exploring the potential for electronic grading of fruit
TRANSCRIPT
1
Exploring the Potential for
Electronic Grading of Fruit with Canker Lesions
Dr. Tom BurksAssociate Professor
Agricultural and Biological Engineering Dept.University of Florida
46th Annual Packinghouse Days
Citrus Research and Extension CenterUniversity of Florida
2
Progress Report
Projected funded by the USDA-TASC through Florida Fresh Packers, Assoc.Project team
Dr. Tim SchubertDr. Gordon BonnDr. Mark Ritenour
3
Research ObjectiveInvestigate various sensing techniques for detecting citrus canker and differentiating it from other common confounding diseases.
Develop On-line detection technologies that can be commercialized for eliminating cankerous fruit from the packing line.
4
First Year Objectives
Determine spectral characteristics of citrus canker and confounding diseases.Identify candidate sensing technologies which could be implemented for commercial packingline grading systemsDevelop relationship with Equipment manufacturer and acquire prototype one-lane system
5
Sample Collection and Preparation
Identified source of Grapefruit which exhibited canker and other typical disease conditionsCollected samples from field in Spring 2007Sorted, washed, sanitized, waxed at Indian River REC on their packingline under supervision of Dr. Ritenour, FMC, and a local packer.Transported to UF-Gainesville under advise of FDACS-DPI.
6
Citrus Samples & Skin ConditionsGrapefruits (Ruby Red), 2006/07 Harvest Season
Washed + ChlorineWashed + Chlorine
Washed + Chlorine + SOPP + Wax
Washed + Chlorine + SOPP + Wax
Washed + Chlorine + SOPP
Washed + Chlorine + SOPP
Copper Burn
Greasy Spot
Wind Scar
Cake Melanose
Specular Melanose
Insect damage
UnwashedUnwashed
Field Fruit with Other DiseasesField Fruit with CankerMarket Fruit
7
Citrus Data CollectionFour different vision-based systems:
Reflectance spectra using spectrophotometer ***
Color images using RGB color camera
Visible & near-infrared multispectral images
Hyperspectral reflectance & fluorescence images
8
Objectives of Spectral Reflectance Studies1. To acquire the spectral reflectance of citrus canker and other conditions.
2. To identify the significant wavelengths that have the maximum discriminatory potential.
3. To derive discriminant function that could classify citrus canker from other conditions.
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800700
600500
400
300200
Wavelength(nm)
Ref
lect
ance
(%)
BaselineCanker MosaicMarket Fruit
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800700
600500
400
300200
Wavelength(nm)
Ref
lect
ance
(%)
BaselineCanker MosaicMarket Fruit
Machine Vision Grading System
CANKER
• RGB camera• NIR camera• Multispectral Camera• Hyperspectral Camera
Machine Vision Grading System
CANKER
• RGB camera• NIR camera• Multispectral Camera• Hyperspectral Camera
Main GOAL
9
Market fruit Fruit with canker Fruit with greasy spot
Fruit with copper burn Fruit with wind scar Fruit with cake melanose
Fruit samples
10
Sample holder
Spectrophotometer- Measures spectral reflectance from 200nm to 2500nm
CARY Spectrophotometer
2 inches
11
Market fruit peel
Baseline material
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
100090
080
070
060
050
040
030
020
0
Wavelength(nm)
Ref
lect
ance
(%)
Baseline
Market Fruit
Sample reflectance result from spectrophotometer
12
1. Color image acquisition of whole fruit
2. Preparation of peel sample in holder
3. Sample with spectrophotometer
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900800
700600
500400300200
Wavelength(nm)
Ref
lect
ance
(%)
Baseline
Canker MosaicMarket Fruit
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900800
700600
500400300200
Wavelength(nm)
Ref
lect
ance
(%)
Baseline
Canker MosaicMarket Fruit 4. Statistical analysis to identify wavelengths with
high discriminant potential
Flowchart for spectral reflectance acquisition an analysis
13
Different faces of canker
• different distribution of canker infection• spectrophotometer averages thearea of the sample holder• canker “mosaic” was prepared
14
0
10
20
30
40
50
60
70
80
90
100
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800700600
500400
300200
Wavelength(nm)
Ref
lect
ance
(%)
BaselineCanker Mosaic
Market Fruit
Visible
Visible/NIR
Whole spectrum
Statistical analysis of reflectance data
3 Spectral Ranges
1. Visible (400-750)
2. Visible/NIR (400-1100)
3. Whole spectrum(400-2500)
15
0
10
20
30
40
50
60
70
2502300210019001700150013001100900700500300
Wavelength(nm)
Ref
lect
ance
(%)
MarketCankerWindscarCake MelanoseCopper BurnGreasy Spot
Spectral reflectance of the different conditions
• Different spectral characteristics • Potential for separation
16
Interpreting Initial Results
These results are encouraging although caution must be used.
Disease sample sizes were largeSpectrophotometer results are not always implementable under different light sourcesGood results required 7 to 10 bands
Results provide a starting point for on-line system development, more detailed experiments in multi-spectral and hyper-spectral detection.
17
0
10
20
30
40
50
60
70
25002300210019001700150013001100900700500300
Wavelength(nm)
Ref
lect
ance
(%)
Late Market SampleEarly Market Sample
0
10
20
30
40
50
60
25002300210019001700150013001100900700500300
Wavelength(nm)
Ref
lect
ance
(%)
Late CankerSample(Average)Early CankerSample(Average)
Sampling DatesEarly : March 12, 2007Late : April 16, 2007
• No difference was observed with market fruit and other conditions
• There was a shift of the reflectance curve of the canker samples
Comparison of samples taken on different dates
18
0
10
20
30
40
50
60
25002300210019001700150013001100900700500300
LateEarly
Greasy Spot
0
10
20
30
40
50
60
25002300210019001700150013001100900700500300
LateEarly
Wind Scar
Cake Melanose
0
10
20
30
40
50
60
25002300210019001700150013001100900700500300
LateEarly
Copper Burn
0
10
20
30
40
50
60
25002300210019001700150013001100900700500300
LateEarly
19
Industry ContactWe have visited numerous Florida packinghouse’s and have sought close working relationships and advise.We have made contact with a prominent grading equipment manufacturer and have discussed the potential for collaboration.We have purchased a one-lane grading system prototype, where we will integrate our sensing/grading technology on their system to test design, detection and control concepts.
20
Year Two Objectives
Further develop On-line sensing technologies Integrate sensing technologies with one-lane grading system.Monitor canker spectral characteristics over the entire growing season on grapefruit to see if there is a seasonal difference.Explore spectral characteristics of other citrus varieties: navel, and tangerine are likely candidates.
21
Conclusions
Spectral reflectance of citrus canker and other disease conditions were acquired, and key wavelengths have been identified in the visible and the NIR bands.
A discriminant function was able to separate canker from other conditions with overall accuracies of 93% for all conditions, and 100 % for canker.
Results have shown a shift in the reflectance curves between “early” samples with canker and “late” samples with canker. Further spectral reflectance test and analysis will be conducted to study influence of time of season on canker spectral response.