Journal of Plankton Research Vol.18 no.8 pp.1471-1484, 19%

Diffraction patterns as a tool to recognize copepods

Victor Antonio Zavala-Hamz, Josud Alvarez-Borrego1 and Antonio Trujillo-

Ortfz2

Centro de Investigacidn Cientifica y de Educacidn Superior de Ensenada,Divisidn de Oceanologia, Departamento de Ecologla Marina, 'Divisidn de FlsicaAplicada, Departamento de Optica, Km. 107 Carretera Tijuana-Ensenada,Ensenada and 2Universidad Autdnoma de Baja California, Facultad de CienciasMarinas, Km. 103 Carretera Tijuana-Ensenada, Ensenada, BC, Mixico

Abstract Photographs of three species of calanoid copepods were digitized, binarized and edited toremove the debris present in the images and to simulate different degrees of segmentation. Twenty-eight segmented images of the three species were generated after pre-processing to obtain their digitalFourier transform or diffraction pattern. Ouster analysis was carried out for the images of the cope-pods and for their diffraction patterns to determine which set of images discriminated genus, speciesand sex of the selected organisms. The dendrograms of the diffraction patterns had more power of dis-crimination and 50% less variability. This method is promising, but more research is necessary in orderto implement an automated system for plankton identification in the near future.

Introduction

Even for the most competent plankton taxonomist, the correct identification of amarine microorganism can take a long time (Ortner et ai, 1979; Omori and Ikeda,1984). In recent years, there has been a decrease in the number of competentplankton taxonomists (Simpson et ai, 1992), which could be the reason for someof the reported synonymies for calanoid copepods. Bjornberg (1981) reports thatAcartia lilljeborgii is the same as Acartia denticomis and Nannocalanus minor isthe same as Calanus vulgus, and Trujillo-Ortfz (1990) reports that Acartia cali-forniensis is the same as Acartia bacorehuisensis.

A broad group of researchers have been trying to implement automatedsystems in order to identify and count organisms in a faster and more accurateway. These systems include the Coulter-Counter (Sheldon and Parsons, 1967), thein situ Coulter-Counter (Herman and Dauphinee, 1980; Herman and Mitchell,1981), silhouette photography (Ortner, 1979) and the video camera (Liacopoulos,1983; Latrous, 1984; Rolke and Lenz, 1984). However, none of these systemsensure complete effectiveness of the taxonomic classification due to the orien-tation of the organism or the resolution of the image.

Another group of researchers have begun to use coherent optical processing forthe recognition of planktonic organisms (Cairns et al., 1972; Almeida et al., 1976;Jeffries et al., 1980,1984). A very important part of these techniques is the acqui-sition of a Fourier transform or diffraction pattern of the organism. The rapiddevelopment of computational sciences has made image-processing softwareavailable to almost everyone in order to obtain diffraction patterns and otherimage-processing tools. Coherent optical processing techniques were introducedby Vander Lugt (1964). The objects they used to test their systems were images of

O Oxford University Press 1471

V.A.Zavala-Hamz el al.

SELECTION OFORGANISMS AND

PHOTOGRAPHS

DIGITAL BINARYCONTOURS AND

SCANNING

DIGITAL DIFFRACTIONPATTERNS

CORRELATIONS ANDDENDROGRAMS OFORGANISMS AND

DIFFRACTION PATTERNS

Complate Linkage

Organism*

Fig. L Flux diagram of the different steps involved in the methodology.

1472

Diffraction pattern recognition of copepods

letters or airplanes. All of these projects were supported by the military defenceprogrammes of countries such as the USA or Canada (Caulfield and Maloney,1969; Casasent and Psaltis, 1976; Leclerc et al., 1991).

The Plymouth group, in the UK, has recently proposed the use of differentneural networks and multivariate analyses for biological pattern recognition(Simpson et al., 1992; Culverhouse et al., 1994; Williams et al., 1994). They havetested their systems successfully using pre-processed images of dinoflagellates andtintinnids (Fourier transforms as a function of the shape of the organism), obtain-ing >70% correct categorization. The network requires a pool of representativetraining data provided by plankton taxonomists which categorized the imageswith >91% consistency. In order to create an accurate data bank, the imageslacking 100% consistency amongst the experts were discarded. They state thatlarger sets of data will become more representative of reality so they are workingto address the 'real time' identification of microorganisms, yet still face problemssuch as scale or orientation in the image.

The calanoid copepods are holoplanktonic organisms that dominate most ofthe seas of the world. In temperate and cold regions, they are extremely abun-dant. The genus Calanus is the principal representative, as it inhabits all of theoceans and is abundant in excess (Raymont, 1983). Calanus pacificus is the mostabundant copepod in the zooplankton samples from the California Currentregion (Frost, 1972). Acartia californiensis is one of the most abundant copepodsin coastal lagoons of the north-eastern Pacific Ocean and Acartia tonsa is a mostwidely spread calanoid copepod, but also lives in coastal lagoons of the north-eastern Pacific Ocean. In some lagoons, it is the dominant copepod (Trujillo-Orti'z, 1986).

The aim of this paper is to explore the possibility of using digital diffractionpatterns as a basis for an automated system for plankton identification. Clusteranalysis is an approach to assess the power of discrimination of the proposedmethod.

Method

Figure 1 shows the different steps involved in the methodology.

Photographs

Male and female adults of the calanoid copepods Calanus pacificus, Acartia tonsaand A.californiensis were separated from several plankton samples. Each organ-ism was photographed in the same orientation, keeping its natural proportion,using a Wild Heerbrugs MSA model stereomicroscope with power source andKodak Technical-Pan 2415 film. After all the organisms had been photographed,the film was developed and the negatives printed on paper. We chose not to usereplicates because, at the moment, we are not interested in the individual varia-tions in morphology of the selected species. The most typical specimen of eachspecies was selected to continue the pre-processing (Figure 2).

1473

V.A.Zavala-Hamz el al

T TB

T

Fig. 2. Contours of the three selected species of calanoid copepods. (A) Cpacificus, female; (B)Cpacificus, male; (C) A.califomiensis, female; (D) A.californiensis, male; (E) A.lonsa, female; (F)A.tonsa, male.

Pre-processing

The next step was to digitize the six printed images with a Hewlett Packard ScanJet Plus scanner (all the images were 256 X 256 pixels in size) and draw binarycontours (white outside and black inside) with a personal computer to avoid noisefrom the background of the sample resulting from air bubbles or debris. Com-mercial drawing software was used to simulate different degrees of segmentationin the images of the organisms (Table I). Finally, all the images were transferredto a SUN system to obtain their fast Fourier transform or diffraction pattern usingimage-processing software.

Table L Segmentation variables for the selected organisms

Genus

AcartiaAcartiaAcartiaAcartiaAcartiaCalanusCalanusCalanusCalanus

Variable no.

113-4511S4

Status of the organism

CompleteWithout one first antennaWithout both first antennaeWithout the urosomeWithout antennae and urosomeCompleteWithout appendagesWithout appendages and first antennaeWithout appendages and urosome

1474

Diffraction pattern recognition of copepods

Table IL Colour code for the dendrograms of the organisms and the diffraction patterns

JFemale of Arartia r.alifnrniensis

Male of Acartia raliln

Species

Female of Arartia tonsa

Male of Arwrtia Inntit

Spedes touu

Genus Acartia

Female of Calanus pacificus

Male of Calanus nacifinns

Genus r«i»mig

Statistical analysis

Correlation is the best way to compare at least two sets of data. In our case, theimages of the organisms were the first set of data and their corresponding dif-fraction patterns were the second set of data. A cross-correlation for each set ofdata was carried out using the STATIST!CA for Windows program.

Several dendrograms or clusters were elaborated in order to determine whichset of data could discriminate genus, species and sex of the selected organisms,depending on the variables of segmentation considered (Table I). Table II pre-sents the colour code selected to show the discrimination of genus, species and sexamong the dendrograms for the images of the organisms and for their diffractionpatterns.

1475

V.A.Zavala-Hami tt al

A1

A2

A3

A4

A5

Fig. 3. Segmentation pattern for A.californiensis, female. Al = complete; A2 = without one firstantenna; A3 = without both first antennae; A4 = without the urosome; A5 = without the first antennaeand urosome (just the body).

B1

B2

B3

B4

B5

Fig. 4. Segmentation pattern for A.califomiensis, male. Bl = complete; B2 = without one first antenna;B3 = without both first antennae; B4 = without the urosome; B5 = without the first antennae andurosome (just the body).

1476

Diffraction pattern recognition of copepods

C1i

C2

C3

\C4

C5

Fig. 5. Segmentation pattern for A.tonsa, female. Cl = complete; C2 = without one first antenna; C3 =without both first antennae; C4 = without the urosome; C5 = without the first antennae and urosomeGust the body).

The goal of a dendrogram is to form groups, each containing more homo-geneous characteristics inside it and at the same time being more heterogeneousamong themselves (Zupan, 1982). Using STATISTICA for Windows, the Com-plete Linkage method with Euclidean distances method was used.The smaller thedistance represented, the more similarity amongst the objects.

Results and discussion

A total of 56 images were generated, 28 for the organisms and 28 for their respec-tive diffraction patterns (Figures ?-8).

It is important to note that all of the organisms have the same orientation, sothe differences in the correlations are due to the organisms themselves, not totheir angle of orientation.

A 28 X 28 correlation matrix was generated for each set of data (organisms anddiffraction patterns). In general, the correlation values for each image of theorganisms were higher than those for the images of the diffraction patterns Acomparison between both sets of correlations was carried out to test the hypoth-esis that there was no difference between the data sets The result was 0.66(P < 0.05), so there was no significant difference between the images of the organ-isms and the images of their diffraction patterns.

A total of 12 dendrograms were computed, six for each data set. They showeddifferent associations amongst both sets of data and were coloured to discriminatecharacteristics such as sex, species and genus of the different organisms (Table II).

1477

V.A.Zivala-H«mz et aL

D1

D2

"~ D3

-i D4

• D5

Fig. 6. Segmentation pattern for A.tonsa, male. Dl = complete; D2 = without one first antenna; D3 =without both first antennae; EM = without the urosome; D5 = without the first antennae and urosome(just the body).

E1

E2

E3

E4

Fig. 7. Segmentation pattern for Cpadficus, female. El = complete; E2 = without appendages; E3 =without appendages and the first antennae; E4 = without appendages and urosome (just the body).

1478

Diffraction pattern recognition of copepods

F1

F2

F3

F4

Fig. 8. Segmentation pattern for C.pacificus, male. Ft = complete; F2 = without appendages; F3 :without appendages and the first antennae; F4 = without appendages and urosome (just the body).

The x-axes represented the different segmentation variables involved (Table I)and the y-axes the linkage (Euclidean) distances (Figures 9-11).

Table III presents the linkage distances for the six complete copepods. Both setsof data discriminated the genus Acartia (Al, Bl, Cl and Dl) from the genusCalanus (El and Fl). The linkage distances for the images of the diffraction pat-terns were lower than the distances for the images of the organisms. Diffractionpatterns discriminated the female and male of A.californiensis (Al, Bl; 0.76 dis-tance) from the female and male of A.tonsa (Cl, Dl; 0.83 distance), as well as from

Table IIL Linkage distance matrixes for the six complete copepods and their diffraction patterns

Diffraction patterns

Variable A l Bl Cl Dl El Fl

AlBlCl

mElFl

Organisms

AlBlO

mElFl

0.000.760.830.830.940.87

0.001.271.021.131.432.43

0.000.750.781.050.97

0.000.781.091.242.44

0.000.801.040.97

0.000.801.292.64

0.001.121.04

0.001.062.51

0.000.73

0.001.77

0.00

0.00

1479

V.A.ZavaIa-H«mz el al

Complete Linkage

Diffraction Patterns

Complete Linkage

Organisms

Fig. 9. Complete linkage dendrograms with all the variables of segmentation for the organisms (a) andtheir diffraction patterns (b).

Complete Linkage Complete Linkage

s

Diffraction Patterns Organisms

b

Fig. 10. Complete linkage dendrograms without variables 4 and 5 for the organisms (a) and their dif-fraction patterns (b).

Complete Linkage Complete Linkage0 66

E4 B5 A5

DlrrsGbon PuUvtiis Organisms

Fig. 1L Complete linkage dendrograms for the organisms with a major degree of segmentation (a) andtheir diffraction patterns (b).

the male of Cpacificus (Fl; 0.87 distance) and from the female of C.pacificus (El;0.94 distance).

Table IV presents the linkage distances for the three complete females. Bothsets of data discriminated the genus Acartia (Al and Cl) from the genus Calanus(El). The linkage distances for the images of the diffraction patterns were lowerthan the distances for the images of the organisms, and both are consistent with

1480

Diffraction pattern recognition of copepods

Table IV. Linkage distance matrixes for the three complete female copepods and their diffractionpatterns

Diffraction patternsVariable

AlClEl

Organisms

AlClEl

Al

0.000.830.94

0.001.021.43

Cl

0.001.04

0.001.29

El

0.00

0.00

Table V. Linkage distance matrixes for the three complete male copepods and their diffractionpatterns

Diffraction patternsVariable

BlDlFl

Organisms

BlDlFl

Bl

0.000.780.97

0.001.092.44

Dl

0.001.04

0.002.51

Fl

0.00

0.00

the distances shown in Table III. Diffraction patterns discriminated the female ofA.californiensis (Al) from the female of A.tonsa (Cl; 0.83 distance), as well asfrom the female of C.pacificus (El; 0.94 distance).

Table V presents the linkage distances for the three complete males. Both setsof data discriminated the genus Acartia (Bl and Dl) from the genus Calanus (Fl).The linkage distances for the images of the diffraction patterns were lower thanthe distances for the images of the organisms, and both are consistent with the dis-tances shown in Table III. Diffraction patterns discriminated the male of A.cali-forniensis (Bl) from the male of A.tonsa (Dl; 0.78 distance), as well as from themale of C.pacificus (Fl; 0.97 distance).

With the dendrograms containing the 28 images of the organisms and their 28diffraction patterns (Figure 9a and b), we were able to discriminate genus, speciesand sex of the organisms. The images of the diffraction patterns grouped the seg-mented female of A.californiensis at a 0.59 distance (Al, A4 and A2 coloured inred), the male at a 0.5 distance (Bl, B4 and B2 coloured in orange), and both sexeswere joined at a 0.8 distance to form the species californiensis (coloured inyellow). The segmented female of A.tonsa was grouped at a 0.7 distance (Cl, C2and C4 coloured in light blue), the male at a 0.42 distance (Dl, D4 and D2coloured in light grey), and both sexes were joined at a 0.8 distance to form the

1481

V.A.Zavals-Hamz el at

species tonsa (coloured in pink). The genus Acartia (coloured in black) wasgrouped at a 0.82 distance. The segmented male of Cpacificus (Fs) was colouredbright green, females (Es) pale green and the genus Calanus in blue. The male ofA.tonsa without both antennae (D3) appeared closer to Cpacificus because atthat level of segmentation it was difficult to differentiate.

The images of the organisms did not discriminate genus, sex and species;however, the diffraction patterns did (Figure 9b). The difference between thelinkage distance scale in the dendrograms of the diffraction patterns and theimages of the organisms was almost 50%.

After extracting segmentation variables 4 and 5 (Table I), the dendrograms forthe diffraction patterns (Figure 10a) still discriminated genus, species and sex. Thelinkage distances for the diffraction patterns were almost equal to the scale ofFigure 10a. The dendrogram for the images of the organisms (Figure 10b) dis-criminated only genus, but not species or sex, and the linkage distance scale wasslightly different to the scale of Figure 9b.

Finally, we tested the power of discrimination using only the images with themajor degree of segmentation. The diffraction patterns (Figure lla) failed to dis-criminate the two species of the genus Acartia. The dendrogram grouped thefemale (A5) of A.californiensis with the male (D5) of A.tonsa at a 0.59 distance,and grouped the male (B5) of A.californiensis with the female (C5) of A.tonsa ata 0.57 distance. Although the dendrogram for the organisms (Figure lib) groupedthe genus Acartia better (coloured in black) at a 0.99 distance and groupedCpacificus at a 0.78 distance (coloured in blue), the variability of the linkage dis-tance (y-axis) is almost 50% less in the diffraction patterns.

The images of the diffraction patterns discriminated genus, species and even sexof the selected species. The segmentation variables gave the method more powerof discrimination. This power of discrimination diminished as more segmentationvariables were extracted.There is 50% less variability in the linkage distance scaleof the dendrograms for the diffraction patterns. This difference in the scaleensures a better discrimination using the diffraction patterns.

Conclusion

(i) Digital diffraction patterns of three species of calanoid copepods wereobtained for the first time.

(ii) At first glance, the correlations of the images of the organisms and the cor-relations of the diffraction patterns seemed to have no significant difference(r = 0.66).

(iii) The cluster analysis permitted us to ensure that the images of the diffractionpatterns have 50% less variability than the images of the organisms.

(iv) The diffraction patterns differentiate the genera Acartia and Calanus. It waspossible to differentiate more information, such as species or even sex, withthe inclusion of more than one segmentation variable.

(v) The potential of the diffraction patterns as a tool for taxonomy is promising.This method could be more useful using rotation or scale invariant correla-tions instead of simple correlations.

1482

Diffraction pattern recognition of copepods

Acknowledgements

TTie authors would like to thank Octavio Meillon Menchaca for the photographicwork and Carlos Lopez Famozo for the software to obtain the diffraction pat-terns.

ReferencesAlmeida,S.P, LuJ.K.T, LaiJ'.E, CairnsJ.Jr and Dickson.K.L. (1976) In Maron,E., Friesem.A.A. and

Wiener- AuneaivE. (eds), Applications of Holography and Optical Data Processing. Pergamon Press,New York, pp. 573-579.

Bj8rnberg,T.K.S. (1981) Copepoda. In BoltovskoyJ). (ed.), Atlas del zooplancton del Atldnticosudoccidental y mitodos de trabajo con el zooplancton marino. Publication especial del INIDEP,Mar de la Plata, Argentina, pp. 587-679.

CairnsJ.Jr, DicksonJ.K.L., Lanza,G.R., Almeida.S.P. and del BalzoJ). (1972) Coherent optical spatialfiltering of diatoms in water pollution monitoring. Arch. Mikrobiol., 83,141-146.

Casasent.D. and Psaltis,D. (1976) Scale invariant optical correlation using Mellin transforms. Opt.Commun., 17,59-63.

Caulfield,HJ. and Maloney.W.T. (1969) Improved discrimination in optical character recognition.Appl Opt., 8,2350-2356.

Culverhouse,P.F., Ellis,R., Simpson.R.G., Williams,R., Pierce.R.W. and TurnerJ.T. (1994) Automaticcategorisation of five species of Cymatocylis (Protozoa, Tintinnida) by artificial neural network.Mar. Ecol. Prog. Ser, 107,273-280.

Frost,B.W. (1972) Effects of size and concentration of food particles on the feeding behaviour of themarine planktonic copepod Calanus pacificus. Limnol. Oceanogr., 17,805-815.

Herman.A.W. and Dauphinee.T.M. (1980) Continuous and rapid profiling of zooplankton with anelectronic counter mounted on a 'Batfish' vehicle. Deep-Sea Res., 28,79-96.

Herman.A.W. and Mitchell,M.R. (1981) Counting and identifying copepod species with an in situ elec-tronic plankton counter. Deep-Sea Res., 28,739-755.

Jeffries,H.P, Sherman,K., Maurer.R. and Katsinis,C. (1980) Computer-processing of zooplanktonsamples. In Kennedy,V. (ed.), Estuarine Perspectives. Academic Press, New York, pp. 303-316.

JeffrieiH.P, Berman,M.S., Poularikas^i.D., Katsinis,C, MelasJ., Sherman,K. and Bivins,L. (1984)Automated sizing, counting and identification of zooplankton by pattern recognition. Mar. BioL, 78,329-334.

Latrous,S. (1984) Pattern Recognition and Automatic Zooplankton Classification by Image Analysis.Rennes-I University, Rennes, France, 117 pp.

Leclerc,L., Sheng,Y. and Arsenault.H.H. (1991) Circular harmonic covariance filters for rotationinvariant object recognition and discrimination. Opt. Commun., 85,299-305.

Liacopoulos,C. (1983) An automatic on board system for identifying and counting plankton. Rapp.P-V. Reun. CIESM, 28,111-112.

Omori.H. and Ikeda.T. (1984) Methods in Marine Zooplankton Ecology. John Wiley and Sons, USA,382 pp.

Ortnerj"., Cummings,S.R., Aftring,R.P. and EdgertonJvl.E. (1979) Silhouette photography of oceaniczooplankton. Nature, TTI, 50-51.

RaymontJE.GJ. (1983) Plankton and Productivity in the Oceans. 2nd edn. VoL 12: Zooplankton.Pergamon Press, New York, 824 pp.

Rolke,M. and LenzJ. (1984) Size structure analysis of zooplankton samples by means of an automatedimage analyzing system. J. Plankton Res., 6,637-645.

Sheldon.R.W. and Parsons,T.R. (1967) A Practical Manual on the Use of the Coulter Counter. MarineScience, Coulter Electronics, 66 pp.

Simpson^R.,Williams,R., Elu^R- and CulverhouseJ'.F. (1992) Biological pattern recognition by neuralnetworks Mar. Ecol Prog. Ser., 79,303-308.

Trujillo-Ortlz^A. (1986) Life cycle of the marine calanoid copepod Acartia califomiensis, Trinast,reared under laboratory conditions. CalCOFI Rep., XXVII, 188-204.

Trujillo-Ortiz^A. (1990) Porciento de eclosi6n, produccidn de huevos y tiempo de desarrollo deAcartia califomiensis,Trinast (Copepoda, Calanoida) bajo condiciones de laboratorio. Cienc Mar,16,1-22.

Vander LugtA- (1964) Signal detection by complex spatial filter. IEEE Trans. Inf. Theory, IT-10,139-145.

1483

VA^irala-Hamz el aL

Williams,R., McCall,H., Pierce,R.W. and TbrnerJ.T. (1994) Speciation of the tintinnid genus Cymaio-cylisby morphometric analysis of the loricae. Mar. EcoL Prog. Ser., 107,263-272.

ZupanJ. (1982) Clustering of LMrge Data Sets. Research Studies Press, New York, 122 pp.

Received on July 6, 1995; accepted on March 26, 1996

1484