Image Analysis in Horticulture

1
Image Analysis in Horticulture Rodney Edmondson
Yu Song Nick Parsons Steve Adams
2
Outline of Talk

• MORNING
• Grading for pot and bedding plants
• Crop scanning
• Disease detection
• AFTERNOON
• Crop scanning and height tracking

3
HDC bedding plant project PC200

• This project was set up in 2003 to study bedding
  plant pack quality
• The project was later broadened to include image
  analysis of pack quality
• The project provided a test-bed for image
  analysis methods for glasshouse crops

4
PC 200 Bedding Plant Quality Assessment
5
Image analysis was developed to measure the
characteristics of individual packs
6
Light-box for overhead stereo-pairs of images
7
Stereo-pairs of pansy and impatiens packs
8
Image processing for overhead stereo-pairs

• Image characteristics
• Leaf Cover () (ii) Flower Cover ()
• Flower Colour (iv) Leaf Colour
• (v) Canopy Height (vi) etc
• Methods
• (i) Segmentation for pack cover of leaf, flower.
  etc
• (ii) Stereoscopy for 3-dimensional structure

9
Segmentation for pansy pack cover
Leaf 62.7 Pack 1.2 Flower 35.5
Compost 0.6
10
Segmentation for dianthus leaf and flower cover
Calculated leaf cover () 73.5 Calculated
flower cover() 13.9
11
Segmentation for cyclamen leaf and flower cover
Calculated Leaf Cover () 73.4 Calculated
Flower Cover () 22.2
12
Practical applications of image analysis in Dutch
horticulture

• Image analysis for automated grading of pot
  plants is well developed in Holland
• The following examples shows image analysis used
  for grading pot plants in Dutch Nurseries

13
Pots are transported to the grading cameras by
moving belts
14
Then moving pots are imaged by fixed cameras in a
light box
15
The pots are graded by image analysis methods and
are electronically tagged.
16
Next the electronically tagged pots are passed
into a grading line
17
Finally, the pots are sorted into grades
according to the tag information
18
Vision applications from the Walking Plant System
website

• The WPS website shows videos of image grading
  systems in Holland
• The following screen dumps show some images from
  their example videos

19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Comment on the Dutch vision systems for pot-plant
grading

• The Dutch system moves the pot-plants to a fixed
  camera in a controlled environment
• The Dutch system can only be used at grading or
  re-spacing of pots
• The system requires major investment in equipment
  and machinery

26
Our vision system envisages overhead scanning of
production crops in situ

• We envisage an overhead moving camera to scan a
  standing crop from above
• The camera will be mounted on a moving boom and
  the crops will be scanned in situ
• The system will use off-the-shelf cameras and
  processors and should be relatively cheap

27
Overhead crop scanning for image analysis
Glasshouses have rigid infrastructures that can
be used to carry overhead booms for camera systems
Overhead scanning can monitor the growth and
development of large uniform areas of crops
28
We have investigated two different overhead
scanning systems

• System 1
• Cameras on a moving boom scanning down the length
  of the bed
• System 2
• Cameras on a stationary boom scanning across the
  width of the bed

29
System1 Fixed cameras on a moving boom
30
Yu Song setting up the scanning software
31
Stereo-cameras scanning down the length of a bed
on a moving boom
32
System 2 Moving camera on a stationary boom
scanning across the width of a bed
33
System 2 requires a motorised camera with an
optical counter to give exact image separation
34
We have investigated two camera types for
overhead crop scanning

• IP web cameras (security cameras)
• USB machine vision cameras
• Both systems provide a fast frame rate for image
  capture approx 5-10 frames per second

35
UIP cameras and image capture
Cabled together on a scanning boom
Router
Camera
Wireless link
Remote processor for storage, analysis and control
36
USB cameras and image capture
Cabled together on a scanning boom
Router
Processor
Camera
Wireless link
Remote processor for storage, analysis and control
37
The three cameras we have tested
3 Megapixel IP camera
1/3 Megapixel USB camera
1/3 Megapixel IP camera
38
Comparison of the three cameras (without
artificial lighting)
1/3 Megapixel USB camera
3.0 Megapixel IP camera
1/3 Megapixel IP camera
39
A test lighting system for cross-bed scanning
camera
40
White and yellow flowers with dark green foliage
using a1/3 megapixel IP camera
Unlit
Lit
41
Pink buds and dark green foliage using a1/3
megapixel IP camera
Unlit
Lit
42
Controlled lighting is essential and will require
a rig with adjustable blinds or screens
These screens would also protect the cameras and
electronics
43
Components for a crop scanning rig

• A carrier for locating the camera over the
  required section of bed
• A precision rack for scanning across the bed
• A controlled environment for the camera system
• A trigger for firing the camera at exact
  separations
• A camera with suitable imaging software
• A pre-programmed processor for automated scanning
  tasks

44
Possible benefits of crop scanning

• Tracking crop development for crop scheduling
• Continuous monitoring of crop quality and
  uniformity
• Routine crop inspection by live video display

Further discussion in the afternoon session
45
Disease detection using imaging methods

• Visual imaging for pest or disease symptoms or
  damage
• Visual imaging of crops by sideways looking
  cameras
• Thermal imaging of temperature anomalies due to
  pests or disease

46
Visible symptoms of white rust pustules on
detached leaves
Can image analysis be used to detect specific
disease symptoms routinely by crop scanning?
47
Segmentation software can be trained to identify
specific disease image signatures
In real crops, pests and diseases can be hidden
beneath the canopy
48
Oblique looking cameras might be used to give a
different perspective on a crop
49
Images from a sideways looking camera
50
Images from a sideways looking camera
51
Images from a sideways looking camera
52
Thermal imaging

• Pests and diseases are known to affect plant
  transpiration and can cause plant temperature
  anomalies
• Temperature effects may appear before visual
  symptoms and this can provide an early warning of
  pest or disease damage

53
TMV infection can cause a thermal signal before
visual effects are seen
Early stage infection
Late Stage infection
Thermal images
Visual images
54
The thermal image of even a healthy crop is
highly variable over the crop area
Visual image
Thermal image

Discrimination of thermal effects will require
combination of information across both the visual
and the thermal spectrum
55
Multispectral imaging
Multispectral sensors could readily be used for
crop scanning from a motorised rigs either in
glasshouses or in the field
56
Conclusions

• Image analysis is powerful for automated
  inspection of crops
• Image analysis has great potential for routine
  crop surveillance
• Image analysis could be linked to other
  technologies for stress, pest and disease
  detection

57
Predictions

• Image analysis will provide a major tool for crop
  management and surveillance
• The most immediate routine applications will be
  in glasshouse and bedding crops
• Intelligent sensors and software will become
  increasingly important for crop monitoring

58
Acknowledgments

• HDC for project funding
• GCRI Trust for visit to Holland
• Double H Nursery and Mike Holmes for crop
  scanning facilities
• Roundstone Nurseries for images
• PC200 Steering group for help and advice