Video Evaluation of Table Olive Damage duringHarvest with a Canopy Shaker

Sergio Castro-Garcia1, Uriel A. Rosa2,7, Christopher J. Gliever2,

David Smith3, Jacqueline K. Burns4, William H. Krueger5,

Louise Ferguson6, and Kitren Glozer6

ADDITIONAL INDEX WORDS. fruit impact, Olea europaea, image analysis, bruising,stereo vision

SUMMARY. Table olives (Olea europaea) traditionally are hand harvested when greenin color and before physiological maturity is attained. Hand harvesting accounts forthe grower’s main production costs. Several mechanical harvesting methods havebeen previously tested. However, tree configuration and fruit injury are majorconstraints to the adoption of mechanical harvesting. In prior work with a canopyshaker, promising results were attained after critical machine components werereconfigured. In this study, stereo video analysis based on two high-speed camerasoperating during the harvesting process were used to identify the sources of fruitdamage due to canopy-harvester interaction. Damage was subjectively evaluatedafter harvest. Fruit mechanically harvested had 35% more bruising and three timesas many fruit with broken skin as that of hand-harvested fruit. The main source offruit damaged in the canopy was the strike-impact of fruit by harvester rods.Implementation of softer padding materials were effective in mitigating fruit injurycaused by the impact of rods and hard surfaces. Canopy acceleration was correlatedwith fruit damage, thus restricting improvements needed for fruit removalefficiency through increased tine frequency.

Olives have been grown inCalifornia since the late1700s. California is the sole

commercial source of olives domesti-cally, currently depending on about1200 growers in the Central Valley(Sacramento and San Joaquin valleys)with biennial production of 115,500tons (olive is strongly alternate-bearing).The average U.S. table olive produc-tion from 2001 to 2006 accountedfor only 5.1% of total world produc-tion; the United States is the largestimporter of olives at 125,000 tons(International Olive Council, 2007).‘Manzanillo’ and ‘Sevillano’ are themost important domestic cultivars,

contributing 76% and 20%, respec-tively, of U.S. production (U.S.Department of Agriculture, 2006).Hand harvesting is the main produc-tion cost, accounting for 65% of thegross return per ton in 2005 (Hester,2006). Unlike other producers out-side the United States, California’stable olives are primarily processedas ‘‘black-ripe canned,’’ with only 5%processed by other methods. Despitethe processed nature of the product,quality of the fresh fruit is the mostimportant factor in developing mechan-ical harvesting in olives destined fortable consumption (Ferguson, 2006).

Mechanical harvesting methodsfor olives destined for oil have beendeveloped over more than 40 years,focusing on trunk shakers withdetached fruit cast over the ground,canvas, or a catch frame. These har-vesters maximize harvesting efficiency

(Fridley et al., 1971; Pellenc, 1993);fruit quality is secondary. However,trunk shaker-type harvesters areimpractical for table olives due todifferent tree structures and condi-tions, as well as the harvest maturity ofthe fruit (black-ripe for oil olives andgreen-immature for table olives).Trees are well-irrigated at harvest fortable olives; thus, ‘‘barking’’ of thetrunks can be problematic (Castro-Garcıa et al., 2007). Trees producingtable olives are often tall, weeping,and old, with fluted, multiple trunks,making trunk attachment difficult orimpossible and requiring greaterenergy input for shaking tall trees(Horvath and Sitkei, 2001). Further-more, the detachment force requiredto remove unripe, small olives, aver-aging 3 to 6 g each, from pendulouswillowy shoots is generally excessive(Kouraba et al., 2004). Fresh greenolives are extremely susceptible tomechanical damage. Industrial pro-cessing for black table olives can mit-igate some damage, but severebruising, cuts, and abrasions are unac-ceptable to the consumer.

Some fruit crops, such as citrus(Citrus spp.) or blackberry (Rubussubgenus Rubus) can use canopy har-vesters for harvesting processed orfresh market fruit (Peterson, 1998;Takeda and Peterson, 1999). Similarcanopy harvesters may prove amena-ble to table olive harvest. Recent trialswith a modified canopy harvesterengineered by AgRight (Madera,CA)/Korvan (Lynden, WA) andmodified by Dave Smith Engineering(DSE, Exeter, CA) (Fig. 1) removedfruit with 90% efficiency where fruitwas accessible (Ferguson et al.,2006), although fruit damage was stillat unacceptable levels. To reduceolive damage, the canopy harvesterwas modified by the incorporation ofpadding material to rods and othersurfaces likely to contact fruit.

The Korvan/DSE harvester isdesigned to remove fruit by vibrating

UnitsTo convert U.S. to SI,multiply by U.S. unit SI unit

To convert SI to U.S.,multiply by

0.3048 ft m 3.28084.4482 lbf N 0.22480.4470 mph m�s–1 2.2369

28.3495 oz g 0.03530.9072 ton(s) Mg 1.10232.2417 ton/acre Mg�ha–1 0.4461

The authors wish to acknowledge the support of theCalifornia Olive Committee and the cooperation ofBell-Carter Olive Company and Musco Family OliveCompany.

1Department of Rural Engineering, University ofCordoba, Cordoba, Spain

2Biological and Agricultural Engineering Depart-ment, University of California, Davis, CA 95616

3Dave Smith Engineering, Exeter, CA 93221

4Horticultural Sciences Department, University ofFlorida, Institute of Food and Agricultural Sciences,Citrus Research and Education Center, Lake Alfred,FL 33850

5University of California Cooperative Extension,Glenn County, Orland, CA 95963

6Department of Plant Sciences, University of Califor-nia, Davis, CA 95616

7Corresponding author. E-mail: uarosa@ucdavis.edu.

260 • April–June 2009 19(2)

the canopy with rods attached radiallyto the axis of three drums (Fig. 2).Drums are oriented parallel to thetree axis or at �45� to the tree axisat the top of the tree. Rods penetratethe canopy on one side of the tree andshake with a predominantly oscilla-tory movement in the plane of therods. While this movement isintended to remove fruit with littledirect interaction between the rodsand fruit, it is inevitable that rods,branches, and olives contact eachother, causing mechanical damage tothe fruit. Padding material encasingthe rods is expected to reduce thatdamage; however, no documentedanalysis of the padding quality, orhow it might be modified, exists.

Harvester–canopy interaction isa fast and complex process in which alarge number of elements are impli-cated. Thus, high-speed image analy-sis allows us to study this interactionbetween short periods of time. Eachelement position can be calculated bya stereo vision method using two

images from different viewpointsbased on the triangular measurementprinciple. This method has beenapplied in agriculture for estimationof plant geometric attributes (Ander-sen et al., 2005), location of fruit ontrees (Jimenez et al., 2000; Takahashiet al., 2002), and implementation ofharvesting robots (Tanigakia et al.,2008; Van Henten et al., 2003).

The main objective of this studywas the identification and evaluationof olive damage sources produced inthe canopy–harvester interaction toevaluate and recommend alterationsto the harvester, while identifying thenature and magnitude of olive fruitdamage as a result.

Materials and methodsHARVEST TRIAL. Olive harvesting

tests were carried out in a single dayduring Oct. 2006 on ‘Manzanillo’trees at University of California River-side’s Lindcove Research and Exten-sion Center in Exeter. Trees wereplanted in 1989 at spacing of 17 ft

in-row and 21 ft between rows andwith an average height of 14 ft. Treeswere homogeneous in size and ofmoderate vigor, producing �5tons/acre when fully cropped. Tenindividual trees were machine-har-vested for this trial. We used a proto-type developed for table olive harvestby AgRight/Korvan and modified byDSE for this work. Because themachine is designed to harvest a sin-gle side of the tree at a time and thetime required to harvest each side ofthe tree was longer than the 2.048 srestriction for recording high-speedvideo, the tree side was split in twosymmetrical quadrants. Each tree wasdivided into four quadrants deter-mined by the intersection of treecenter lines passing on directions par-allel and perpendicular to the tree rowdirection. In all, 21 quadrants wereused at random within the 10 trees forharvest evaluation. Six entire treessimilarly divided into quadrants wereused for hand-harvested comparison,and hand-harvested fruit from these24-tree quadrants were used as con-trols, providing a baseline for mini-mum attainable fruit damage.

Harvesting was carried out with aground speed of 0.25 mph. The twolower drums were oriented parallel tothe tree axis and the higher drum wasinclined 45�. Two different paddingmaterials were tested for use on therods: ‘‘soft’’ (with a hardness meas-urement of 66 Shore A) and ‘‘hard’’(75 Shore A). The hardness of amaterial is measured as its surfaceresistance to penetration of an inden-ter. The relative hardness of elasticmaterials such as rubber can be deter-mined on a Shore A scale. Typically, arubber band (soft) and a shoe heel(hard), give Shore A hardness meas-urements of 30 and 70, respectively.Both padding materials were installedon different rods in each drum in thetested harvester. Drum frequency wasfixed into a small range (180–220rpm) during field tests according tohydraulic machine regulation. Thecolors of the soft and hard rod pad-ding materials were black and red,respectively. In video analysis, thematerials different colors allowed usto differentiate between soft and hardmaterials in the captured frames. Twohigh-speed cameras (FASTCAM-X1024 PCI camera head; Photron,San Diego, CA) were used to recordstereo videos of the canopy shaking

Fig. 1. Experimental olive canopy shaker used during tests on ‘Manzanillo’ tableolives at the University of California Lindcove Experimental Station at Exeter in2006. The harvester originated as an AgRight (Madera, CA)/Korvan (Lynden,WA)-engineered design prototype and has been modified by Dave SmithEngineering (DSE, Exeter, CA).

• April–June 2009 19(2) 261

process to analyze sources of fruitdamage. The cameras were attachedto a platform mounted on a forkliftwith 8-ft separation between cameras.This separation distance was selectedaccording to manufacturer’s recom-mendations to produce the properview angle from the cameras to thetarget. Each camera was connected toits own high-speed, solid-state imagememory controller. After the cameraplatform was lifted into a recordingposition, both cameras were aimedand focused on a customized indexedtarget (Fig. 3). Subsequently, therecorded grid image was processedin the laboratory to calibrate measure-ments. ImageExpress MotionPlustracking software (Sensor Applica-tions, Utica, NY) was used to obtaina three-dimensional solution fromtwo two-dimensional images aimedat the same targeted area on thecanopy.

GREEN OLIVE DAMAGE EVAL-UATION: COMPARISON BETWEEN

MACHINE- AND HAND-HARVESTED

FRUIT. Two forms of olive damagewere considered in this study: bruis-ing and skin injury (cut or abrasion).Bruising occurs when excessive defor-mation causes the olive surface todiscolor due to oxidation of phe-nolics. Bruising results from dropping

or rough contact by hand or machine,and fruit may be softened withoutskin rupture. Skin breakage exposesflesh to the environment and fruitquality is irreversibly degraded. Skininjury during hand harvesting is lesslikely to occur because a relativelysharp edge is normally required toimpact the olive for this type ofdamage.

Machine-harvested olives fellinto a soft cloth tarp installed on theharvester catch frame to reducepotential damage by other sources.Fruit were hand collected from thetarp and stored in plastic containersfor transportation to the processingfacility, where they were visually eval-uated before processing. Hand-harvested trees were sampled byremoving fruit by stripping theshoots, the typical commercial harvestmethod. Fruit from the two harvestmethods were compared and damagewas evaluated as percentage of fruitharvested with bruises and/or skininjury. The Wilcoxon signed-rank testwas used to compare means with JMPstatistical software (version 7.0; SASInstitute, Cary, NC).

VIDEO FRAME ANALYSIS: A TOTAL

OF 15 CAPTURED SEQUENCES WITH

NUMEROUS OLIVES WERE PROCESSED

FOR THIS ANALYSIS. Each sequence

was 2.048 s in duration and allowedolive counting in the field of viewcommon to both cameras. Olivesobserved moving in the sequenceswere either ‘‘free’’ olives falling verti-cally, therefore probably droppingwithout deflection by the harvesteror branches (and probably not signif-icantly damaged) or olives in relativelyhorizontal movement, thus probablystruck by a rod or branch and poten-tially significantly injured. Other fruitdamage studies consider more param-eters to evaluate fruit damage byimpacts, but under highly controlledlaboratory conditions with respect tohow impacts are generated, fruitdevelopmental state, and uniformstage of ripeness (Menesatti andPaglia, 2001). These studies producepredictive models of high accuracy;however, they are not in situ and mayintroduce experimental error andinaccurate conclusions as a result. Inevaluating sources of damage in thisstudy, we assumed that all olives had asimilar weight and maturity status.These assumptions in other fruitstudies have been made for bruisedamage evaluation (Van Zeebroecket al., 2007). Olive velocity and accel-eration values were analyzed as one-way analysis of variance and meansranked with Kruskal-Wallis test at P £0.05.

STEREO VIDEO ANALYSIS: OLIVE

TRACKING PROCESS. Recorded videoswere displayed at low-speed to iden-tify olive damage sources. When dam-age events were identified by bothcameras, the three-dimensional olivetrajectory was determined by a man-ual tracking process using ImageEx-press MotionPlus software. Olivedamage was usually caused by animpact; a transient event that involveda short time period. Instantaneousolive velocity and acceleration beforeand after an impact were consideredto evaluate impact magnitude. Veloc-ity and acceleration measurementswere computed from three-dimen-sional position values for a constantframe period of 2 milliseconds. Themanual tracking process used onimages produced reasonable positionaccuracy; however, the tracking proc-ess generated noise in the databecause olives represented a relativelylarge and homogeneous target,occlusions or shadows occurred insome cases, and low resolution withimage magnification.

Fig. 2. Vertical drum and horizontal rod radial distribution set-up for the lowestdrum of the olive canopy shaker used during experimental tests on ‘Manzanillo’table olives at the University of California Lindcove Experimental Station at Exeterin 2006. The harvester originated as an AgRight (Madera, CA)/Korvan (Lynden,WA)-engineered design prototype and has been modified by Dave SmithEngineering (DSE, Exeter, CA).

262 • April–June 2009 19(2)

RESEARCH REPORTS

The use of short frame times tocalculate velocity values also intro-duced small errors in calculatingaccelerations. Thus, it was necessaryto verify and calibrate accelerationmeasurements. Decelerated uniformmovement of various objects was ana-lyzed to control the methodology

applied as described above and thequality of obtained data. Of theobjects and methods tested for ‘‘cal-ibration’’ purposes, we ultimatelydecided to use the parabolic trajectoryproduced when a free falling olive hita harvester rod. A first point wasmeasured after impact, and a second

one was taken as the highest verticalheight point to define the parabola.Eight olive fruit were tracked onseven different data sets to measurethe vertical velocity deceleration(gravity acceleration). Figure 4 showsthe results of gravity acceleration esti-mation. Noise reduction requiredthree or more points to obtain accel-eration values close to gravity accel-erations and error was reduced byusing more points in calculating theaverage value. Hence, frame timeresolution of 2 milliseconds wasadequate to identify impacts thatoccurred between one or two framesand to get a precise fruit trajectory;therefore, impact velocity and accel-eration were measured as instantane-ous values obtained from the averageof three points after impact.

STEREO VIDEO ANALYSIS: ROD

HARVESTER–CANOPY INTERACTION. Acanopy shaker is designed to removefruit without direct contact. In thisway, a drum transfers energy to thecanopy using rods with a periodicmovement. Rod movement is theresult of canopy-rod interaction, har-vester ground speed, drum movementeccentricity, and drum frequency.Harvester ground speed was con-stant, while drum frequency wasmodified within a small range (180–220 rpm) during field tests accordingto hydraulic machine regulation.Drum eccentricity was a harvester-design parameter constant duringfield tests.

Rod movement through the can-opy was studied by position, velocity,and acceleration measurements at therod tip. In all, nine tracked rod posi-tions in seven different quadrantsselected among those used for thetrial were tracked to study the har-vester setting. For each rod tip cycleinto the canopy, four points placed atdistances of p/2 radians were consid-ered to obtain average values. Canopyvibration was characterized by veloc-ity and acceleration values measuredbefore rods impacted olives. Thisinformation was obtained by trackingolives attached to their stems.

Results and discussionPERCENTAGE OF FRUIT DAMAGED.

Fruit sampled from the machine har-vest had 35% more bruising and threetimes as many fruit with skin injury asthat found in hand-harvested fruit(Table 1). When video frame analysis

Fig. 3. Stereo high-speed camera images focused on the indexed target and grid usedto calibrate frame position measurements of immature green ‘Manzanillo’ olive fruitand harvester rods of the olive canopy shaker used during experimental tests on‘Manzanillo’ table olives at the University of California Lindcove ExperimentalStation at Exeter in 2006. The picture on the left side (A) was taken with the leftvideo camera, and the picture on the right (B) was taken with the right video cameraused to capture images to assess potential for fruit damage during harvest. Theharvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototype and has been modified by Dave Smith Engineering(DSE, Exeter, CA).

Fig. 4. Estimation of the acceleration of gravity (9.81 m�s22) taken with differentvelocity data sets from immature green ‘Manzanillo’ olive fruit showing parabolictrajectories produced by free fall olives hit by harvester (experimental olive canopyshaker) rods. Error bars show mean value and SD (1 m = 3.2808 ft). The harvesteroriginated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered designprototype and has been modified by Dave Smith Engineering (DSE, Exeter, CA).

• April–June 2009 19(2) 263

was used to predict potentially dam-aged fruit, separated by right and leftcamera images, no difference in per-centage of potential damage wasfound (Table 2; Student’s t test: t =2.0518, P > 0.05). In all, 1226 fruitwere evaluated according to theirdirection of descent. Almost 19% ofremoved olives were isolated aspotentially highly damaged olives.The percentage of potentially dam-aged fruit, based on video frameanalysis, was considerably less thanthat actually found in sampled fruitthat were counted. This result wasanticipated because imagery damageevaluation only estimates olive dam-age caused by impacts with branchesor rods in the focused area, while thesampled fruit likely include damageresulting from all sources in the treeand harvester during the detachmentand dropping process.

STEREO VIDEO ANALYSIS: SOURCES

OF FRUIT DAMAGE. Of the 21 quad-rants chosen for stereo video analysisof machine harvest, 10 quadrantswith a total of 42 impacted fruit couldbe used for damage source analysis. Aquadrant was valid to analyze whenillumination condition, video reso-

lution, and the interaction of fruit,canopy, and rod reported a source offruit damage. Canopy movement pro-duced by rods caused slim fruit-bearingbranches to swing at high amplitude(up to 0.2 m displacement). Olivesbecame detached when a branch expe-rienced a whipping motion or whenstruck by an object. Two sources ofstrike-impact were identified: harvesterrods striking attached olives in bearingbranches (Source 1), and olivesattached to different fruit-bearingbranches striking each other (Source2). Impacts with ‘‘hard-padded’’ rods(Source 1a) were separated fromimpacts with ‘‘soft-padded’’ rods(Source 1b). When other damage sour-ces were branches hitting olives orolives hitting olives, these strikes wereseen as barely perceptible impacts anddid not significantly affect attachmentor trajectory of fruit struck in thesemanners. Olives detached solely bycanopy vibration dropped vertically;some of these were observed hittinghard-padded rods, thus becoming sig-nificantly damaged (Source 3). Fruitthat dropped vertically and hit branchesor other still-attached olives whendetached by canopy vibration werenot detected or exhibited insignificantdamage and thus were not detectable.Source 1 damage occurred to 26 fruit,17 of which were impacted with hard-padded rods (Source 1a) and 9 fruitwith soft-padded rods (Source 1b).Seven olives displayed ‘‘Source 2’’ dam-age; nine fruit had ‘‘Source 3’’ damage.Thus, the majority of severe damagewas inflicted by rods.

Table 3 presents olive velocityand acceleration values before andafter impacts. Results showed highvariability due to the particular con-dition of each impact: differentimpact time, angle, and rod velocity(lowest close to drum and maximumat the rod tip). About 60% of sampleshad reduced velocity after impact withrods or other olives. Impacts between

olive and canopy or rod coinciding inthe same movement direction wereless severe. Still-attached olivesshowed acceleration before impactclose to 600 m�s–2. Sources 1a and 3were the most important potentialdamage sources, without significantdifferences between them. All impactsof these kinds were with hard materialpadding. Source 1b (fruit impactswith soft-padded material) showedlower impacts than impacts withhard-padded material. Olives shakenfree and sustaining damage byimpacting hard-padded rods (Source3) also appeared to contribute to theincidence of olives thrown across thecatch frame; this damage sourceshould be reduced to improve harvestefficiency. Possibilities for harvestermodification to reduce Source 3 dam-age include: reducing the number ofrods, selecting soft rod padding mate-rials, and deflecting or interferingwith upper canopy detached oliveshitting the bottom drum rods.

Analysis of the sources of damageand their relative incidence led topreliminary considerations for har-vester modification. Results fromSources 1a- and 3-induced damageindicated advantages to using rodswith soft padding materials, withlarger material deformation, moreimpact energy absorbed, and longerimpact time. We did not evaluatewhether harvest efficiency (ability toremove fruit) was significantly differ-ent between rods padded with hard orsoft materials. Acceleration valuesproduced by olive impacts consideredfor Source 2, before and after impact,did not show significant differencesand had the lowest acceleration valueamong considered sources. Accord-ingly, Source 2 damage was given thelowest priority and we concluded thatharvester modifications to addressthis damage were unwarranted.

STEREO VIDEO ANALYSIS: ROD

HARVESTER–CANOPY INTERACTION.Rod movement is responsible fortransmitting energy to the canopy tocause fruit detachment, but it is alsothe main source of fruit damage. Byanalysis of tracked data, we concludedthat rod trajectory could be reducedto a periodic movement. Rod trajec-tory into the canopy showed thatrod amplitude movement in thehorizontal plane was a predominantmovement component and was keptsimilar along tests. The rod horizontal

Table 1. Damaged fruit by type ofdamage as a result of machine orhand harvest of immature green‘Manzanillo’ olives; machine harvestby AgRight/Korvan/DSEharvester.z

Harvestmethody

Damaged fruit (%)

BruiseSkin injury

(cut or abrasion)

Hand 32.8 ax 9.8 aMechanical 44.4 b 29.8 bzHarvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototypeand has been modified by Dave Smith Engineering(DSE, Exeter, CA).yTotal hand harvest sample equaled 15,688 g (553.4oz) and mechanical harvest sample equaled 9550 g(336.9 oz).xMeans in the same column with different letters differby Wilcoxon test at P < 0.05.

Table 2. Percentage of immature green ‘Manzanillo’ olives judged as having ahigh potential for damage as obtained from video frame analysis of the rodharvester–canopy interaction during harvest by AgRight/Korvan/DSEharvester.z Potential for damage determined by horizontal movement of fruitdeflected by harvester rod or branch.

Left camera (%) Right camera (%) Average (%)

Undamaged olives 81.7 80.4 81.1Potentially damaged olives 18.4 19.6 18.9zHarvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototype and hasbeen modified by Dave Smith Engineering (DSE, Exeter, CA).

264 • April–June 2009 19(2)

RESEARCH REPORTS

amplitude into the canopy was 0.149m with a frequency of 5.26 Hz (315.6rod cycles into the canopy perminute). Rod movement at the tipfeatured a mean tangential velocity of2.4 m�s–1 and mean acceleration valueof 563 m�s–2.

Table 4 presents similar velocityand acceleration values for rods andattached olives in the canopy. Stereolow-speed video inspection allowed usto observe that olives in the branchesfollowed the rod periodic movementwith similar frequency but longeramplitudes. Canopy shaking was sim-ilar to rod movement: velocity valuesclose to 2 m�s–1 and instantaneousacceleration close to 600 m�s–2. Thisapproximates a fruit removal force of 3N for a 5-g olive fruit. Previous studieswith ‘Manzanillo’ table olives meas-ured the mean fruit removal forcefrom 4 to 5 N in California growinglocations (Burns et al., 2008; Denneyand Martin, 1994). The high forcecalculated in each case illustrates thedifficulty in removing olive fruit withthe canopy shaker. While abscissionagent application could reduce theforce required to remove immatureolives, decades of research on tableolives under California conditionshave not resulted in an acceptablecommercial practice. Other researchhas reported better success with abscis-sion agent application and/or mech-anical harvest methods; Sessiz andOzcan (2006) increased harvest effi-ciency from less than 50% to close 96%using abscission treatments and abranch shaker.

MECHANICAL HARVESTING OLIVE

DAMAGE EVALUATION. Data on accel-eration of 33 olives (Sources 1 and 2)obtained from 10 quadrants werecompared with the corresponding per-centage of green olive damage (Fig.5). A positive linear correlation be-tween green olive damage and oliveacceleration measured in the canopyfrom fruit before impact (Pearson cor-relation = 0.674) was found, indicatinghigher olive damage with increasingcanopy acceleration. However, the lin-ear fit showed a low coefficient ofdetermination (r2 = 0.455) as a resultof previously stated assumptions anddifficult-to-control field conditions.Considering only canopy accelerationwas insufficient to obtain a high pre-diction value of green olive damage.Using these analyses, we establisheda damage-acceleration threshold to

Table 3. Mean velocity and acceleration of immature green ‘Manzanillo’ olivesbefore and after impacts as identified damage sources during harvest byAgRight/Korvan/DSE harvester.z Sources of damage were rods of the harvesterwith different padding or attached fruit impacting each other.

Source of oliveimpact damage Impact

Velocity[m�s–1 (SD)]y

Acceleration[m�s–2 (SD)]y

Source 1a: Hard-padded rods Before 1.9 (1.2) ax 648 (362) aAfter 2.5 (0.6) b 855 (327) b

Source 1b: Soft-padded rods Before 2.4 (1.0) b 691 (303) bAfter 2.6 (0.8) b 587 (317) a

Source 2: Olives ondifferent branchesstriking each other

Before 1.8 (1.0) a 587 (281) a

After 1.7 (0.6) a 441 (196) a

Source 3: Vertical drop,hitting hard-padded rods

Before 2.9 (1.0) b 599 (328) aAfter 2.1 (0.8) a 866 (263) b

zHarvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototype and hasbeen modified by Dave Smith Engineering (DSE, Exeter, CA).y1 m = 3.2808 ft.xMeans in the same column with different letters differ by Kruskal-Wallis test at P < 0.05.

Table 4. Comparison of velocity and acceleration between harvester rods andimmature green ‘Manzanillo’ olives before detachment from the canopy as ameasure of vibration transference obtained by stereo vision analysis of videorecorded during harvest by AgRight/Korvan/DSE harvester.z

Harvester rods Attached olives

Velocity m�s–1 (SD)y 2.4 (0.4) 2.0 (0.9)Acceleration m�s–2 (SD)y 563 (279) 647 (324)zHarvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototype and hasbeen modified by Dave Smith Engineering (DSE, Exeter, CA).y1 m = 3.2808 ft.

Fig. 5. Measured acceleration of immature green ‘Manzanillo’ olive fruit beforedetachment from the canopy with stereo video cameras and computer softwareversus actual green olive damage induced by the experimental olive canopy shaker.The harvester originated as an AgRight (Madera, CA)/Korvan (Lynden, WA)-engineered design prototype and has been modified by Dave Smith Engineering(DSE, Exeter, CA); 1 m = 3.2808 ft.

• April–June 2009 19(2) 265

isolate and prevent high olive damagefor the harvester’s current configura-tion. Olive acceleration levels up to800 m�s–2 generated more than 80%damaged olives. Thus, changingmachine configuration by increasingdrum frequency to obtain higher oliveacceleration is limited by green olivedamage.

Other alternatives such as druminclination, rod length, drum verticalseparations (to avoid re-entry of de-tached olives into another drum), andabscission treatments must be evaluatedto improve machine-harvested olivequality and increase harvest efficiency.Additional considerations importantto reduce fruit damage include clar-ifying the relationship between drumfrequency and rod density with can-opy acceleration and evaluation ofdifferent ground speeds. In additionto setting an acceptable damagethreshold for green olives, it will benecessary to optimize the harvesterfor high removal efficiency and lowolive damage.

Literature citedAndersen, H.J., L. Reng, and K. Kirk.2005. Geometric plant properties byrelaxed stereo vision using simulatedannealing. Comput. Electron. Agr.49:219–232.

Burns, J.K., L. Ferguson, K. Glozer, W.H.Krueger, and R.C. Rosecrance. 2008.Screening fruit loosening agents for blackripe processed table olives. HortScience43:1449–1453.

Castro-Garcıa, S., J.A. Gil-Ribes, G.L.Blanco-Roldan, and J. Aguera-Vega.2007. Mode shapes evaluation of trunkshakers used in oil olive harvesting. Trans.Amer. Soc. Agr. Biol. Eng. 50:727–732.

Denney, J.O. and G.C. Martin. 1994.Ethephon tissue penetration and harvesteffectiveness in olive as a function ofsolution pH, application time, and BA orNAA addition. J. Amer. Soc. Hort. Sci.119:1185–1192.

Ferguson, L. 2006. Trends in olive fruithandling previous to its industrial trans-formation. Grasas Aceites 57(1):9–15.

Ferguson, L., J.K. Burns, K. Glozer, C.Crisosto, V. Bremer, W.H. Krueger, andR. Rosecrance. 2006. Mechanical har-vester efficiency and damage evaluations.6 Jan. 2008. <http://groups.ucanr.org/olive_harvest/Research_Proposals_&_Reports/Reports_for_2006.htm>.

Fridley, Y., J. Hartman, P. Chen, and P.Whisler. 1971. Olive harvest mechaniza-tion in California. California Agr. Expt.Sta. Bul. 885.

Hester, A. 2006. Olive growers ‘‘wrapup’’ and begin plans for 2006. OliveGrower Council Nwsl., Visalia, CA.

Horvath, E. and G. Sitkei. 2001. Energyconsumption of selected tree shakersunder different operational conditions. J.Agr. Eng. Res. 80:191–199.

International Olive Council. 2007. Worldtable olive figures. 14 Sept. 2007.<http://www.internationaloliveoil.org/web/aa-ingles/corp/AreasActivitie/economics/AreasActivitie.html>.

Kouraba, K., J.A. Gil-Ribes, G.L. Blanco-Roldan, M.A. De-Jaime, and D. Barranco.2004. Suitability of olive varieties formechanical shake harvesting. Olivae101:38–43.

Jimenez, A.R., R. Ceres, and J.L. Pons.2000. A survey of computer vision meth-ods for locating fruit on trees. Trans.Amer. Soc. Agr. Biol. Eng. 43:1911–1920.

Menesatti, P. and G. Paglia. 2001. Devel-opment of a drop damage index of fruit

resistance to damage. J. Agr. Eng. Res.80:53–54.

Pellenc, R. 1993. Mechanical harvestingof olives with shaker frequency and ampli-tude control. Proc. Fruit Nut Veg. Pro-duction Eng. 4:327–330.

Peterson, D.L. 1998. Mechanical har-vester for process oranges. Appl. Eng.Agr. 14:455–458.

Sessiz, A. and M.T. Ozcan. 2006. Oliveremoval with pneumatic branch shakerand abscission chemical. J. Food Eng.76:148–153.

Takahashi, T., S. Zhang, and H. Fukuchi.2002. Measurement of 3-D locations offruit by binocular stereo vision for appleharvesting in an orchard. Proc. Amer. Soc.Agr. Biol. Eng. Ann. Intl. Mtg. Paper No.021102.

Takeda, F. and D.L. Peterson. 1999.Consideration for machine harvestingfresh-market eastern thorn less blackber-ries: Trellis design, cane training systemand mechanical harvesting developments.HortTechnology 9:16–21.

Tanigakia, K., T. Fujiuraa, A. Akaseb, andJ. Imagawac. 2008. Cherry-harvestingrobot. Comput. Electron. Agr. 63:65–72.

U.S. Department of Agriculture. 2006.California olive probability survey report. 17Sept. 2007. <http://www.nass.usda.gov/Statistics_by_State/California/Publications/Fruits_and_Nuts/indexfor.asp>.

Van Henten, E.J., B.A.J. Van Tuijl, J.Hemming, J.G. Kornet, J. Bontsema,and E.A. Van Os. 2003. Field test of anautonomous cucumber picking robot.Biosystems Eng. 86:305–313.

Van Zeebroeck, M., V. Van Linden, P.Darius, B. De Ketelaere, H. Ramon, andE. Tijskens. 2007. The effect of fruitfactor on the bruise susceptibility of ap-ples. Postharvest Biol. Technol. 46:10–19.

266 • April–June 2009 19(2)

RESEARCH REPORTS