1 Improved fish detection probability in data from split- beam sonars. Helge Balk and Torfinn Lindem. Department of Physics. University of Oslo.

1

Improved fish detection probability

in data from split-beam sonars.

Helge Balk and Torfinn Lindem.

Department of Physics. University of Oslo.

2

Method and material. Method and material. Development of sonar software. Most data collected with Simrad EY500. Some data collected with HTI modell-243 Experience from fieldwork.

River Tornio (Finland summer 97) Lake Semsvannet. (Norway winter 98, 99) River Tana (Norway summer 98, 99) (Data from various other rivers and lakes.)

3

Traditional counting method.Traditional counting method.

Single Echo Detector (SED)

4-ChTVG

Phase detector

EnvelopeDetector

Tracking

Raw-echogram(time / range)

X/Y position diagram

SED-echogram(time / range)

fish-tracktrack statistics

split-beamtransducer

4

Horizontal application Horizontal application in shallow rivers:in shallow rivers:

The traditional method tends to faile because of:The traditional method tends to faile because of:

Increased noise-level. (Rain, silt, running water, bottom and surface reflections, air-bubles, debris )

Increased phase and amplitude fluctuations in the echo-signal. ( side-aspect, reflection, transducer-vibrations, moving sound-media )

Multiple object problem.(fish, debris, stones....;- Classification nessesary)

5

SED;- the main problem.SED;- the main problem.

Noise is too easily detected as single targets. Important information is removed.

1) Increased fluctuation in the echo-signal increases the rejection of echoes from fish.

2) The shape of a track.

3) Echoes below a fixed treshold.

4) Echoes from fish-schools.

RAW-Echogram

ClassificationTrackingSED

SED-Echogram

6

Tracking algorithm.

a) Missing echoes results in rejection of fish-tracks.

b) Noise-echoes results in creations of artificial fish-like tracks.

Four salmons a rainy day in Tana.Four salmons a rainy day in Tana.

Tracking result SED-echogram

ClassificationTrackingSED

7

How to improve the method.How to improve the method.

Four salmons on a rainy day in Four salmons on a rainy day in Tana.Tana.

Raw-echogramSED-echogram

ClassificationTrackingSED

Collect more data. Extract more information from existing data.

8

?

?

9

Image Image analysisanalysis..

Contour detection

Shape analysis Filters

Segmentation

Classification

Morphologic operations

Intensity

Texture

Image

10

Convolution and Convolution and window operations.window operations.

Echogram array F(m1, m2)

I H GF E DC B A

I H GF E DC B A

F (n1, n2 ) = Input image arrayH(m1, m2 ) = Impulse response arrayQ(m1,m2) = Output image array

Q m m F n n H m n m nnn

( , ) ( , ) ( , )1 2 1 2 1 1 2 222

1

Windowproducing one output pixel.

1 1 11 1 11 1 1

1 1 1 0 0 0-1-1-1

0 1 0 1-4 1 0 1 0

Hit-miss, mean, roberts-c, laplace

0 0 00 1 00 0 0

Q F H

11

Filtering.Filtering. Many well-known filters availabel.

Low-pass: Median, Mean, Knn, Sigma, High-Pass: Sobel, Robert’s, Prewitt, Gradient, Laplace. Morphologic filters: Hit-miss, Hit-add.

Not always an improvement.

Filter dimension is important.

Original Median 9x1 Median 1x9 Robert’s col. Robert’s col.Echogram Low-pass Low-pass High-pass + Median 3x5

12

Segmentation. Segmentation.

Edgebased. Detecting edges.

(high-pass: gradient, Laplace.) Linking.

Region based. Tresholding. Growing and shrinking. Seeds. Split and merge. Relaxation.

Separating background and foreground.

13

Shape analysis.Shape analysis.

Central moments. Radius of gyration. Orientation. Topological features. Area. Contour length and

smoothness. Compactness. Eccentricity.

Small shapes originates from surface noise and airbubbles.

High fluctuation in contour and large area may indicate bottom structures.

Fish and debris seen to produce thin and smooth tracks.

p,q

G2,0 0,2

0,0

1,1

2,0 0,2

R =

( ) ( ) ( , )

tan

x x y y f x y

=

yx

p q

12

21

14

Putting things together.Putting things together.

Single EchoDetector

Track analysis

Classification

Raw-echogram Image

processor

Regions of fish,

debris or stones.

Contour detector

Shape analysis

Tracking algorithm

Phase detector

Envelopedetector

15

Median 3x15 Treshold Region Contour Single echo filter -52 dB growing detection detection

Two fishesStones Drifting

debris

Testing a difficult case. Testing a difficult case.

Original raw- and SED-echogram

16

Conclusion:Conclusion:Combining image analysis with

the traditional metod is

promising!

The traditional SED is still reducing the fish detection probability. Difficult to find one parameter setting that manage to handle all kinds of

tracks and noise. More research is needed!

However we have shown that this method manages to:

* Extract and use important information lost by the SED.

* Reduce the creation of noise-based fish tracks.

The overall ability to detect fish in sonar data with low signal to noise ratio has been improved!