Malcolm Thomson

1
Classification of maerl beds using video images
SUMARE Workshop Underwater Robotics for Ocean
Modelling and Monitoring

• Malcolm Thomson

International Centre for Island Technology Heriot
Watt University(Orkney Campus) Old
Academy Stromness Orkney Scotland, UK
2

• Use of video in marine habitat mapping

SUMARE and the maerl case study
Recognition of different maerl features
Influence of altitude on classification
Data outputs problems
3
Video in marine habitat mapping

• Widely used by divers and in ROVs for seabed
  survey
• Human interpretation required
• Simple data processing, e.g. animal counting
• Used to ground truth acoustic survey results,
  e.g. Sound of Arisaig SAC

4
Unsupervised video processing

• Used by Lebart et al. (2000) to detect features
  in sea floor video transects
• looking for discrete features
• Seabed habitat mapping is a priority in marine
  research e.g. ICES, OSPAR, Habitats Directive
• unsupervised classification tools have great
  potential
• large data outputs

5
Project SUMARE - maerl application

• Recap-
• Marine alga
• Non-jointed calcareous structure
• Can form large deposits on the seabed
• Found in or near strong water currents
• Is exploited commercially in France, the UK and
  Ireland
• Very high species diversity - high conservation
  value

6
10 cm
7
(No Transcript)
8
Maerl mosaic from Wyre Sound, Orkney Islands
9
Information requirements for maerl

• Dimensions of maerl beds
• Variation in area coverage of maerl
• variation in amount of living maerl may indicate
  the health status
• SUMARE - use autonomous sensors to
• provide information on the boundaries of maerl
  beds
• estimate the coverage of living (and dead) maerl
  within these beds.
• Practical application
• conservation exploitation

10
Characteristics of maerl habitats
Analysis of video footage collected during SUMARE
sea trials, August 2000
4 features

• Living maerl

• Dead maerl

• Macroalgae

• Sand

11
Recognition of maerl features

• Visual discrimination
• Analysis of selected examples of maerl features,
  e.g. living maerl
• examine greyscale properties for each feature
• greyscale histograms characteristic of different
  features
• histograms produced by MatLab
• combined effort from biologists and computer
  programmers

12
Living and dead maerl...
Dead maerl occupies the lighter portion of the
greyscale histogram
Living maerl occupies the darker portion of the
greyscale histogram
13
Sand...
14
Macroalgae.
15
Altitude and classification

• Greyscale values vary with ROV altitude
• Some confusion between different features with
  similar greyscale histograms
• To improve classification
• collect images from different altitudes
• compare greyscale histograms

16
Living maerl...
8.4m
6.5m
4.6m
1.1m
0.5m
17
Dead maerl...
8.4m
5.3m
2.3m
0.9m
0.7m
18
Sand...
5.3m
4.5m
3.7m
2.5m
0.6m
19
Macro-algae
6.9m
4.8m
2.8m
1.6m
0.8m
20
The result

• Histogram sets
• for each of the 4 maerl features
• living maerl
• dead maerl
• sand
• macroalgae
• for varying altitudes (0.5 - 8m)

Reference database
21
Computer algorithm

• Written in Visual C
• Analyses maerl bed video footage
• Identifies maerl features by reference to
  histogram database
• Accuracy of classification improves with the
  number of images in each database
• Quantify area of seabed covered by living and
  dead maerl
• application in exploitation and conservation of
  maerl

22
Problems

• Variation in image exposure
• depth
• light conditions (sun, cloud)
• water clarity
• Indistinct boundaries between features
• e.g. sand and dead maerl
• Presence of other features
• e.g. rock, other species of algae

23
Future work

• Continued development of classification algorithm
• Field trials in 2002 (Orkney)