Evaluation of potential bias in observing fish with a DIDSON acoustic camera

Page 1

Ea

MTa

b

c

d

a

ARRAH

KDFLHA

1

hbrid2

(bp

h0

Fisheries Research 155 (2014) 114–121

Contents lists available at ScienceDirect

Fisheries Research

j ourna l ho me pa ge: www.elsev ier .com/ locate / f i shres

valuation of potential bias in observing fish with a DIDSONcoustic camera

ichal Tusera,b, Jaroslava Frouzováa, Helge Balka,c, Milan Muskaa,omás Mrkvickaa,d, Jan Kubeckaa,∗

Institute of Hydrobiology, Biology Center of the Academy of Sciences of the Czech Republic, Na Sádkách 7, 370 05 Ceské Budejovice, Czech RepublicFaculty of Science, University of South Bohemia, Branisovská 31, 370 05 Ceské Budejovice, Czech RepublicDepartment of Physics, University of Oslo, P.O. Box 1048, Blindern, 0316 Oslo, NorwayFaculty of Economics, University of South Bohemia, Studentská 13, 370 05 Ceské Budejovice, Czech Republic

r t i c l e i n f o

rticle history:eceived 4 July 2013eceived in revised form 13 January 2014ccepted 22 February 2014andling Editor George A. Rose

eywords:IDSONish detectionength measurementorizontal body orientationcoustic data processing

a b s t r a c t

The DIDSON multi-beam sonar has become a popular tool for monitoring fish. The aim of this study wasto determine how well the DIDSON sonar can detect and measure fish during stationary applications inlakes. Cyprinid fish of known sizes (10–60 cm long) and with horizontal body orientations were deployedin known positions at two ranges (6.3 and 9.5 m) within a DIDSON high-frequency array of beams. A newmethod for estimating fish length from multi-beam sonars was thus developed and applied.

At both ranges, all the deployed fish were invariably detected when they were aligned with their sidesperpendicular to the beam. Increasing fish body angle, however, reduced the ability of the system to detectfish. Only the largest fish were detectable when the fish were parallel to the beam (head or tail aspect).The estimated fish length agreed well with the actual length when the fish were positioned perpendicularto the center of the multi-beam array. The lengths were underestimated for fish that were at the edges ofthe array, further away from the transducer or that had an increased body aspect, especially small-sizedfish. Additionally, we observed that the wide girth of large fish can shadow the rest of the body, resulting

in a decreased estimated length.

We showed that determining the actual length of fish is challenging and is not a trivial task, and we raisethe question of where to define fish length along the echo intensity of fish. We conclude that including theerror functions for length estimates allows the DIDSON to obtain more reliable and accurate biologicalinformation.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

Over the past decades, real-time acoustic-imaging technologyas rapidly developed, including mechanically scanned or multi-eam systems with a wide variety of frequencies, ranges andesolutions (Simmonds and MacLennan, 2005). The dual-frequencydentification sonar (DIDSONTM) system, which was originallyeveloped for the US Department of Defense (Belcher et al., 2001,002), is at the forefront of all imaging sonars.

This sonar is based on the concept of the ‘acoustic camera’

Jacobs, 1965; Smyth et al., 1963) and uses principles of multipleeams and scanning to generate an array of visualized points (i.e.,ixels). Due to the underlying complex beam-forming techniques

∗ Corresponding author. Tel.: +420 604344267; fax: +420 385310248.E-mail address: [email protected] (J. Kubecka).

ttp://dx.doi.org/10.1016/j.fishres.2014.02.031165-7836/© 2014 Elsevier B.V. All rights reserved.

that involve acoustic lenses, DIDSON can create high-resolutiondigital images, approaching optical quality, in which biologicallymeaningful specimens can be positively identified in a given envi-ronment. The operating frequencies are typically 0.7–1.8 MHz,which unfortunately reduces the observable range to a few tensof meters; however, this range is still substantially better thanwhat optical devices can achieve in dark or turbid water. Occupy-ing a niche between fisheries assessment (conventional) sonars andoptical systems; DIDSON can potentially be useful where detailedimaging is required over short ranges.

Video-like images, nevertheless, may lead users to believe thatthe obtained magnitude of size is easily measurable and reflectsreality. However, users should be aware that DIDSON images

remain acoustic-based and are consequently sensitive to charac-teristics such as beam arrangement, range and even the propertiesof the observed targets. Moursund et al. (2003) determined thatthe best geometry for obtaining high-resolution images of fish and

Page 2

M. Tuser et al. / Fisheries Research 155 (2014) 114–121 115

F framo

fatvmCcaFmlbSoaofidES

2

2

o(mwemd

2

ah

surface (i.e., measured from the surface to the midpoint of the DID-SON lens, Fig. 1). At this position of the horizontally aimed sonar,the garage floaters, water surface and bottom did not disturb thedata acquisition.

ig. 1. Scheme depicting the acoustic experiment construction with a rotatable fishf two vertical metal rods (A and B) and the middle of a horizontal rig (C).

or distinguishing them from other objects is when fish are alignedlong the 96-beam plane, as was accomplished in spawning migra-ion studies. In riverine research, many studies have attempted toerify how reliable fish sizes are, particularly in spawning escape-ent estimates (Baumgartner et al., 2006; Burwen et al., 2007;

ronkite et al., 2006). The advantage of riverine acoustic appli-ations is that fish swim predominantly aligned with the currentnd are therefore perpendicular to the acoustic beam (Burwen andleischman, 1998; Kubecka et al., 2000; Lilja et al., 2000). Moreover,igration studies usually examine single-species migratory popu-

ations with relatively large sizes detected in acoustically strongody aspects. In this study, we investigated the performance of DID-ON fish recording in lake conditions, where smaller fish could bebserved from acoustically less-reflective aspects. The aim was toscertain how the detection probability and length measurementsf fish using a DIDSON beam array are influenced by the actualsh size, spatial orientation and distance from the transducer. Weeployed fish of known sizes and species that typically occur inuropean lakes and reservoirs. The results address whether DID-ON can be used for quantitative studies in multispecies systems.

. Materials and methods

.1. Study site

The acoustic experiments were directly implemented in thepen water of the lacustrine region of the Rímov reservoirCzech Republic). The entire infrastructure for the experiment was

ounted on a floating boat garage (Fig. 1), which was a suitableorking platform that provided both electricity and shelter for the

quipment. The water column beneath the garage was approxi-ately 7.5 m deep, and its daily temperature profile was constant

uring the experimental period from June 20 to 23, 2007 (Fig. 2).

.2. Equipment

The acoustic data were collected using DIDSON, which gener-tes a fan-shaped multi-beam array across a 14◦ vertical and 29◦

orizontal sector. The sonar was operated at a 1.8 MHz-frequency

e. The anesthetized fish were tethered onto the fish frame between the lower ends

mode for a 0.3◦ horizontal resolution, which is the highest res-olution possible; in this mode, the array is horizontally dividedinto 96 elements (beams). The DIDSON is limited to 512 samplesper frame for each beam. We used a 10 m window length, wherethe down-range resolution per sample was 2 cm. The DIDSON unitautomatically selected a pulse length of 32 �s as a function of therange resolution and frequency. The image window was set to start2.9 m in front of the transducer, and the frame rate was adjusted toseven frames per second.

The DIDSON unit was mounted on a sub-Atlantic pan and tilt unitto allow precise aim and movement. Both devices were attachedto a vertical steel rod and deployed at a depth of 1.5 m below the

Fig. 2. The temperature-depth water profile depicts temperature layering duringthe acoustic experiment.

Page 3

116 M. Tuser et al. / Fisheries Resea

Table 1Fish deployed in the experiment with their measured body parameters. SL, standardlength; TL, total length.

Species SL (mm) TL (mm) Weight (g)

Bream Abramis brama 140 175 49250 310 373

Carp Cyprinus carpio 480 580 3468

Roach Rutilus rutilus 80 100 9118 145 36

miro(

2

tcictrd

cafFjtTrwitTtF

FD = pos

F(1)

F(e

225 270 253

The final piece of utilized equipment was a rotatable fish frameounted to an electric engine, which turned the fish frame around

ts vertical axis. The fish frame consisted of a wide horizontal metalig, which bore two metal rods vertically down into the water,ne on each side. These rods supported the lines that held the fishFig. 1).

.3. Experimental setup

This study included six fish from the family Cyprinidae, namelywo bream, Abramis brama (L.), one common carp, Cyprinus carpioarpio L., and three roach, Rutilus rutilus (L.), that were 10–60 cmn length and that weighed 9 g to 3.5 kg (Table 1). The fish wereaptured by electrofishing in a tributary of the Rímov reservoir. Allhe fish were enclosed in a special live pen placed directly in theeservoir to keep them in natural conditions until they could beeployed.

Prior to the experiment, each fish was anesthetized with tri-aine mesylate (MS-222), and their standard length, total lengthnd weight were measured. The anesthetized fish were then care-ully tethered in the fish frame with 0.3 mm fishing lines (also seerouzová et al., 2005). Briefly, three fishing lines were sewn into theaw, tail and spine of the fish. The head and tail lines were stretchedo the lower ends of the frame’s vertical metal rods (Fig. 1A and B).he spine line was attached to the middle of the frame’s horizontalig above the water (Fig. 1, point C). The head, tail and spine linesere slightly stretched to hold the fish in a straight upright position

n the water. The frame with the tethered fish was then lowered intohe water and moved to the required distance from the transducer.

he fish was brought to a depth of approximately 1.5 m. The dis-ance from the transducer was first set to 6.3 m and then to 9.5 m.or both distances, we performed two types of experiments.

ig. 3. Left: A DIDSON frame depicting a fish trace with the operator’s drawn line througright-left). The vertical dimension of the frame represents the range from the transducstimated length of the marked target using the intelligent line ruler, including the setup

rch 155 (2014) 114–121

In experiment 1, the fish frame was rotated to align the fishwith its side toward the transducer and with its head to the leftof the transducer. This position was locked to maintain this aspectconstantly throughout the experiment. First, the DIDSON’s tilt andpan angles were trimmed with the sub-Atlantic pan and tilt unitto have the fish vertically and horizontally centered in the beamarray. Second, the unit was panned to the left (anticlockwise) untilthe whole body of the fish was no longer visible on the DIDSONimages. After starting the recording, the transducer was succes-sively panned from left to right (clockwise) so that the fish initiallyappeared, passed through and then disappeared out of the beamarray on the other side as observed on the computer screen. Then,the rotation was reversed (from right to left) for a new recording sothat the fish passed the beam array back to its initial position beforethe second recording was stopped. Each recording took about oneand a half minutes. By rotating the DIDSON sonar, we ensured thatthe fish was constantly recorded with the same side aspect anddistance from the transducer throughout the entire beam array.

In experiment 2, the DIDSON’s pan angle was trimmed back sothat the fish was exactly centered in the middle of the multi-beamarray (horizontally and vertically). This time, the transducer washeld fixed in this position while the fish was rotated. Initially, thefish was positioned perpendicular to the transducer with its headpointing to the left of the transducer. At the start of the recording,the fish frame was rotated anticlockwise so that the fish was turnedwith its head toward the transducer first, then to the other sideaspect before approaching the tail-first position, and then back tothe initial position. One complete turn took 7 min, and two turnswere recorded per fish.

After completing both experiments at a distance of 6.3 m fromthe transducer, the frame with the fish was moved 9.5 m from theDIDSON unit, and both experiments were repeated.

2.4. Data processing

All the acoustic data were processed with the Sonar5 Pro post-processing software (Balk and Lindem, University of Oslo).

The probability of fish detection (FD) was determined as theproportion of positive observations of the fish compared to all theinvestigated frames, as follows:

F

tot

where Fpos is the number of frames with a positive observation ofthe fish and Ftot is the number of all investigated frames in 2.5◦

h the target. The horizontal dimension of the frame represents the X range domainer (Z domain). Right: The intensity of the fish trace along the drawn line with the

of the filter and threshold.

Page 4

Research 155 (2014) 114–121 117

o9taWwit

DfistawmsT“Ptrdanism(n

apfmolwtttil

wtpli(

puiawnD

sotimfs

Fig. 4. Observed lengths of all the fish images at side aspects deployed along theDIDSON’s horizontal plane for two distances from the transducer. The horizontalline indicates the true fish size. The vertical line depicts the middle of the DIDSON

M. Tuser et al. / Fisheries

ff-axis angle increments for Experiment 1 and 5◦ aspects (i.e.,0–94◦, 95–99◦ and such) for Experiment 2. A frame was consideredo be positive when a fish was conclusively observed on the recordnd measurable in accordance with the methods described below.hen a frame failed to meet the criteria explained below, the frameas considered to be negative and marked as “investigated” for

nclusion in the analysis. Each increment usually contained morehan 10 investigated frames.

The common method for measuring fish length and aspect inIDSON data is to use a computer mouse to draw a line along ash trace observed in a DIDSON frame viewer on the computercreen. An operator presses and holds down the mouse button athe beginning of the trace, drags the mouse cursor along the tracend releases the mouse button at the end of the trace. The soft-are then calculates the length and aspect of the drawn line. Thisethod relies on the operator’s ability to judge where a fish trace

tarts and ends as well as to be accurate in the drawing process.o ease the processing task, a new measuring method called theintelligent ruler” was developed and implemented in the Sonar5-ro software. With this approach, the operator still manually drawshe same line through the fish trace, but instead of being accu-ate with the start and end of the fish trace, the operator startsrawing somewhere in front of the fish trace in the image noisend finishes the line somewhere behind the fish trace in the imageoise. The software subsequently determines where the fish trace

s located and truncates the drawn line to this part based on theame criteria for all fish traces rather than on a subjective judg-ent of the operator. The operator immediately sees the result

Fig. 3) and can evaluate and overrun the automatic detection ifecessary.

An algorithm to determine the presence of fish uses a filter and threshold. The detection algorithm initiates by running a low-ass filter (mean window operator) through the frame to reducerequent variations in the intensity along the line. Minimum and

aximum levels are detected along the line, providing an estimatef local noise and signal levels. The threshold for determining theength is defined on the basis of a fixed level or a dynamic level,

hich is expressed as a ratio of minimum to maximum levels alonghe drawn filtered line. The start of the fish trace is defined ashe point where the filtered intensity along the line penetrates thehreshold level for the first time. The end of the trace is obtainedn the same way, but the search starts from the opposite end of theine.

In this study, a 3 × 3 (range × beam samples) running meanindow operator was applied for filtering. Each sample along

he drawn line corresponded to the average of its own valuelus those from eight surrounding samples. The signal to noise

evel threshold was adjusted to 50%, cutting the fish tracentensity halfway between the local noise and signal levelsFig. 3).

The data showed a stable noise level throughout the recordingeriod starting at about 80 DIDSON units (i.e., DIDSON returns val-es in the range from 0 to 240, which represent the returned echoes

n volt output from the AD-converter) at close range decreasing tobout 40 DIDSON units at 4 m and outward. At both ranges wheree positioned the fish the noise level was about the same. The sameoise levels were recorded both in the center and side beams of theIDSON beam array.

The length estimates were run with data that were uncompen-ated for acoustic transmission losses. The threshold and intensityf the DIDSON viewer display were tuned to optimize the con-rast of the fish image and the background. In cases when the fish

mage was somehow obscured by signs of the tether or freely swim-

ing fish, the operator skipped that particular frame and searchedor a better fish image in successive frames. Additionally, at theide beams of the DIDSON’s beam array, the fish was subject to

beam array. A = 10 cm roach, B = 14.5 cm roach, C = 17.5 cm bream, D = 27 cm roach,E = 31 cm bream, F = 58 cm carp.

processing only when the complete fish echo was imaged. Before afish image first encountered the edge of the DIDSON’s beam array,this image was the last one we measured and included in theanalyses. From this point onward, we stopped measuring any fishimage.

For statistical evaluation of the acoustic data, we used multipleregressions of Generalized Linear Models in the Wolfram Mathe-matica software (Oxfordshire, United Kingdom) to assess the error

of observed fish length as a function of its position within the beamarray, horizontal body orientation and distance from the trans-ducer.

Page 5

118 M. Tuser et al. / Fisheries Research 155 (2014) 114–121

the off

3

3

esfd(eft

L

wtrite

aete

3

ad(t

TSota

Fig. 5. A model for the error in the observed length depending on

. Results

.1. Position in the beam array

In Experiment 1, when the fish bodies were perpendicular tovery angle of the DIDSON transducer, all the fish deployed in thetudy were invariably detected at both 6.3 and 9.5 m. In addition,or all the fish, the length values were highest at the center andeclined as the fish were moved to the edges of the beam arrayFig. 4). This pattern was least apparent for the largest fish. Therror between the observed and true lengths was modeled as aunction of the off-axis position and distance, as best described byhe following quadratic equation:

E(�, R) = −(0.025 × �2) + (0.008 × �) − (0.098 × R) + 1.207 (2)

here LE (�, R) is the error of the observed fish length in centime-ers for a given off-axis angle and distance from the transducer, �epresents the off-axis angle of the DIDSON array in degrees and Rs the distance from the transducer in meters. In the model, onlyhe effect of � significantly accounted for the variability in lengthrror (Table 2).

At both ranges, the observed length was slightly overestimatedt the center of the DIDSON beam array and underestimated at thedges of the beam array (Fig. 5). The decline of the model curveoward the edges was slightly sharper at a distance of 9.5 m, but noffect of distance was significantly proven (Table 2).

.2. Fish body aspects

In Experiment 2, when fish in the middle of the beam array devi-

ted from the side aspect, the overall detection probability of fishepended on fish size and also on distance from the transducerFig. 6a). Fish aspect has also a strong influence on the detec-ion probability as shown in Fig. 6b and c. At the closer distance,

able 2ummary table of the regression results of a model given in Eq. (1) (DF, degreesf freedom; SS, sum of squares; MS, mean square), showing the model parametershat can best explain the variability in observed lengths of all the fish images at sidespects along the DIDSON’s horizontal plane. Significant results are in bold (p < 0.05).

DF SS MS F-Statistic p-Value

Range 1 9.8 9.8 1.4 0.246Off-axis angle 1 28.3 28.3 3.9 0.049Off-axis angle2 1 894.8 894.8 123.1 6.86E−26Error 534 3880.1 7.3Total 537 4813.1

-axis angle of the DIDSON at (a) a 6.3 m range and (b) 9.5 m range.

all the fish except for two of the smallest ones were successfullydetected at all body aspects. The 10 cm (smallest) and 14.5 cm roachfailed to be detected with the head-or-tail orientations (160–180◦,Fig. 6b). The effect of body aspect significantly increased when thefish were 9.5 m from the transducer. Fish detectability diminishedwith decreasing body size and increasing body aspect relative to thetransducer. The smallest fish, for instance, was only detected whenit was oriented at a near-side aspect (90–105◦). In contrast, onlythe 58 cm long carp was completely detectable in all body aspects(Fig. 6c).

Furthermore, the observed fish lengths were influenced by bodyaspect. Specifically, observed lengths decreased with an increase inbody aspect, more markedly with smaller fish and with fish thatwere further away from the transducer (Fig. 7). Additionally, thedecrease in length was asymmetrical in larger fish, most likely dueto the different body dimensions along the head-to-tail axis. Forexample, the observed length of the carp gradually declined as a“dome-shaped” curve when approaching the tail aspect; towardthe head aspect, the observed length was more stable until thepoint where it abruptly declined to half (see Fig. 7E). The widergirth of the frontal part of the body likely shadowed the thinnercaudal part. Furthermore, the 10 cm long roach was not included inFig. 7 because upon rotation, the fish images became only a singlepoint (pixel) devoid of an apparent prolonged body shape, and theyconsequently disappeared.

Using the following quadratic function, we modeled the errorof the length estimate as a function of fish body aspect and thedistance from the transducer:

LE(˛, R) = −(0.002 × ˛2) + (0.418 × ˛) − (1.822 × R) − 4.449 (3)

where LE (˛, R) is the error of the observed fish length in centime-ters for a given fish aspect and distance from the transducer, ˛represents the body aspect in degrees and R is the distance fromthe transducer in meters. For this model, all the tested effects sig-nificantly explained the observed variability in the data (Table 3).In the model, the observed lengths in the near-side aspects wereoverestimated at 6.3 m and slightly underestimated at 9.5 m (Fig. 8).

4. Discussion

In this study, we have shown that the DIDSON sonar is capable

of observing all deployed fish ranging from 10 to 60 cm in length.The detection and length estimates of the fish, however, markedlydepended upon their actual size, body orientation and distancefrom the transducer.

Page 6

M. Tuser et al. / Fisheries Research 155 (2014) 114–121 119

F veragw mirrori

wtpblbne

booobet

TSe

ig. 6. Detection probability of all the deployed fish at all body aspects for (a) the aere pooled because aspect relationships of the intervals 0–90◦ and 90–180◦ are

nterval 90◦ (side aspect) to 180◦ (head-tail aspect).

The detectability of all the fish posed no problem when the fishere perpendicular to the transducer at any point from the center

o the edges of the beam array. There can be no doubt that DIDSONossesses the best resolution for detecting a target along its 96-eam array horizontal plane (Moursund et al., 2003). However, the

ength estimates diminished for fish approaching the edges of theeam array. Fortunately, the error analysis showed that there waso additional effect of distance from the transducer on the lengthstimate.

The reason for the reduced fish length estimates at off-centeream positions can be explained by a combination of the intensityf individual beams and the manner of fish reflection. The intensityf individual beams drops with increasing distance from the center

f the beam array, ultimately resulting in a nearly a 20 dB differenceetween the center and side beams (Fig. 9). When we studied thecho intensity along a line drawn through a fish trace, we observedhat the intensity of the fish trace forms an arc rising up from the

able 3ummary table of regression results of a model given in Eq. (2) (DF, degrees of freedom; Sxplain the variability in observed lengths of all the fish images at all body aspects. Signifi

DF SS

Range 1 5263.8

Body aspect 1 787.7

Body aspect2 1 29 084.0

Error 1328 34 801.5

Total 1331 69 937.1

e of both distances from the transducer, (b) 6.3 m and (c) 9.5 m distances. The data images of each other (Frouzová et al., 2005); we converted all the aspects to the

noise, where the signal is robust before decaying back into the noise.If the arc trajectory is lowered so that it is mostly obscured by noise,the length of the observable part of the arc representing the observ-able signal will decrease. Relocating a fish from the center beamswith a high signal-to-noise ratio (SNR) to the side beams with alower SNR (i.e., weaker fish signal and the same noise, Fig. 10) willnaturally have the same effect, explaining the observed reductionin the estimated fish length with increased off-center position.

Another potential cause for the variable length estimates couldhave been the increased spacing between the beams at the edgesof the array (E. Belcher, pers. comm.). Nevertheless, this effect iscompensated in the software and is therefore not regarded as amajor factor.

As shown by the results from Experiment 2, when fish bodyaspect was altered, the detection probability and observed lengthof the fish dramatically declined with diminishing fish size andincreasing distance from the transducer. This effect could be due

S, sum of squares; MS, mean square), showing the model parameters that can bestcant results are in bold (p < 0.05).

MS F-Statistic p-Value

5263.8 200.9 1.45E−42787.7 30.1 5.01E−08

29 084.0 1109.8 2.21E−17726.2

Page 7

120 M. Tuser et al. / Fisheries Research 155 (2014) 114–121

Fig. 7. Observed lengths of five fish at all body aspects for both distances fromthe transducer. The 10 cm long roach was not included in this figure because uponrotation, the fish images became only a single point (pixel) devoid of an apparentprolonged body shape, and it consequently disappeared. The side aspect corre-sponds to 90◦ , the head aspect to 0◦ and the tail aspect to 180◦ . The horizontalline indicates the true size of the fish. The vertical line indicates the side aspect ofthe fish relative to the transducer. A = 14.5 cm roach, B = 17.5 cm bream, C = 27 cmroach, D = 31 cm bream, E = 58 cm carp.

Fig. 9. The intensity of the DIDSON 96-beam array calculated according to a ref-erence intensity (provided by MacArtney A/S on the behalf of Sound MetricsCorporation). In the experiment for the determination of DIDSON intensity pattern,the sonar was mounted on a tilt/pan rotator in a tank that was 4 m long, 1.5 m wideand 1 m deep. A small spherical stainless steel target (smaller in diameter than thebeam width of the sonar under test at the target range of 3 m) was mounted mid-depth at 3 m from the transducer. The sonar was rotated over the pan axis over arange from −20 to +20◦ , and the image was recorded. The target appeared as a brightspot in the image, moving across the field of view. The image file was processed toselect the maximum sample value at the target range as a function of angle. Thebeam pattern was thus a two-way pattern (transmit and receive combined), andthe signal level in dB was relative to a dynamic range of 0–90 dB and not an absolutesound pressure level (W.H. Hanot, pers. comm.).

Fig. 10. An illustration of the same fish at two cross-beam positions producingdifferent trace lengths due to different signal-to-noise ratios.

Fig. 8. A model for the error of the observed length for all body aspects at a (a) 6.3 m and (b) 9.5 m distance from the transducer.

Page 8

Resea

tbfiWrl

clpos

fitcpeeeoahdtlrmcdrttfiiiFoc

eaiNtfiwdhtfir

dtws

M. Tuser et al. / Fisheries

o a consequent decrease in scattered energy and also a decrease inoth the down- and cross-range DIDSON resolutions because thesh were further away from the transducer (Burwen et al., 2010).ith an increase in body aspect and distance from the transducer,

esolution is reduced because single beams collide with the targetess.

Interestingly, girth variation along the fish head-to-tail axisaused a drastic underestimation of the observed length. Thus, aong fish in head and tail orientations could be easily misinter-reted as a quite smaller fish. This acoustic shadowing effect wouldbviously be relevant for larger sized fish and would also likely bepecies-dependent.

Furthermore, we encountered another problem in determiningsh length. Although we used 0.3 mm thick fishing lines to tetherhe fish, we observed that they were visible to DIDSON in someases. At low fish signal conditions and when the tethers were per-endicular to the beam, it was sometimes difficult to resolve thechoes from the tethers from those of the fish. As a result, thesechoes from the tethers might have been incorporated into the fishcho and unintentionally elongated it. This incorporation of twobjects raises the important question of where to define fish lengthlong the intensity arc of the fish echo. The length measurementere was based on the echo intensity of the fish, firmly set to aynamic level of 50% between minimum and maximum values;his level seemed to be an appropriate threshold to obtain goodength estimates. A dynamic threshold level seemed to be moreobust toward noise than a fixed threshold level. Nevertheless,easurements of small-bodied fish can be particularly compli-

ated because their echo intensity is low relative to the noise. Theifference between a terminal pixel of the fish outline and the sur-ounding noise might be small in such cases, and this may lead tohe misinterpretation of the actual beginning and termination ofhe fish image. Additionally, SNR may also be affected by manualne-tuning of the DIDSON’s display controls (i.e., the threshold and

ntensity) for better user-defined visual contrast (i.e., altering min-mum and maximum levels for the resultant echo arc of the fish).or defining the true length of small-sized fish, advanced filteringf the DIDSON data would be helpful to enhance SNR, such as aross-filter analysis (Balk et al., 2009).

The intelligent ruler developed for this study worked well andased the work of measuring the fish lengths. Because the algorithmpplies the same detection criteria each time, it is possible that thentelligent ruler provides more objective results than an operator.evertheless, the length measurements still required an operator

o draw over the fish trace in the correct manner. In our study, thesh body was firmly attached in a straight, upright position, whichas advantageous for using the line approach. Yet, if the line wasrawn so that it cut through the fish trace and missed a part of theead or tail, the length would be underestimated. An alternativeo the line method would be to draw a rectangle around the entiresh trace, leaving the rest to the system. This rectangle mode hasecently been implemented in the software.

The current DIDSON capabilities and functionalities for both

ata handling and processing cannot be used for quantitative detec-ion and sizing of fish smaller than 20 cm from various aspects. Aseaker signals are not detected, both the detection probability and

izing accuracy are biased in a similar way. The study showed that

rch 155 (2014) 114–121 121

there is a need to improve SNR and the resolution of the system. Inspecial cases, the findings from this study can be applied to correctrecording biases or to estimate the risks of false observations. Theequations from this study can be used to recalculate observed sizesfor fish in side beams or given body aspects. For fish sizes that arenot detectable in certain body aspects, we can proportionally assesshow many fish could be missed in DIDSON recordings. However,this assessment of missed observations will require knowledge offish body orientations in a lake or its particular location, as bodyorientation can be dynamic and highly variable.

Acknowledgments

This study was financed by project No. CZ.1.07/2.3.00/20.0204(CEKOPOT) of the Ministry of Education, Youth and Sport of theCzech Republic. We would like to thank E. Belcher for his valuablecomments on our data, W.H. Hanot for providing information aboutDIDSON testing, Prof. M. Godlewska, Dr. M. Schmidt and anony-mous reviewers for useful comments to the manuscript and alsoDr. Hassan Hashimi for his careful editing of English.

References

Balk, H., Lindem, T., Kubecka, J., 2009. New cubic cross filter detector for multi-beam data recorded with DIDSON acoustic camera. In: Papadakis, J.S., Bjorno, L.(Eds.), Underwater Acoustic Measurements, Technologies and Results. Herak-lion, Greece, June, pp. 1461–1467.

Baumgartner, L.J., Reynoldson, N., Cameron, L., Stanger, J., 2006. Assessment of adual-frequency identification sonar (DIDSON) for application in fish migrationstudies. Fisheries Final Report Series No. 84. NSW Department of Primary Indus-tries, PO Box 21, 2230 Cronulla, NSW.

Belcher, E.O., Fox, W.L.J., Hanot, W.H., 2002. Dual-frequency acoustic camera: acandidate for an obstacle avoidance, gap-filler, and identification sensor foruntethered underwater vehicles. In: Proceedings of Oceans ‘02 MTS/IEEE Con-ference, Biloxi, MS, USA, 29–31 October, pp. 2124–2128.

Belcher, E.O., Matsuyama, B., Trimble, G., 2001. Object identification with acous-tic lenses. In: Proceedings of Oceans 2001 Conference, Honolulu, Hawaii, 5–8November, pp. 6–11.

Burwen, D.L., Fleischman, S.J., 1998. Evaluation of side-aspect target strength andpulse width as potential hydroacoustic discriminators of fish species in rivers.Can. J. Fish. Aquat. Sci. 55 (11), 2492–2502.

Burwen, D.L., Fleischman, S.J., Miller, J.D., 2007. Evaluation of a dual-frequency imag-ing sonar for detecting and estimating the size of migrating salmon. Fishery DataSeries No. 07-44. Alaska Department of Fish and Game, Division of Sport Fish,Anchorage.

Burwen, D.L., Fleischman, S.J., Miller, J.D., 2010. Accuracy and precision of salmonlength estimates taken from DIDSON sonar images. Trans. Am. Fish. Soc. 139,1306–1314.

Cronkite, G.M.W., Enzenhofer, H.J., Ridley, T., Holmes, J., Lilja, J., Benner, K., 2006. Useof High-Frequency Imaging Sonar to Estimate Adult Sockeye Salmon Escape-ment in the Horsefly River, British Columbia, iv+47 pp.

Frouzová, J., Kubecka, J., Balk, H., Frouz, J., 2005. Target strength of some Europeanfish species and its dependence on fish body parameters. Fish. Res. 75, 86–96.

Jacobs, J.E., 1965. The ultrasound camera. Science (Lond.) 1, 60–65.Kubecka, J., Frouzová, J., Vilcinskas, A., Wolter, C., Slavík, O., 2000. Longitudinal

hydroacoustic survey of fish in the elbe river supplemented by direct capture. In:Cowx, I.G. (Ed.), Management and Ecology of River Fisheries. Blackwell ScienceLtd., pp. 14–25.

Lilja, J., Marjomäki, T.J., Riikonen, R., Jurvelius, J., 2000. Side-aspect target strength ofAtlantic salmon (Salmo salar), brown trout (Salmo trutta), whitefish (Coregonuslavaretus), and pike (Esox lucius). Aquat. Living Resour. 13, 355–360.

Moursund, R.A., Carlson, T.J., Peters, R.D., 2003. A fisheries application of a dual-

frequency identification sonar acoustic camera. ICES J. Mar. Sci. 60, 678–683.

Simmonds, E.J., MacLennan, D.N., 2005. Fisheries Acoustics: Theory and Practice.Blackwell Science, Oxford, UK, pp. 437.

Smyth, C.N., Poynton, F.Y., Sayers, J.F., 1963. The ultrasound image camera. Proc. Inst.Electr. Eng. 110, 16–28.