Processing of Mandarin Leaf Multispectral Re fl ectance Data for the Retrieval of

Processing of Mandarin Leaf Multispectral Reflectance Data

for the Retrieval of Leaf Water Potential Information

Janos Kriston-Vizi PhDKyoto University

Acknowledgement

This research was conducted by financial support of Japanese Society for Promotion of Science (JSPS).

Dr. Kumi Miyamoto senior researcherWakayama Research Center of Agriculture, Forestry and FisheriesFruit Tree Experiment Station

Professor Mikio UmedaKyoto University, Laboratory of Filed robotics and Precision Agriculture

source: Yakushiji, H. et al. (1996):

1. Water stress induce sugar accumulation in mandarin fruit

2. Mulching induce water stress

3. Japanese mandarin farmer: „Leaf reflectance indicates water stress”…

Physical and Physiological Background

Sugar and Acid Content Change due to Water Stress – Japanese Local Growers

Sugar content [degrees Brix]

Acid content [%]

Orchard properties

Place

Variety

Sugar and Acid Content Change due to Water Stress – Experimental Orchard

°Brix Acid [%]

Control tree 1. 11.1 0.8

Control tree 2. 11.1 0.7

Control tree 3. 11.1 0.7

Mulched tree 1. 13.3 1.1

Mulched tree 2. 13.1 1.3

Mulched tree 3. 13.6 1.3

Satsuma Mandarin (Citrus unshiu Marc. var. Satsuma) rootstock and variety: Miyagawa Wase Mulch: plastic cover with DuPont Tyvek

Wakayama Research Center of Agriculture, Fruit Tree Experiment Station (near Osaka)

Experimental FieldData Collection Equipments

Silvacam multispectral digital video camera

490-580 nm Green580-680 nm Red760-900 nm NIR

Pressure Chamber made by Pms Instrument Company, Model 600

GNU/Linux capture and non-linear DV editor software

1. Capture data from MiniDV to .dv file

2. Export .dv file to .png image sequence

Capture and export process

Capture by Kino video: 1_53s_mpeg1_Kino_demo_xvidcap_screen-video_capture_HDV.mpeg

NIR R G

760 - 9

00 n

m

490 - 5

80 n

m

580 – 680 nm

Silvacam false color image and bands

http://rsb.info.nih.gov/ij/index.html

Advantages:

customizable, open source

code many algorithms available free

Linux image processing program

2. Customized java script for SegmentingAssistant plugin to be able to segment image sequence

1. Customizable, free software

SegmentingAssisstant plugin

1. Setting segmenting parameters for image sequence

2. Automatically segmenting image sequence

Segmentation workflow

Automatized segmenting process video:2_10s_mpeg1_ImageJ_SegmentingAssisstant_XVidCap_screenshot_video_2005-11-24_coT2L1.mpg

Result file after analyzing an image sequence

NIRframe 1 R

GNIR

frame 2 RG

etc.

Python script to format ImageJ output file and preprocessing for statistical analysis: calculate abs. reflectance

1. Boxplot for initial comparison

Statistical analysis

2. Histogram, Kernel Density Estimates and Stem-and-leaf chart to find outliers

Rank experiments by box and whiskers plot - 2003

Rank experiments by box and whiskers plot - 2006

F p-value t df p-value Mean Diff. Diff.St.Err.2003. VIII. peach A 2,218 0,143 -6,543 46 0.000 -3,12 0,11 G refl. [%] B -8,091 29,657 0.000 -3,12 0,072003. VIII. peach A 0,648 0,425 -1,865 46 0,069 -1,17 0,47 R refl. [%] B -2,134 24,636 0,043 -1,17 0,382003. VIII. mandarin A 13,055 0,001 -4,367 44 0.000 -2,49 0,57 G refl. [%] B -3,125 12,656 0,008 -2,49 0,792002. XI. mandarin A 0,863 0,356 -1,864 62 0,067 -1,48 0,79 G refl. [%] B -1,597 20,644 0,125 -1,48 0,92

Levene-test Significance testing

G refl. – 490-580 nmR refl. – 580-680 nm

A – assume equal variancesB – assume non-equal variances

Reflectance of mulched leaves are higher than reflectance of control leaves.

Significance Testing – Reflectance Difference

between control and mulched leaves - 2003

Mean differenceIX.27. peach LWP 1.96 MPaIX.27. peach G refl. -4.66%IX.27. peach R refl. -4.42%IX.28. mandarin LWP 0.35 MPaIX.28. mandarin G refl. -2.60%IX.28. mandarin R refl. -1.55%IX.29. mandarin LWP 0.48 MPaIX.29. mandarin G refl. t-test not passedIX.29. mandarin R refl. -2.32%IX.30. peach LWP 0.91 MPaIX.30. peach G refl. t-test not passedIX.30. peach R refl. -1.91%XI.24. mandarin LWP 1.04 MPaXI.24. mandarin G refl. -3.39%XI.24. mandarin R refl. -4.02%XI.25. mandarin LWP 0.99 MPaXI.25. mandarin G refl. -1.83%XI.25. mandarin R refl. -1.52%

Significance Testing – Reflectance Difference between control and mulched leaves - 2005

formula: LWP=IX.27. peach G refl.-LWP 0.29*(-0.26) G refl. 0.41IX.27. peach R refl.-LWP -0.79*(-0.23)R refl. 0.33IX.28. mandarin G refl.-LWP -0.43*(-0.09)G refl. 0.35IX.28. mandarin R refl.-LWP -0.79*(-0.12)R refl. 0.40IX.29. mandarin G refl.-LWP t-test not passed -IX.29. mandarin R refl.-LWP -1.7*(-0.10)R refl. 0.40IX.30. mandarin G refl.-LWP t-test not passed -IX.30. mandarin R refl.-LWP -0.98*(-0.20)R refl. 0.22XI.24. mandarin G refl.-LWP -0.02*(-0.20)G refl. 0.51XI.24. mandarin R refl.-LWP -0.71*(-0.17)R refl. 0.45XI.25. mandarin G refl.-LWP -0.53*(-0.14)G refl. 0.17XI.25. mandarin R refl.-LWP -1.22*(-0.096)R refl. 0.12

R2

Linear regression results: equations - 2005

LWP = -0.02 • (-0.2)G refl.Multiple R2: 0.51p = 1.15e-08

LWP = -0.71 • (-0.17)R refl.Multiple R2: 0.53p = 3.76e-09

Linear regression results: plots – 2005

Linear regression results: plots – 2003peach

peachLWP = 0.19 • ( - 21.02 )G refl.Multiple R2: 0.63

2002. XI mandarin G-LWP

.18.16.14.12.10.08.06.04.02

-.5

-1.0

-1.5

-2.0

-2.5

-3.0

-3.5

-4.0

-4.5

várható

gyakoriságok

megfigyelt

gyakoriságok

Linear regression results: plots - 2002

LWP = - 0.19 • ( - 21.02 )G refl.Multiple R2: 0.29

2002. XI mandarin R-LWP

.10.09.08.07.06.05.04.03.02

-.5

-1.0

-1.5

-2.0

-2.5

-3.0

-3.5

-4.0

-4.5

várható

gyakoriságok

megfigyelt

gyakoriságok

Linear regression results: plots – 2002

LWP = - 0.45 • ( - 29.15 )R refl.Multiple R2: 0.28

Whole Mandarin Orchard Image Segmentation – manual

2005. 09. 29. 10h

Manual segmentation by ImageJ:

Green channel,threshold intensity for ROI pixels = 30-70

Whole Mandarin Orchard Image Segmentation – automatic

4 class k-means canopy segmentation of multispectral orchard image

Infrared thermography

Objective: Find optimal conditions to detect water stress by infrared thermography.Hardware tool: Avio Nippon Avionics, Neo Thermo TVS-600

Thermal image on whole mandarin orchard image - 2005

LWP difference between mulched and control area: Mulched area: -2.552 MPa Control area: -2.071 MPaMean difference: 0.481 MPa

Temperature difference between mulched and control area: Mulched area: 29.2 °C (mean) Control area: 26.4 °C (mean)Mean difference: 2.8 °C

2005. 09. 29. 10h

Need large (6-8 rows) area to detect temperature difference.

Thermal image on whole mandarin orchard image - 2006

Temperature difference between mulched and control area: Mulched area: 28.9 °C (mean) Control area: 26.8 °C (mean)Mean difference: 2.1 °C

2006. 09. 27. 11:15h

Current work and near future research plan

Hyperspectral reflectanceObjective: Find optimal bandwith at visible range to detect LWP, that narrower than R,GHardware tool: Specim Imspector with Hamamatsu camera (400-1000 nm)

Current work and near future research plan

Severe water stress effect on peach leaves’s reflectance at visible spectral range. LWPs mu: -4.0 MPa

co: -0.9 MPa

PhD: 2005 (age of 29) Hungary, Corvinus University of Budapest Crop Sciences and Horticulture

Research: 2002 - Present Kyoto University, Japan Precision Agriculture Mandarin Water Stress

Author’s Bio

Thank you for your attention