Broadband Fish Identification of Great Lake Fishes

Broadband Fish Identification of Great Lake Fishes

Patrick Simpson and Mike Tuohey

Scientific Fishery Systems, Inc.

Anchorage, AK

Guy Fleischer and Ray Argyle

Biological Resources Division - U.S. Geological Survey

Great Lakes Science Center

Ann Arbor, MI

OverviewOverview

• Why Broadband Sonar?• System Overview• Data Collection

• Classification Results• Future Work

Broadband vs. NarrowbandBroadband vs. Narrowband

Benefits of Broadband Fish Identification

Benefits of Broadband Fish Identification

• Full-Column Assessment

• Continuous Assessment

• Remote Assessment

• Cost Savings

Broadband Sonar Fish Identification System Prototype

Broadband Sonar Fish Identification System Prototype

SIGNALIN

RD Instruments

HP 33120A

OUT

COAX SONAR TRIGGER

COAX SONAR DATA

COAX FUNGEN TRIG

RS-232

COAX XMIT CLOCK

FUNGEN CTRL

RS-232

RDI CTRL

SciFish CPU - Orca

RDI VM Chassis

HP FunGen

RDI SonarTransceiver

• Pentium 90 MHz CPU - 16 MB RAM

• 16 bit, 770 kHz ADC-DSP board

• PCI SCSI and 2 MB video adapters

• 1.0 GB SCSI-2 Hard Disk

• 1.2 GB Magneto-Optical Drive

• Quad-speed SCSI-2 CD-ROM

• 1.44 MB Floppy Disk Drive

• ORCA.EXE acquire & process s/w

MULTI-CONDUCTORUNDERWATER CABLE

SONAR DATA AND CONTROL

Broadband TransducerBroadband Transducer

• Resonant Frequency 153,600 Hz• 3 dB Operation Band 45 kHz (138 - 183 kHz) Q=3.4• Active Surface 177-mm disc• Beam Pattern 4.1° beamwidth, sidelobes @ 20 dB• Rated Power 80 W transducer & wet electronics• Source Level 216 dB re 1 uPa @ 1 m @ 153.6 kHz• Transmit Sensitivity peak TVR of 181 dB re 1 uPa-m/V @ 169

kHz • Receive Sensitivity peak OCVR of -180 dB re 1 V/uPa @ 169

kHz

Processing PlatformProcessing Platform

• Mid-tower computer case with 230 W power supply• Plato motherboard with ISA/PCI bus adapters, 256 kByte cache• Intel Pentium 90 MHz CPU with 16 Mbyte DRAM• PCI SCSI-2 host adapter controlling three SCSI-2 devices below• 1.0 GB hard disk, 1.3 GB magneto-optical drive, and quad-speed

SCSI-2 CD-ROM• 12 bit 770 kS/s ADC/DSP card with 486DX2/66 and 4 MB on-board

DRAM• 1280 x 1024 video monitor and video adapter card with 2 MB DRAM• DOS 6.22 running ORCA.EXE Interface & Processing

Data ProcessingData Processing

ACOUSTIC RECEIVER

A/D

CONVERSION

ECHO

DETECTION

FEATURE EXTRACTIO

N

CLASSIFICATIONMAN-MACHINE

INTERFACE

SIGNATURE

DATABASE

FeaturesFeatures

• Single Ping Parameters

• Multiple Ping Parameters

• Environmental Parameters

Neural Net ClassifierNeural Net Classifier

o o o

o o o

o o o

F A

F B

F C

HiddenNodes

InputNodes

OutputNodes

a 1 a 2 a 3 a n

b 1 b 2 b 3 b p

c 1 c 2 c 3 c m

Features

Classes

Tethered Individual FishTethered Individual Fish

BroadbandTransceiver

TetheredFish Suspendedin Sonar Beam

AnchoredTether Line

Surface

Bottom

Small MeshNetting Holds

Live Fish

Free-Swimming FishFree-Swimming Fish

Tethered Data CollectionTethered Data Collection

Species N Mean Length(mm)

Numberof Echoes

MeanEcho

Length

MeanMaximum

mV

MeanRMS mV

Alewife 24 178 1231 262.8 118.9 51.8

RainbowSmelt

21 128 1236 258.6 162.2 78.4

Stickelback 9 65 427 266.6 55.7 26.8

Classification ResultsClassification ResultsProb. Threshold = 0.48

Alewife Smelt Stickleback

Alewife 0.80 0.16 0.04

Smelt 0.05 0.91 0.04

Stickleback 0.06 0.22 0.72

Echoes Retained 0.94 0.98 0.87

Prob. Threshold = 0.7

Alewife 0.91 0.05 0.03

Smelt 0.04 0.93 0.03

Stickleback 0.05 0.07 0.88

Echoes Retained 0.64 0.67 0.42

Tethered Vs. Free-Swimming Results

Tethered Vs. Free-Swimming Results

Composition of Fish Trawled from Lake Charlevoix

Alewife Smelt SticklebackFish Count 0 84 38Pct. of Total 0% 69% 31%

Composition of Free-Swimming FishComputed by Neural Network

Alewife Smelt Stickleback UnknownFish Count 3 25 29 2%. of Total 5% 44% 49% 2%

Future WorkFuture Work

• Wider Beam (15 degrees)

• Dual / Split Beam

• Greater Source Level

• Streamline Data Storage and Analysis

• More Data Collection