1 Malcolm Thomson International Centre for Island Technology Heriot Watt University(Orkney Campus) Old Academy Stromness Orkney Scotland, UK SUMARE Workshop:

1

Malcolm Malcolm ThomsonThomson

International Centre for Island TechnologyHeriot Watt University(Orkney Campus)

Old AcademyStromness

OrkneyScotland, UK

SUMARE Workshop: Underwater Robotics for Ocean Modelling and Monitoring

Classification of maerl beds Classification of maerl beds using video imagesusing video images

Classification of maerl beds Classification of maerl beds using video imagesusing video images

2

Use of video in marine habitat mapping

Data outputs & problems

Influence of altitude on classification

Recognition of different maerl features

SUMARE and the maerl case study

3

Video in marine habitat mappingVideo 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 processingUnsupervised” 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 Project SUMARE - maerl applicationapplication

• 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

8

~2m

Maerl mosaic from Wyre Sound, Orkney Islands

9

Information requirements for Information requirements for maerlmaerl

• 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 Characteristics of maerl habitatshabitats

Analysis of video footage collected during SUMARE sea trials, August 2000

4 features:4 features:Living maerlDead maerlMacroalgaeSand

Survey requires recognition of these features

11

RecognitionRecognition of maerl features 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...Living and dead maerl...

Living maerl occupies the darker portion of the greyscale histogram

Dead maerl occupies the lighter portion of the greyscale histogram

13

Sand...Sand...

Sand appears similar to dead maerl at certain altitudes

San

3.7 m

0

0.1

0.2

0.3

0.4

0.5

1 2 4 5 6 7 8 9 10 11 12 13 14 15

Sand 3.7m

Sand has a uniform greyscale range

14

Macroalgae.Macroalgae.

Macroalgae6.9 m

0

0.1

0.2

0.3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Macroalgae exhibits a broad greyscale range at high altitudes

15

Altitude and classificationAltitude 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 Living maerl...maerl...

DecreasingAltitude

L iv e M a e r l

8 .4 m

0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

L iv e M a e r l

6 .5 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Live m a e rl

4.5 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

L i v e M a e r l

1 . 1 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

L iv e M a e rl

0 .5 m

0

0.1

0.2

0.3

0.4

0.5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

8.4m8.4m

6.5m6.5m

4.6m4.6m

1.1m1.1m

0.5m0.5m

17

Dead Dead maerl...maerl... 8.4m8.4m

5.3m5.3m

2.3m2.3m

0.9m0.9m

0.7m0.7m

DecreasingAltitude

D e a d M a e r l

8 .4 m

0

0 .1

0 .2

0 .3

0 .4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

D e a d M a e r l

5 .3 m

0

0.1

0.2

0.3

0.4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

D e a l M a e rl

2 .3 m

0

0.1

0.2

0.3

0.4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

D e a d M a e rl

0 .9 m

0

0.1

0.2

0.3

0.4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

D e a d M a e l

0 . 7 m

0

0 .1

0 .2

0 .3

0 .4

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

18

Sand...Sand...5.3m5.3m

4.5m4.5m

3.7m3.7m

2.5m2.5m

0.6m0.6m

DecreasingAltitude

Sa n d

5 .3 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

S a n d

4 .5 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

S a n d

3.7 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

S a n d

2 .5 m

0

0 .1

0 .2

0 .3

0 .4

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

S a n d

0 . 6 m

0

0 .1

0 .2

0 .3

0 .4

0 .5

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

19

Macro-Macro-algaealgae 6.9m6.9m

4.8m4.8m

2.8m2.8m

1.6m1.6m

0.8m0.8m

M a c r o a l g a e

6 . 9 m

0

0 .1

0 .2

0 .3

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

M a c r o a l g a e

4 . 8 m

0

0 .1

0 .2

0 .3

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

M a c r o a lg a e

2 .8 m

0

0 . 1

0 . 2

0 . 3

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

M a c ro a lg a e

1 . 6 m

0

0 .1

0 .2

0 .3

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

M a c r o a l g a e

0 . 8 m

0

0 . 1

0 . 2

0 . 3

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5

D ecreasingA ltitu de

20

The resultThe 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 algorithmComputer 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

ProblemsProblems

• 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 workFuture work

• Continued development of classification algorithm

• Field trials in 2002 (Orkney)