacity: 160 Ah • Primary Voltage: 48 VDC • Secondary Voltage: 12/24 VDC • Weight: 430 kg • Operation time: 11 h • Power: 7 kW • Average Power Consumption: 450 W • Computer: Frobobox • Controlling Software: MobotWare /

FroboMind Armadillo Scout



WP 1 – Autonomous Platform In the 3D-Mosaic, an algorithm was developed for mapping trees in an orchard. The algorithm is currently being implemented in a SLAM solution for our platforms

Armadillo with Sensor Tower equipped with vision systems



WP 2 – Vision Systems

Partners: Agroscope Reckenholz-Tänikon ART Research station (CH), Universität Kassel (DE), Zürcher Hochschule für angewandte Wissenschaften (CH)

Contact: Thomas Anken, ART [email protected]

Competencies of work groups 2D leaf and fruit analyses Plant recognition system (e.g. SmartWeeder) Real Time 3D data acquisition in high resolution Real Time 3D plant recognition



CCD camera and LiDAR were mounted with vertical top-down viewing direction Heights: 2.7 m (CCD) and 3.5 m (LiDAR) Vision NIR GigE cameras were applied, providing 1600x1200 pixels, 8bit intensity images

In the 3D-Mosaic, systems were integrated on the platform for data acquisition and software tools were developed for canopy analyses.

WP 2 – Vision Systems



WP 3 – Fruit Information

Partners: Politecnico di Milano (IT), Leibniz Institute for Agricultural Engineering Potsdam-Bornim, ATB (DE), Sintéleia S.r.l. (IT), Aristotle University of Thessaloniki (GR)

Contact: Alessandro Torricelli, Politecnico di Milano [email protected]

Competencies of work groups (Commercial) multispectral fruit sensors Fluorescence lifetime imaging and spectroscopy of diffusive media Physical models for photon transport in biological material



WP 3 – Facilities time-domain workstation for diffuse spectroscopy of turbid media in the 600-1100 nm spectral range based on super-continuum laser and time-correlated single photon counting

In the 3D-Mosaic, software for fitting of photon transport in biological samples with one or more layers. Novel sensors for measuring fruit water content were approached.



WP 3 – Background



WP 4 – Field tests

Partners: Leibniz Institute for Agricultural Engineering Potsdam-Bornim, ATB (DE), University of Cukurova (TR), all partners

Contact: Manuela Zude, ATB [email protected]

Competencies Fruit physiology Mechanical and optical properties of fruits in-situ fruit analysis by means of sensors



Experimental orchards

Prunus domestica Marquardt, Germany

Citrus paradisii Adana, Turkey



Field trial 1 intro plattform vision system fruit sensor field trials GIS PRECISION FRUTICULTURE

Citrus paradisii Adana, Turkey



Variation of evapotranspiration values measured by Bowen ratio-energy balance method (BREB) at the period from DOY:249 (2011) to DOY:97 (2012)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

13

.09

.20

11

19

.09

.20

11

25

.09

.20

11

01

.10

.20

11

07

.10

.20

11

13

.10

.20

11

19

.10

.20

11

25

.10

.20

11

31

.10

.20

11

06

.11

.20

11

12

.11

.20

11

18

.11

.20

11

24

.11

.20

11

30

.11

.20

11

06

.12

.20

11

12

.12

.20

11

18

.12

.20

11

24

.12

.20

11

30

.12

.20

11

05

.01

.20

12

11

.01

.20

12

17

.01

.20

12

23

.01

.20

12

29

.01

.20

12

04

.02

.20

12

10

.02

.20

12

16

.02

.20

12

22

.02

.20

12

28

.02

.20

12

05

.03

.20

12

11

.03

.20

12

17

.03

.20

12

23

.03

.20

12

29

.03

.20

12

04

.04

.20

12

Date

Ev

ap

otr

an

sp

ira

tio

n_

Bo

we

n, m

m/d

ay

Total ET=345.3 mm

Citrus paradisii – Environment



18

Field trial 2

Prunus domestica Marquardt, Germany




Field trial 2 …automatical and not so automatical data acquisition

ww

w.a

tb-p

ots

dam

.de

/3D

-MO

SAIC



Soil pattern stable over 3 years

Small scale variability due to sand lenses



Plum – Statistics Plant parameter ECa

topsoil Elevation

F p < F F p < F

Fiv

e y

ears

old

tre

es Flower set 2.25 0.037 2.47 0.023

Fruit set 1.15 n. s. 1.21 n. s.

Fruit drop 0.91 n. s. 4.63 0.002

# Fruits 0.72 n. s. 0.33 n. s.

Fruit height 7.02 < 0.001 5.31 < 0.001

Yield 0.33 n. s. 0.34 n. s.

Fruit NDVI 0.75 0.594 1.65 0.184

DA-index 1.36 0.264 2.34 0.074

Six

years

old

tre

es

Flower set 3.78 0.007 2.63 0.038

Fruit set 3.46 0.011 1.80 0.133

Fruit drop 3.78 0.007 0.47 n.s.

# Fruits 9.75 0 0.58 n. s.

Fruit height* 10.72 < 0.001 8.20 < 0.001

Yield 1.84 0.128 0.65 n. s.

Fruit NDVI 0.46 0.800 1.72 0.184

DA-index 0.70 0.630 0.98 0.441

1



Spatial pattern of fruits per tree and soil electrical resistivity, R = 0.46 (2011)

Spatial pattern of soil and yield

1

10

20

30

40

50

Number



Right side drive 137

137

137 Top view

Vision system: LiDAR and camera

Left side drive



.

(a) (b) (c)

Figure 1 – Processing steps for the determination of the per cent leaf cover of two exemplary plum tree

images taken under different lightning conditions: (a) unprocessed image; (b) fragmentation into

touching area objects with the help of a watershed procedure; (c) segmented binary image

Vision system: LiDAR and camera

(a) unprocessed image

(b) fragmentation using watershed procedure

(c) segmented binary image



Platform: Software Configuration



Tree Mapping using LiDAR & GNSS

Calculated tree positions, marked with green +, and the path driven by the tractor, marked with blue o. Axis are in meters, (0,0) is equivalent to UTM zone 36 S; N 4100245.8869 E 711859.8143

Using border data of trees – detected by histogram - and direct least square ellipse fitting on resulting point cloud, the position of each tree can be calculated.



Fruit sensors / information

GOAL Non-destructive, continuous monitoring of fruit parameters in the orchard aiming at optimising cultivation processes METHOD Optical multispectral sensors in the canopy PROBLEM Varying scattering coefficient in the growing fruit SOLUTION Physical calibration correction of VIS/NIR optical sensors by means of advanced optical techniques and models



Novel sensors – laboratory data



Novel sensors in the orchard

2. Fruit sensor “spider” equipped with Xbee Pro 2.4GHz RF modules for data transfer 1. Fruit sensor “DA-Meter” for manual use in

the orchard



Novel sensors in the orchard



Data transfer of fruit sensor „spider“: Radio path loss

Wave propagation modes in the presence of trees:

• around tree canopies - diffracted wave (black)

• directly through tree foliage - scattered wave (blue)

• ground reflected wave (red)



• The prediction of radio path loss and ultimately the maximum range among the nodes of wireless sensor networks is critical for the successful deployment of this technology in orchards.

• Our goals were to:

1. Evaluate the influence of leaves on electromagnetic attenuation at various transmission heights.

2. Compare the prediction power of empirical radio propagation models

3. Compare empirical and computational models.

32

Radio path loss



33

Empirical models were used to predict propagation losses through vegetation.

There was significantly higher attenuation at 1.5 m when leaves were present.

The MED, ITU-R, and FITU-R could not account for the presence of leaves.

More advanced computational modelling is needed.

Radio path loss



34

InertiaCube3 3-DOF orientation tracking sensor

Euler angles (roll, pitch, yaw) of main tree branches

Length and circumference of branches

Digitization of cherry tree geometry with computer software using the collected data as input data for all executed experiment scenarios.

COMSOL MULTIPHYSICS: EM solution for simplified case.

Geometric tree models



GIS

Tree Unit Plant Life BBDD

Geo-positioned Sensor Data

Management Zones Expert Analysis

OLAP Analysis

3D-Mosaic Graphic

DSS 3D-Chart

Farmer

Tile organization

WP 5 – Data management



Management Zones Map

Farmer Tile organization

Raw sensor data -Tree

- Leaf Area - Canopy - Fruit count - Fruit quality

-Soil measurements - ECa,..

-Weather station - Temperature - Humidity, ..

-Irrigation - L/m², ..

Expert

Spatial Analysis

Irrigation Map

Soil Map

Trees + cover

WP 5 – Layer structure



3D Graphic User Interface

OLAP Analysis

Decision Support System

Google plug-in

Dashboard

Measurement Data GUI: • Multidata web upload.

• Object (tree, MZ…) & Dimensions (ECa,

yield…) dynamic representation.

• GIS web interface.

Features: • Correlation & dependences

between measures

• Dispersion Analysis

In 3D-Mosaic, the commercial application were extended by algorithms for hotspot analysis to enable DSS

WP 5 – DSS Tool



WP 6 – Management Zones Goal: A methodology for defining management zones in orchards based on spatial statistical analysis Method: Spatial variability can be used as the basis for delineating agricultural fields into sub-units know as management zones that exhibit homogeneity in environmental or plant parameters that are considered yield-defining Problem: Existing clustering algorithms typically rely on the data structure to recognize natural groupings (clusters) and to partition n data observations into k clusters based on methods of similarity. However, while the location of data in the character space is taken into consideration, its geographical spatial context (spatial neighborhood) is not accounted for. Solution: The proposed method is based on quantifying the distribution in space of attributes associated with trees or with environmental parameters and evaluating whether or not, and to what extent, a recognized pattern is significantly clustered or dispersed



WP 6 – SDSS

BUILDING A GEOGRAPHICALLY WEIGHTED REGRESSION MODEL for field trial 2; Basemap from ESRI World Map Background, Copyright © 1995 - 2012 ESRI (ESRI 1982)

Trunk circumference + ECa vs. yield

R2 = 0.57 R2 = 0.59

Management zones Based on local R2

Trunk circumference vs. yield



Spatial correlation analysis such as a geographically weighted regression (GWR) will be applied using the data collected in field trial II to recognize the parameters that most influence and/or represent yield variability

3D-Mosaic: The developed methodology captures (i) evaluation of global spatial variability in the orchard using the General G statistic, (ii) detection of local spatial clusters using the Gi* statistic (hot-spot analysis), and (iii) delineation of management zones based on the identified clusters

WP 6 – Achievements & Outlook



Conclusion Consistent with findings in precision agriculture of field crops, correlation was found between soil electric conductivity and plant parameters. In preliminary experiments in the subtropics, interactions of soil and vegetative growth, yield, and fruit quality were indicated. The data were supplied by means of novel sensors or new approaches in data processing, providing the necessary prerequisites for numerous spatially resolved measurements.

Finally some notes from (i) a FARMER perspective: “farming with sensors is so much easier”; and (ii) the external project advisor: “the level of enthusiasm was infectious”

I feel that by the bringing facilities together we, were able to provide an integrated approach that benefits the AUTOMATION of agricultural processes and supports the PRECISION FRUTICULTURE concept .

Hope to see you soon on similar questions,

Manuela Zude



a)Pforte F, Selbeck J, Hensel O. (2012). Comparison of image analysis and laser scanner technique for automated estimation of the leaf area on tree level. Biosystems Engineering Volume 113, Issue 4, 325-333 b)Vougioukas S; Anastassiu HT; Regen C; Zude M (2012). Influence of foliage on radio path losses (PLs) for wireless sensor network (WSN) planning in orchards. Biosystems Engineering 114: 454-465 c)Zude M; Peeters A; Selbeck J; Käthner J; Gebbers R; Ben-Gal A; Hetzroni A; Jaeger-Hansen C; Griepentrog HW; Pforte F; Rozzi P; Torricelli A; Spinelli L; Ünlü M; Kanber R (2012). Methoden für die präzise obstbauliche Produktion / Advances in precise fruit production. Landtechnik / Agricultural Engineering 67 (5): 338–341 d)Lorente D; Zude M; Regen C; Palouc L; Gómez-Sanchis J; Blasco J (2013). Early decay detection in citrus fruit using laser-light backscattering imaging. Postharvest Biology and Technology, in press e)Mollazade K; Omid M; Akhlaghian T; Rezaei Y; Mohtasebi KS; Zude M (2013). Analysis of texture-based features for predicting mechanical properties of horticultural products by laser light backscattering imaging. Computers and Electronics in Agriculture, in press f)Torricelli A;Spinelli L; Kaethner J; Selbeck J; Franceschini A; Rozzi P; Zude M (2012). Non-destructive optical assessment of photon path lengths in fruit during ripening: implications on design of continuous-wave sensors. CIGR-AgEng International Conference of Agricultural Engineering, Proceedings 84-88. ISBN-10: 84-615-9928-4 g)Hetzroni, A., Peeters, A., Ben-Gal, A. (2012). Towards precision management of orchards: using automated monitoring to build a GIS-based spatial decision support system. International Conference of Agricultural Engineering, Proceedings 73-78. ISBN-10: 84-615-9928-4 h)Thomas Anken, Andrea Battiato, Dejan Seatovic, Vincent Meiser, Jörn Selbeck, Florian Pforte. “Canopy-Area Measurement of Plum Trees using Laser and Near-Infrared Imaging” EFITA-WCCA-CIGR Conference “Sustainable Agriculture through ICT Innovation”, Turin, Italy, 24-27 June 2013, p 861-868 i)Jana Käthner, Werner Herppich, Rolf Adamek and Manuela Zude: Influence of soil variability and topography on plant growth and yield parameters in Prunus domestica orchard, EFITA-WCCA-CIGR conference, 23-27 June, 2013 Turin, Italy ………. ……….

Literature – please visit webpage



This work was supported by the ICT-AGRI project “3D-Mosaic – Advanced Monitoring of Tree Crops for Optimized Management – How to Cope with Variability in Soil and Plant Properties?” which is funded by the European Commission’s ERA-NET scheme under the 7th Framework Program for Research. We thank the contributing funders group from seven countries.

Acknowledgement

final project overview contact: mzude@atb-potsdamict-agri project “3d-mosaic” final project...

Documents