Measurement of Spaghetti Speck Count, Size, and Color Using an Automated Imaging System 1

Transcription

1 Mesurement of Spghetti Speck Count, Size, nd Color Using n Automted Imging System 1 S. J. Symons, 2,3 G. Venor, 4 L. Vn Schepdel, 2 nd M. A. Shhin 2 ABSTRACT Cerel Chem. 86(2): An objective imging method ws developed to count drk specks in spghetti. The method simultneously mesured individul speck size nd color nd the overll color of the spghetti. Spghetti smples were prepred from durum whet smples collected from the Pririe Registrtion Recommending Committee for Grins (PRRCG) durum whet coopertive trils during four consecutive crop yers from 2002 inclusive to Differences in speck counts were found between smples within ech yer. From yer to yer, the bseline for speck counts vried with the highest numbers in 2005 nd the lowest numbers in For comprison, three technicins lso counted the number of specks in ech smple. These visul counts were not consistent between technicins or technicin to the imging method, supporting the need for this objective pproch. Spghetti speck counts did not relte to the speck counts of the semolin subsequently used to prepre the product. Speck sizes were consistent cross smples nd between yers, indicting consistent milling method for ll the smples. Differences in speck count numbers could not be ttributed to differences in speck color or pst color. The imging method gve very consistent speck counts nd color mesurements over the four yers. Spghetti is n extruded nd dried food product mde from durum whet semolin. Dried spghetti qulity for the consumer is initilly bsed on ppernce nd then, following cooking, on texture nd tste. The biggest fctor in ppernce is product color, lthough excessive drk or light regions (specks) on the surfce cn led to low product ppel. In the lbortory, color is typiclly determined using spectrometer giving color vlues in the L**b* color spce (Dexter et l 1997). Speckiness of the dried pst product my not be directly ssessed but indirectly determined by counting the specks of the rw semolin. During durum whet milling to produce semolin, the imperfect seprtion of semolin (strchy endosperm) from brn cuses brn frgments to be imprted s impurities in the milled product. Any discolortion or contmintion on the surfce of the brn cn increse the visibility of these impurities, which re typiclly visible s specks in the semolin. Brn specks re normlly yellow to brown in color, while specks with dditionl contminnts re often brown to blck in color. The presence of specks in the semolin hs negtive impct on semolin vlue becuse they pper s brown or drk specks in the pst. Correspondingly, due to consumer perception tht specks in pst re sign of poor qulity, the product vlue is deprecited. Considering this significnce, count of the number of specks in semolin is usully conducted s prt of the qulity control process during semolin production nd is often primry milling specifiction tht is set by the processor. The exct method for counting visible specks in semolin vries (Vsiljevic nd Bnsik 1980; Dexter nd Mtsuo 1982; El Bouziri nd Posner 1988), lthough it cn be chieved objectively using imge nlysis bsed instruments (Hrrign 1995; Symons et l 1996). Using fluorescence imging pproch, the Brnscn F4000 ( instrument detects nd mesures brn contminnts in semolin during the milling process. No known objective systems re in use for instrumentlly determining the speck count nd size distribution in pst. The detection of brn frgments hs lso been reported in common whet flours using the imging methods bsed on white light illumintion (Evers 1993; Whitworth 1994) nd commercilized using fluorescence methods (Hrrign 1995). 1 GRL publiction number Cndin Grin Commission, GRL, Winnipeg Mnitob R3C 3G8 Cnd. 3 Corresponding uthor. Phone: Fx: E-mil ddress: stephen. symons@grinscnd.gc.c 4 Stzione Sperimentle di Grnicoltur per l Sicilli, Cltgirone, Itly. doi: / CCHEM This rticle is in the public domin nd not copyrightble. It my be freely reprinted with customry crediting of the source. AACC Interntionl, Inc., Routine visul speck counting of semolin is subjective nd bsed on the experience of the observer nd is influenced by such fctors s ftigue nd the knowledge of customer requirements. This process remins subject to bis tht is bsed on speck size nd color (Symons et l 1996). Similrly, visul speck counting in spghetti is lso ffected. Spghetti is trnslucent nd dense product resulting from the hydrtion, extrusion, nd drying process involved to mke it. Brn frgments below the surfce of spghetti my be detected in contrst to the opque semolin. It is lso uncler how the speckiness of semolin reltes to the speckiness observed in the finl pst product, lthough there is n ssumption of reltive direct reltionship between the products. The objective of this work ws to estblish n objective imging bsed method for counting specks in spghetti tht cn clerly define the speck size nd drkness. Additionlly, to combine this determintion with the mesurement of spghetti color, which is n dditionl qulity prmeter, currently determined using Minolt spectrophotometer. MATERIALS AND METHODS Spghetti Smples Durum whet smples from the 2002, 2003, 2004, nd 2005 Pririe Registrtion Recommending Committee for Grins (PRR CG) durum whet coopertive trils (Clrke et l 1998) were used in this study. The durum whet ws milled ccording to Dexter et l (1990). Spghetti ws prepred ccording to the GRL method (Mtsuo et l 1972) nd ws dried t 70 C using n Afrem lbortory scle drier (CGC Methods; Clrke et l 1998; Ingelbrecht 2002). Spghetti ws produced in duplicte from ech semolin. Spghetti strnds were tped to stndrd white crd (23 60 mm) using double-sided tpe, such tht djcent stnds were touching. The strnds were ligned prllel to the longest dimension. Color ws determined using Minolt CM-525i spectrophotometer. The procedure is described online t ( c/qulity/whet/methods_tests/ spghetti_ clr_mthd1-e.htm). Semolin Speck Counting Semolin speck counts were determined by the GRL method by ssessing smooth 25 cm 2 surfce through n illuminted 1.5 mgnifier (Dexter nd Mtsuo 1982). Spghetti Visul Speck Counting The crd-mounted spghetti smples were visully inspected by three technicins under lmp mgnifier using 5,000 k fluorescent lighting. A mm blck msk ws plced centrlly over 164 CEREAL CHEMISTRY

2 the spghetti smple before counting. Becuse no stndrdized spghetti speck counting procedure is vilble, following trining, ech technicin ws llowed to determine speck counts independently, introducing the widest possible vrition between evlutions in this process. For 2002 to 2004 inclusive, the three ssessing technicins were the sme. For the 2005 smples, only technicin 1 ws common to the previous yers nd only one dditionl technicin ws vilble. Objective Speck Counting The retired imging method developed in our lbortory for semolin (Symons et l 1996) nd dpted to Asin noodles using flt-bed scnner for imging (Htcher et l 2004) ws modified to detect nd count specks in spghetti strnds. The spghetti crd ws scnned t 400 dpi in 24-bit color over n re of mm using ScnMker 4 flt-bed document scnner (Microtek USA). The scnned imge ws color-corrected using correction curve generted from reference Kodk Q60 pper chrt ccording to the method of Shhin nd Symons (2003). The imge ws smoothed nd corrected for ny vrition in illumintion intensity nd the drk specks segmented from the underlying bckground using gry level imge. Two binry msks were creted from this grey level imge, one identifying the drk specks s regions of interest nd the other defining the spghetti less the specks s region of interest. From trils over severl smples, grey vlues for the segmenttion process were selected to represent the best generlized identifiction of the drk specks in spghetti strnds. As previously detiled, segmenttion ws n utomted process determined by the density of the bse smple, nd specks were determined by drkness reltive to this bse (Htcher et l 2004). Color of spghetti smples ws mesured on the color-corrected originl smple imge. The color of the specks or spghetti, less the specks, ws determined using binry msks creted from the grey level imge. Color mesurements were mde using both RGB (red, green, blue) nd HLS (hue, lightness, sturtion) color models s implemented in the KS400 softwre. Imge Processing All imge processing nd mesurement ws performed in the KS400 mcro imge processing environment (Crl Zeiss, Oberkochen, Germny). Experimentl Design nd Sttisticl Anlyses All experiments were performed in rndom block design. Mesurement errors were determined by repetedly mesuring the sme smple without movement from the scnner bed three times. One spghetti crd per smple ws vilble for imging for ech extrusion process. A comprison of spghetti from duplicte processings for the 2003 crop yer smples demonstrted consistency in mesured fetures for ll smples. Only single processing ws evluted cross ll yers. All sttistics were clculted using softwre (v.9.1, SAS Institute, Cry, NC). Smple identifiction numbers chnged ech yer, lthough severl smples would be expected to be represented ech yer. Smples were therefore nested within yer for nlysis. RESULTS AND DISCUSSION Spghetti Speck Counts: Visul Given tht stndrd method ws not vilble within the lbortory for the visul determintion of specks in spghetti, n greement between three technicins ws reched s to the visul interprettion of speck. Specks were defined s esily visible under the viewing conditions. While the overll trend for speck counts between the technicins ppered consistent (Fig. 1 shows 2002), generlly only low correltions were found (Tble I). The highest nd lowest overll correltions for visul speck counts ws for the 2004 (high) nd the 2003 nd 2005 (low) smples, respectively. For 2002, both high nd low correltion were found (Fig. 2). Even with high correltion between technicins, it is cler tht the ctul number of specks counted differs; technicin 1 counted higher thn technicin 3. The sctter of results between technicin 2 nd 3 is typicl of the results found. In 2005, one smple contined very high numbers of specks: 192, 325, nd 224, respectively, for ll ssessment methods. This smple ws removed from the sttisticl nlyses s it creted dumbbell effect in the dt, resulting in bised high correltions. Nevertheless, the counts for both technicins were in greement with ech other s to the nomly of this smple. To test the possibility tht the technicins were lerning nd becoming more consistent over time given overll high correltions for 2004 (Tble I), the 2003 smples were ressessed in 2004 by the sme technicins. The ressessment results were similr to those originlly found with no significnt chnge (dt not shown), indicting tht the higher correltions found for the 2004 smple yer were smple-bsed nd could not be ccounted for by chnge in visul ssessment. Fig. 1. Visul speck counts for 26 spghetti smples (2002) for three technicins. Dt sorted in scending order (technicin 1 gry brs). TABLE I Correltion Between Visul Counts for Three Technicins Independently Assessing Spghetti Smples for Four Crop Yers Technicin 1 Technicin Technicin Technicin process 1 Technicin Technicin process 2 Technicin Technicin Technicin Technicin Fig. 2. Comprison between technicins on smple-by-smple bsis for visul speck counts in 2002 spghetti smples. (3 vs. 2 r = 0.44; 3 vs 1 r = 0.84) Technicin , n = 26; 2003(1) n = 28; 2003(2) n = 28; 2004 n = 16; 2005 n = 30. Vol. 86, No. 2,

3 To test the consistency of n individul technicin, technicin 1 counted the 30 smples for 2005 three times in rndomized pttern on different dys. The men speck count vlues for ech of the three replictes (53.1, 52.6, 52.2) hve no significnt difference with n F vlue of 0.01 nd Pr > F of The speck counts for the smples remined significntly different (F = , Pr > F ), the speck counts for the three replictes within ech smple remin consistent (F = 0.98, Pr > F 0.33). This demonstrted tht consistency in speck count evlution of spghetti cn be chieved for n individul lthough, s discussed, consistency between individuls is greter chllenge. For the visul evlutions, ech technicin viewed the sme spghetti surfce s the strnds were fixed by double-sided tpe to the smple crds, contrsting with the erlier work of Symons et l (1996) for semolin speck counting. Highly correlted speck count results between the technicins were nticipted. Clerly this hypothesis ws not supported. From discussions fter-thefct, the technicins greed tht identifying lrge nd drk specks in spghetti ws esy; however, they hd difficulty in determining the inclusion threshold for lighter, smll specks embedded in the spghetti. Commercilly, the lighter smll specks re not typiclly counted in spghetti. This result is similr to our erlier semolin speck-counting experiences nd reinforces the need for n objective speck-counting system. Spghetti Speck Counts: Imging Spghetti speck counts using the imging system were highly repetble for ech smple, (ANOVA F vlue 0.02 nd Pr > F = 0.82 for 97 smples over 4 yers) when ech smple ws imged three times. This consistency ws not influenced by yer. The imging system lso detected white spots in the smple. These were not included in the drk speck nlysis. While the overll number of specks vried from yer to yer depending on both environmentl effects nd cultivrs grown, there ws generl greement between the visul speck determintions nd the imging (IA) count method (Fig. 3). For ll yers pooled, technicins 1 nd 2 hd excellent greement with ech other (Person r = 0.86, P > 0.001) nd with the IA counts (Person r = 0.93 nd r = 0.79, P > 0.001, respectively) nd rnked the smples in highly consistent mnner with ech other (Spermn r = 0.83) nd the IA count method (Spermn r = 0.94 nd r = 0.82, respectively). Technicin 3 (yers ) hd generl disgreement with both technicins 1 nd 2 nd the IA counts (Person r = 0.07 to 0.27, P > 0.001) due to different speck count numbers which ffected smple rnking when compred with technicins 1 nd 2 nd the IA counts (Spermn r = 0.26, r = 0.14, nd r = 0.36), gin emphsizing tht visul determintion is influenced by mny fctors, including personl interprettion. For the individul 2002 nd 2004 crop yers, the reltionship between technicins counts to ech other nd to the imging method tended to be highly significnt (Tble II). Only technicin 1 lso hd greement with the imging results for 2003 nd 2005, showing tht individuls cn be consistent in their own pproch to such subjective process s speck counting cross significnt period of time. In contrst, technicins 2 nd 3 for 2003 nd the new technicin in 2005 hd highly vrible counts. This shows tht n objective imging bsed system tht demonstrtes highly repetble mesurements would hve n dvntge in commercil environment contrsting with the humn evlution of specks which is highly influenced by the person performing the nlysis. A mchine system could be djusted to production qulity stndrd nd be expected to consistently pply the stndrd to spghetti speck count evlution. Fig. 3. Comprison of visul nd IA-determined spghetti drk speck count for 2002, 2003, 2004, nd 2005 yer smples. Brs represent technicins 1, 2 nd 3, nd then IA counts, except for 2005, where brs represent technicins 1, 4, nd IA counts. TABLE II Correltion of Speck Counts Between Technicins nd Between Technicin nd Imging (IA) Count Method Within Ech of Four Crop Yers b Comprison P S P S P S P S Tech 1 / Tech Tech 1 / Tech Tech 2 / Tech Tech 1 / IA Tech 2 / IA Tech 3 / IA P, Person correltion coefficient; S, Spermn rnk correltion coefficient. b In 2005, technicin 2 ws replced by technicin CEREAL CHEMISTRY

4 Semolin Compred with Spghetti Speck Count Commercil pst qulity is controlled by setting stndrds for the semolin before mnufcture. The underpinning ssumption is tht visully ssessed semolin speck counts relte to spghetti speck counts nd, hence, qulity of rw mterils ffects spghetti qulity. Semolin speck counts were determined visully by the GRL method over n re of 25 cm 2 nd spghetti speck counts were determined both by imging nd visully over n re of 19.8 cm 2. For ll of the yers combined, visul spghetti speck counts for technicins 1, 2, nd 3 were not correlted to visul semolin speck counts (r = 0.19, r = 0.12, nd r = 0.21, respectively). Imging spghetti speck count lso hd low correltion with visul semolin speck count (r = 0.22). This low reltionship between speck counts in semolin nd spghetti remined low when evluted on yerly bsis (Tble III). Semolin is n opque product where specks re only counted on the surfce, while spghetti is trnslucent nd specks re visible on the surfce nd within the product. In generl, bout twice s mny specks were detected in spghetti per unit surfce re compred with semolin (dt not shown). While such difference could influence totl speck visibility, it would not be expected to ffect reltive numbers of specks counted nd cnnot ccount for the low correltion between semolin nd spghetti. It is possible, tht the processing conditions for spghetti chnge the physicl properties of the brn frgments, mking them drker nd, hence, more visible. Spghetti Speck Size Imging hs the bility to mesure the size of ll the drker nd lighter regions or specks within the spghetti strnd. Only drker specks were considered here. Speck size in spghetti ws skewed towrd the smller sizes (Fig. 4). There were no sttisticl differences in drk speck size profiles cross ll the smples (ll yer nlysis or within yers) (Tble IV), suggesting tht ny differences in visul counting were not influenced by significntly lr- TABLE III Correltion (Person) of Spghetti Drk Speck Counts (3 technicins nd imging method [IA]) to Semolin Drk Speck Counts (GRL visul method) for Smples in Ech of Four Yers Source of Spghetti Drk Speck Counts Yer (n) Technicin 1 Technicin 2 Technicin 3 IA 2002 (26) (25) (16) (29) Fig. 4. Speck size (μm 2 ) distribution s determined by imging for ll spghetti smples from TABLE IV ANOVA Results for All Yers Combined nd By Yer for Speck Sizes in Spghetti Smples s Determined by the Imging Method Yer Source F Vlue Pr > F All Smple 2.34 < Rep Smple 2.96 < Rep Smple Rep Smple Rep Smple Rep Fig. 5. Red (R, ), green (G, ), nd blue (B, ) color vlues for whole spghetti from TABLE V ANOVA Results for All Yers Combined nd By Yer for Speck Color Prmeters in Spghetti Smples s Determined by Imging R G B H L S Yer Source F Pr > F F Pr > F F Pr > F F Pr > F F Pr > F F Pr > F All Smple < < < < < <0.001 Rep Smple 14.7 < < < < < <0.001 Rep Smple < < < < < <0.001 Rep Smple < < < < < <0.001 Rep Smple < < < < < <0.001 Rep R, red chnnel; B, blue chnnel; G, green chnnel; H, hue (clculted); L, lightness (R+G+B)/3; S, sturtion (clculted). Vol. 86, No. 2,

5 TABLE VI Correltion of Individul Speck Are to Color Mesurement Attributes for Spghetti Smples in 2005 Speck R Speck G Speck B Speck H Speck L Speck S Speck re R, red; G, green; B, blue; H, hue; L, lightness; S, sturtion. TABLE VII Comprison of Imging-Derived Color Vlues for Spghetti Smples for Four Crop Yers Yer R G B H L S c d b b 215.6c 150.4b 29.1c 197.4b 216.5b b b b b b Mens with different letters re significntly different t = R, red; G, green; B, blue; H, hue; L, lightness; S, sturtion. ger specks in one yer s compred with nother. As the size of the brn frgments ws primrily the result of the milling process, this confirms tht the durum whet ws milled consistently into semolin over the four yers. Spghetti Speck Color Another fctor tht could influence visul speck counting is the color or drkness of the specks. Agin, the imging llowed both color nd density to be determined in spghetti for individul specks. Color mesurements were constnt, with little vrition cross replictes within ech yer or ll four yers combined for ll the color prmeters mesured (low F vlue) (Tble V). This consistency cross replictes shows tht the color-clibrted imging system ws consistent nd repetble. This is highly significnt, prticulrly when hue vlues of 2 3 grey vlues cn indicte significnt shift ( 4 o ) in color. Speck color vried from yer to yer (high F vlues for smple for ll yers combined) (Tble V). Vrition in speck color between smples within yers ws mesured for ll color prmeters. The 2005 smples hd the gretest vrition in mny of the color components nd the lest vrition in brightness (L). The high vribility in hue vlues compred with the other yers shows wide vrition in speck color nd this my relte to incresed speck visibility nd ccount for the higher speck counts in the 2005 smples (Fig. 3). Speck Color nd Size The milling process produces drk specks (brn frgments) of similr size cross multiple yers nd smples. Speck counts vry between smples s does speck color depending on how the brn frgments during milling. The mesurement of individul speck size nd color ttributes llows for the investigtion of the reltionship of speck size to color. No reltionship of speck size to speck color ws found. Tble VI (2005 crop yer) shows typicl set of correltion dt for the speck re nd color. Spghetti Color Vrition in the spghetti color could directly influence speck counting both visully nd by imging. A drker spghetti mtrix could potentilly msk lighter specks tht would otherwise be visully detectible on lighter bckground. For totl spghetti color, the lrgest vrition ws noted in the blue color plne (Fig. 5 shows typicl exmple). Agin, the humn eye is prticulrly sensitive to shifts in blue, so this would result in perceptions of lighter nd drker spghetti smples, independently of the red nd green color plnes which show little vrition. The color vlues show only generl trends, with the 2005 smples hving the highest sturtion (S) nd the drkest blue (B) nd mid-brightness (L) vlues (Tble VII). In contrst, 2003 hs the lowest sturtion (S) nd is the drkest smple yer (lowest L). Differences in hue (H) between the yers indictes tht nnul color shifts occurred; however, it is not cler whether this hd n influence on speck count. The correltion between speck nd whole spghetti color for ll smples ws highest for the green color plne (r = 0.63) nd hue (r = 0.51) with insignificnt reltionships for lightness (r = 0.29) nd sturtion (r = 0.27). This indicted lck of linkge between speck color nd spghetti color. When speck color prmeters were compred with spghetti color prmeters on yer-by-yer bsis, correltions in 2002 were ll r < 0.35 nd strongest in 2005 where r 0.6 for green, blue, hue, nd lightness. A similr but less strong reltionship ws seen in 2003, nd correltions were typiclly wek in 2004 (r 0.25). There is no cler reltionship between speck color nd spghetti color. CONCLUSIONS Spghetti nd spghetti speck color vritions re observed between smples (cultivrs) ech yer nd from yer. Spghetti speck size profiles re consistent nd relte to the consistency of the milling process. Vrition ws found in speck counts when performed by trined technicins lthough consistency ws observed when the smples were reevluted by one technicin. There ws little reltionship of visul counts to instrumentl counts. The utomted imging procedure ws highly consistent for spghetti speck counting nd ws simultneously ble to mesure both speck nd spghetti color s well s speck size. Color nd size could not be evluted by the humn observer. No reltionship ws found between semolin speck counts nd spghetti speck counts, lthough semolin speck counts re used to control the rw mteril qulity for pst mnufcture. There re cler benefits to using n imging pproch for spghetti speck counting regrding reproducibility of the method in ddition to color nd size mesurement. ACKNOWLEDGMENTS We cknowledge Brin Mrchylo nd Lind Schlichting of the Durum Qulity Lbortory, Grin Reserch Lbortory, for the spghetti smples prepred s prt of the nnul qulity ssessment of plnt breeder lines. GF received support from his reserch institute to initite this project in the GRL during the summer of LITERATURE CITED Clrke, J. M., Mrchylo, B. A., Kovcs, M. I. P., Noll, J. S., McCig, T. N., nd Howes, N. K Breeding durum whet for pst qulity in Cnd. Euphytic 100: Dexter, J. E., nd Mtsuo, R. R The spghetti-mking qulity of developing durum whets. Cn. J. Plnt Sci. 57:7-16. Dexter, J. E., nd Mtsuo, R. R Effect of smudge nd blck-point, mildewed kernels, nd ergot on durum whet qulity. Cerel Chem. 59: CEREAL CHEMISTRY

6 Dexter, J. E., Mtsuo, R. R., nd Kruger, J. E The spghettimking qulity of commercil durum whet smples with vrible α- mylse ctivity. Cerel Chem. 67: Dexter, J. E., Mrchylo, B. M., Cler, R. M., nd Clrke, J. M Effect of fusrium hed blight on semolin milling nd pst-mking qulity of durum whet. Cerel Chem. 74: El Bouziri, M., nd Posner, E. S Frin milling: Development of n experimentl flow sheet nd speck counting method. Assoc. Opertive Millers Bull. 2:5393. Htcher, D. W., Symons, S. J., nd Mnivnnn, U Developments in the use of imge nlysis for the ssessment of orientl noodle ppernce nd colour. J. Food Eng. 61: Ingelbrecht, J. A., Brijs, K., Schlichting, L., Mrchylo, B. A., Tweed, T., Ebbinghus, P. B., nd Delcour, J. A Effects of endoxylnse ddition on pst processing with short mixing time. Cerel Chem. 79: Mtsuo, R. R., Brdley, J. W., nd Irvine, G. N Effect of protein content on the cooking qulity of spghetti. Cerel Chem. 49: Shhin, M. A., nd Symons, S. J Color clibrtion of scnners for scnner-independent grin grding. Cerel Chem. 80: Symons, S. J., Dexter, J. E., Mtsuo, R. R., nd Mrchylo, B. A Semolin speck counting using n utomted imging system. Cerel Chem. 73: Vsiljevic, S., nd Bnsik, O. J Qulity testing methods for durum whet nd its products. North Dkot Stte University: Frgo ND. [Received September 4, Accepted December 29, 2008.] Vol. 86, No. 2,