Towards Automated Detection of Stress in Tree Fruit Production J. Park, H. Ngugi, M. Glenn, J. Kim & B. Lehman
May 11, 2015
Towards Automated Detection of
Stress in Tree Fruit Production
J. Park, H. Ngugi, M. Glenn, J.
Kim & B. Lehman
The CIA monitors world-wide, agricultural production with satellite-based, remote sensing.
During the Cold War, the U.S. used this information in the sale of wheat to Russia
World food production is monitored to anticipate governmental instability as well as markets.
From a global scale to a farm scale, this technology can be used to improve grower productivity.
Potential applications of monitoring
technology in tree fruit production
• Detection of tree stress
– Moisture stress (drought or excess water)
– Nutrient stress
– Disease and insect stress
• Estimation of expected yield
• Any other use?
Sensor technology for use in
tree fruit production
All sensor-based systems rely on reflected light from
a portion of the electromagnetic spectrum (EMS)
0
10
20
30
40
50
60
70
80
90
100
40
04
24
44
74
71
49
45
18
54
15
65
58
86
12
63
56
59
68
27
06
72
97
53
77
68
00
82
38
47
87
08
94
91
79
41
96
49
88
Ref
lect
ion
(%
)
Wavelength (nm)
well watered stressed
Changes in chlorophyll activity
Reflection spectrum of apple leaves
Reduced water content
Visible light Near Infra-red radiation
Types of sensors being evaluated
in the CASC project
• Thermal cameras
• NDVI sensors
• Hyperspectral cameras
• Color cameras
Detecting fire blight in orchards
Bacterial disease caused by Erwinia amylovor
Often leads to death in young trees
Factors determining successful
fire blight management
• Once infection occurs, successful
management depends on:
– Early detection
– Application of appropriate control measures
such as cutting out infected shoots
– Continued monitoring
All the factors point to the need for
regular scouting!
Options for
scouting orchards
for fire blight
Current CASC Project Research
• Identification and evaluation of suitable
sensors for automated detection
• Preliminary detection experiments
– Can we detect fire blight with sensors?
– How early can we detect lesions?
• Development of detection algorithms
Potential rapid detection systems
for fire blight
• Biological-based detection systems
– Molecular-based techniques
– Can be quite rapid
• Main challenge is sampling (very large numbers of samples)
– How many shoots (all a potential infection sites)
– Destructive sampling
– Would be very labor-intensive with current technology
– Currently restricted to confirming pathogen identity
Potential rapid detection systems
for fire blight cont.
• Sensor-based detection systems
– Rely on sensors to detect plant response to infection
– No destructive sampling or sample preparation
– Can be as rapid as real-time
– Can cover a large area over a short time
• Main challenge: the right sensors and developing
the detection algorithms
• This is the approach followed in the CASC project
Sensors evaluated for blight detection
700 nm
700 nm
Target for early detection:
<10 cm of diseased tissue
(~7 days after infection)
Inoculated plants in the
green house at: 14, 10, 7, 4
and 2 d before image
acquisition
Hyperspectral images 300
to 1100 m
Detection of fire blight with
hyperspectral sensor
Detection of fire blight with
hyperspectral sensor
Sensors mounted on the APM
What we hope to accomplish
• Detection of diseased shoots within 7
days after infection for fire blight
– No more than 1-3 leaves have visible
symptoms for virulent strains
– Over 85% accuracy rate
• Detection of other types of stress
• Develop a database that to help identify
causes of tree stress
Acknowledgments