1 Fish detection based on spectral differences in the echogram's range and temporal domain Helge Balk and Torfinn Lindem University of Oslo Department.

1

Fish detection based on spectral Fish detection based on spectral differences in the echogram's range and differences in the echogram's range and

temporal domaintemporal domainHelge Balk and Torfinn Lindem

University of OsloDepartment of Physics

[email protected] / [email protected]

Proceeding for the FAST meeting in Bergen June 2003

2

Think of a fish-track in an Think of a fish-track in an echogram and echogram and

- imagine the echo signal along a horizontal line

- imagine the echo signal along a vertical line

3

MaterialMaterial

Simrad EY500 120kHz 4x10 deg Split beam transducer Sonar5-Pro post processing tool

Simrad EY500 Sonar5-Pro

4

Split beam echo sounderSplit beam echo sounder

Phase detector

AmplitudeDetector

4-ChTVG

Single echo detector

Position diagram

SED-Echogram

Amp-Echogram

5

Results

Fish detection

Conclude

Single echo detector

Principle

Welcome

Introduction

Think of a fish Material and method

Importance

Cross-filter detector (CFD)

Problems

Human perception

6

Single echo detection (SED) is Single echo detection (SED) is important in three situationsimportant in three situations a) for fish counting a) for fish counting b) for fish behaviour study b) for fish behaviour study when fish echoes can be resolved as single targetswhen fish echoes can be resolved as single targets

7

c) in abundance estimation of c) in abundance estimation of schools schools where the detections serve as awhere the detections serve as a link between link between the fish and the Echo Integral (EI)the fish and the Echo Integral (EI)

A link can be established by: SED Tracked SED Catch statistics

8

Linking EI and SED to Linking EI and SED to obtain the fish densityobtain the fish density

SED detections surrounding a school gives the size distribution

TS

n

Size distribution (nj)School

SED

sed

tot

sedj

totj

EIEI

,

,

jv

jbsjsedj sN

n

,

,,

SED

=fish density

j =Size class

n =number in size class j

N =all SED

9

The parametric single echo The parametric single echo detector (SED)detector (SED)

Analyse one ping at a time Describe a single echo with parameters

Echo length Shape Phase deviation Angular position Threshold

10

Two problems with SEDTwo problems with SED

1. Rejection of echoes from single fish

2. Detection of false echoes from fluctuations in the background reverberation level

Especially profound with low signal to noise ratio.

11

Results

Fish detection

Conclude

Single echo detectors

Principle

Welcome

Introduction

Think of a fish Material and method

Importance

Cross-filter detector (CFD)

Problems

Differ from SEDHuman

perception How to copy

12

Human perceptionHuman perception

extremely noisy, but we can still see the tracks

Horizontal, stationary recording of spawning Bream

13

Human perception Human perception

This is what the parametric SED evaluatesThis is what a parametric SED seesThis is what we can see

a) We look at more than one ping at a time.

b) We apply information from the background

14

Next question- Next question-

How can we copy these two How can we copy these two elements in a computer elements in a computer algorithm?algorithm?

Answer: They can be copied with filters

15

Results

Fish detection

Conclude

Single echo detector

Principle

Welcome

Introduction

Think of a fish Material and method

Importance

Cross-filter detector (CFD)

filtering in time

Problems

Problems

Differ from SEDHuman

perception

Echogram freq. spectre

filtering in range

combining

How to copy

16

Factors influencing on the Factors influencing on the echogram’s frequency componentsechogram’s frequency components

(mainly controllable factors)

ship velocity

ping ratesample rate

beam widthsound speed

Range

band width

pulse length

Time

17

An example An example from lake from lake d’Annecy (Fr)d’Annecy (Fr)

18

Filtering in timeFiltering in time

1 Remove temporal noise pulses

2 Remove fluctuations in the fish track

Combine information from multiple pings (mean filter)

Missing echoes

threshold

Noise pulseFish echo

threshold

Magnitude

Freq.

Frequency specter

Energy from fish

Energy from noise

19

Filtering in time improves the Filtering in time improves the fish trackfish track

20

Filtering in time, Filtering in time, equation and filter impulse response arrayequation and filter impulse response array

3

3

1

021211 ),1(),(),(

t r

tmrmHtrFmmQ

F Q1

H= [1/3 1/3 1/3]

21

Filtering in rangeFiltering in range

A low-pass filter can remove the fish and detect the background signal

Magnitude

Freq.

Frequency specter

Energy from fish

Energy from background

22

Filtering in range Filtering in range removes the fishremoves the fish

1

0

21

2121212 ),1(),(),(

t r

tmrmHtrFmmQ

.21

121

121

1

H

Echogram F Range filtered echogram Q2

23

Combining the two results by letting Combining the two results by letting Q2 define the threshold in Q1Q2 define the threshold in Q1

0),( 6dB ),(),(

),(),( 6dB ),(),(

321

321

jiQjiQjiQ

jiFjiQjiQjiQ

Echograms

F= Original

Q1= Time filtered

Q2= Range filtered

Q3 =Result

24

Combining the two filtered echogramsCombining the two filtered echograms

Filtering in time

H1Filtering in range

H2Cross filter

Named due to the orientation of the two filter impulse response arrays H1 and H2

Combiningdetector

25

Problems with false detections Problems with false detections

noise fish bottom/schools

Fortunately the size of the detected regions differ

size < size < size

Size filter

26

Problems with echo qualityProblems with echo quality

threshold

time filtered

Faint echoes in a track can be detected in the

process.

Range will be correct

Estimates based on phase such as TS and velocity

may be unreliable

Faint echoes

time signal from a fish

original

threshold

27

Solution to the Solution to the quality problemquality problem

Mark each echo with a quality stamp

A parametric “SED” can do this

all

TS and position

Quality estimation

Cross-filter detected echoes

Tracking

high quality

Combines the best from the two detectors

low qualityhigh quality

28

Implementation of the Cross-Implementation of the Cross-filter detector in Sonar5-Pro filter detector in Sonar5-Pro

softwaresoftware

Amplitudedetector

Phasedetector

TimeFilter

RangeFilter

Combinesignal

backgroundlevel

Sizefilter

Qualityestimation

Detections

low qualityhigh quality

29

Results

Fish detection

Conclude

Single echo detector

Principle

Welcome

Introduction

Think of a fish Material and method

Importance

Cross-filter detector (CFD)

filtering in time

Problems

Problems

Differ from SEDHuman

perception

Echogram Freq. spectre

filtering in range

combining

How to copy

Herring school Spawning

Bream

Migrating Salmon

30

Visual results, spawning BreamVisual results, spawning Bream

Amplitude echogram Parametric single echo detectionCross-filter detection before size filterCross-filter detection after size filter

31

Visual results, herring schoolVisual results, herring school

Rotate

Time filter

Range filter

Combine

Rotate + Size filter

32

Comparing thresholding, parametric Comparing thresholding, parametric SED and Cross-filteringSED and Cross-filtering

Original Amp-echogram Thresholded Amp-echogramParametric SED echogram Cross-filtered SED-echogram

33

Numerical resultsNumerical results

D2

D3

D4 D5

D6

F1

F2

F3

Bottomlines

dB

D1

D=Debris

F=Fish

B=Bottom

Horizontal stationary recording River Tana summer 1999

Amp-

echogram

CFD-echogramSED-echogram

34

NumericalNumericalresultsresults Targets SED

TQ (%)CFD

TQ (%)Fish 1 54 100Fish 2 41 100Fish 3 49 65Debris 1 7 -Debris 2 5 74Debris 3 19 55Debris 4 43 100Debris 5 11 100Debris 6 - 54Bottom 1 33.01m 27 -Bottom 2 34.65m - 100Bottom 3 35.65m - 97Bottom 4 39.52m 41 100Bottom 5 40.06m 25 100Bottom 6 40.87m 37 100Bottom 7 41.37m 51 100Sum in tracks 410 1245Tot. no. of det.TNR 50 % 97%

TQRatio between actual

and possible number of detections

TNRTrack to noise ratio,

number of detections in tracks rel total number

of detections

35

Results

Fish detection

Conclude

Single echo detector

Principle

Welcome

Introduction

Think of a fish Material and method

Importance

Cross-filter detector (CFD)

filtering in time

Problems

Problems

Differ from SEDHuman

perception

Echogram Freq. spectre

filtering in range

combining

How to copy

Herring school Spawning

Bream

Migrating Salmon

Acknowledgement

Conclusion

36

ConclusionsConclusions

The Cross-filter detector is a good alternative to the common parametric single echo detector

The Cross-filter detector is superior in situations with low SNR

37

AcknowledgmentAcknowledgmentData has been provided by…..

Marie Prchalova (Cz)Horizontal recording of spawning Bream

Nathalie Gaudreau (Ca)Vertical recording from

Lake d’Annecy

We thank all for their contribution!

Frank R Knudsen at Simrad

Assisted in the horizontal recording of salmon in River Tana

JimVertical recording of Herring schools outside Vancouver

island

38

Fish detection based on spectral Fish detection based on spectral differences in the echogram's range differences in the echogram's range

and temporal domainand temporal domain

Test CD available by writing to...

[email protected] Balk,

Department of Physics PB1048,

University of Oslo0316 OSLO,

NORWAY

Thank you for your attention!