O

Ds

QSa

b

c

a

ARR3A

KCPSWHS

1

fiaowr(

0d

Computers and Electronics in Agriculture 76 (2011) 175–182

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journa l homepage: www.e lsev ier .com/ locate /compag

riginal papers

evelopment of prototype automated variable rate sprayer for real-timepot-application of agrochemicals in wild blueberry fields

amar Uz Zamana,∗, Travis J. Esaua, Arnold W. Schumannb, David C. Percival c, Young Ki Changa,cott M. Reada, Aitazaz A. Farooquea

Engineering Department, Nova Scotia Agricultural College, Truro, B2N 5E3 NS, CanadaCitrus Research and Education Centre, University of Florida, Lake Alfred, FL 33850, USAEnvironmental Sciences Department, Nova Scotia Agricultural College, B2N 5E3 NS, Canada

r t i c l e i n f o

rticle history:eceived 13 July 2010eceived in revised form0 December 2010ccepted 28 January 2011

eywords:ontrollersrecision agricultureensorseeds

erbicidespot-application

a b s t r a c t

An automated prototype variable rate (VR) sprayer was developed for control of 8 individual nozzles on a6.1 m sprayer boom for in-season, site-specific application of agrochemicals on weeds. The sprayer boomwas divided into 8 sections and mounted behind an all-terrain vehicle (ATV) at 76.2 cm above the ground.The variable-rate control system consisted of 8 ultrasonic sensors (one per spray section) mounted ona separate boom in front of the ATV, DICKEY-john Land Manager II controller and flow valve, solenoidvalves and an 8-channel variable rate controller interfaced to a Pocket PC (PPC) using wireless Bluetooth®

radio with Windows Mobile® compatible software. This type of VR sprayer does not use prescriptionmaps, but relies on sensors to provide real-time weed detection information which is used to dispensecorrect agrochemical rates for the weeds. The sprayer can be used for in-season, spot application (SA) ofagrochemicals by activating specific boom sections where the weeds have been detected.

Two wild blueberry fields have been selected in central Nova Scotia to evaluate the accuracy of the VRsprayer. Water sensitive papers (targets) were stapled to weeds randomly selected in two tracks of eachfield. The papers were orientated parallel to the ground. The percent area coverage (PAC) of the sprayedtargets with both SA and uniform application was calculated by an imaging system. Non significanceof the t-test for uniform versus SA targets PAC indicated that there was no significant bias in the SAand that the SA was accurate. The PAC with both applications ranged from 10.01% to 81.22% and from5.39% to 72.67% in field 1 and field 2, respectively. Weed heights at selected points were measured andrelated with percent area coverage to examine the sprayer performance for spray application. Linear

regression analysis showed that weed heights were significantly correlated (R2 ranged from 0.60 to0.75) with percent area coverage of the targets in selected fields with both applications. It is proposedthat herbicide should be applied at early stage of weed growth (weed height ranged from 35 cm to55 cm and plant height ranged from 12 cm to 30 cm) for appropriate application with these specific VRsprayer arrangements. Based on these results, the VR sprayer was efficient and accurate enough for

hemi

spot-application of agroc

. Introduction

Weeds are the major yield-limiting factor in wild blueberryelds (Yarborough, 2006). Weed flora in blueberry fields tradition-lly consisted of slow spreading perennial species whereas many

f the new species invading blueberry fields are common annualeeds of arable fields that produce large number of seeds and

equire control with herbicides both in prune and production yearJensen and Yarborough, 2004; McCully et al., 1991). Traditionally,

∗ Corresponding author.E-mail address: qzaman@nsac.ca (Q.U. Zaman).

168-1699/$ – see front matter © 2011 Elsevier B.V. All rights reserved.oi:10.1016/j.compag.2011.01.014

cals usage in wild blueberry fields.© 2011 Elsevier B.V. All rights reserved.

herbicides are applied uniformly in wild blueberry fields, but weedsare not distributed uniformly within fields. In these situations,spatial information management systems hold great potential forallowing producers to fine-tune the locations, timings, and rates ofherbicide application.

Many researchers have attempted to develop variable rate (VR)technologies for various crops (Rockwell and Ayers, 1994; Giles andSlaughter, 1997; Steward and Tian, 1999; Tian, 2002; Carrara et al.,

2004; Miller et al., 2005; Zaman et al., 2005, Schumann et al., 2006;Dammer et al., 2008) to date little attention has been paid to wildblueberry production systems. Michaud et al. (2008) developed aVR prototype sprayer to deliver pesticides based on prescriptionmaps, developed in GIS software, using aerial spectral scans of wild

1 ctron

bbpd

wHZoasodantbwt

pewwtp

habcs(IfiSededAPuwrusb

oMwrb

2

2

oVVPm

76 Q.U. Zaman et al. / Computers and Ele

lueberry fields. The system was sensitive to positional error causedy global positioning system (GPS) and obtaining up-to-date aerialhotography was expensive, the quality was quite variable, andata processing for weed detection was also intensive and difficult.

Several machine vision systems have been developed to detecteeds in different cropping systems (Sui et al., 1989; Shearer andolmes, 1990; Zhang and Chaisattapagon, 1995; Tian et al., 1997;hang et al., 2009), because real-time weed detection at the timef spot spraying could be very valuable for cutting chemical costsnd reducing environmental contamination. However, these visionystems, based on morphological or textural weed detection meth-ds, generally needed a relatively high image resolution, and theetection algorithms were quite complicated and computation-lly expensive (Meyer et al., 1998; Zhang et al., 2009). There is aeed to develop spot-specific herbicide application technologieshat do not require high resolution image processing techniques,ut rely on sensors to sense the weed in real-time and provide theeed detection information to fast VR controllers for spray at right

argets.Ultrasonic sensors are widely accepted for quantification of

lant heights (Sui et al., 1989; Schumann and Zaman, 2005). Swaint al. (2009) developed and tested low-cost ultrasonic system foreeds (taller than plants) and bare spot mapping in real-timeithin wild blueberry fields during growing season. They reported

hat ultrasonics performed well to detect tall weeds (taller thanlants) and bare spots in wild blueberry fields.

Advances in sensing technology and VR control systemsave offered new opportunities for detecting weeds and spot-pplication of agrochemicals in a specific section of the VR sprayeroom where the weeds have been detected. Many commercialontrollers have been developed to deliver agrochemicals on site-pecific basis using GPS guided prescription maps within fieldDICKEY-john Land Manager II: DICKEY-john Corporation, Auburn,L; MidTech Legacy 6000 controller: Midwest Technologies, Spring-eld, IL; Raven 660 controller: Raven Industries Inc., Sioux Falls, SD).chumann and Hostler (2009) with the partnership of a machin-ry manufacturer (Chemical Containers, Inc., Lake Wales, FL, USA)eveloped an 8-channel computerized VR controller consists oflectronic hardware with internal firmware and matching Win-ows Mobile 6.0 software on a handheld pocket PC computer (PPC).computerized 8-channel variable rate controller is linked with

PC using wireless Bluetooth®. Typically this controller does notse prescription maps, but relies on sensors to provide real-timeeed information which is used to dispense the correct herbicide

ate for the weed eradication within field. The reliable and fastltrasonics and VR controllers could be used to develop VR rateprayer for in-season, spot-application of agrochemicals in wildlueberry cropping system.

In this study, an automated VR prototype sprayer consistingf ultrasonic sensors, 8-channel computerized VR controller, Landanager II controller, solenoid valves, PPC with operating soft-are was developed. The performance of VR sprayer for in-season

eal-time weed detection and spot spraying was evaluated in wildlueberry fields.

. Materials and methods

.1. Development of prototype variable rate (VR) sprayer

The prototype VR sprayer was developed for spot-application

f herbicides on tall weeds in wild blueberry cropping system. TheR sprayer is consisting of ultrasonics, computerized 8-channelR controller (VRC), Land Manager II controller (LMC), handheldocket PC (PPC) with operating software, servo valve and floweter, solenoid valves, nozzles and a tank capacity of 209 l (Fig. 1).

ics in Agriculture 76 (2011) 175–182

The VR sprayer was mounted on an all-terrain vehicle (ATV). The6.1 m sprayer boom was divided into eight sections (76.2 cm eachsection) and mounted behind the ATV at 76.2 cm above the ground.The boom height was adjustable so that weed sensing area andspray could be fine-tuned to crop conditions. Eight solenoid valvesand nozzles (one valve and one nozzle in each section) weremounted on the boom with a uniform (76.2 cm) interval betweenthem. The nozzles were TeeJet TP8004 nozzles (Maritime SuppliesLimited, Monton, NB, Canada) with a spray angle of 110◦. Using aseries of T joints, the line connecting the distribution valve to eachsection was then connected to each solenoid valve to which a nozzlewas fitted as closely as possible. The model 2201A solenoid valve(Delware Pump and Parts Limited, Delware, ON, Canada) was oper-ated on 12-V and consumed only 13 W. The feed line from the pumpwas going through a flow valve and flow meter then was separatedinto two lines, each line (right and left) feeding four sections of theboom. The pump was operated by a Honda gas engine (Honda Inc.,NS, Canada).

The wide angle beam, long range and fast measurementcycle Maxbotix LV-MaxSonar-EZ1 Sonar Module ultrasonic sensors(Robotic Inc., Boisbriand, QC, Canada) were incorporated verticallyinto individual boom section to detect weeds taller than blueberryplants in real-time within wild blueberry fields. The 6.1 m long sen-sor boom was mounted in front of the ATV at 0.90 m height aboveground surface. The sensors were connected to VRC (Chemical Con-tainers, Inc., Lake Wales, FL, USA). The VRC consists of electronichardware with internal firmware and matching Windows Mobile6.0 software on a PPC. VRC was interfaced to a PPC and could beoperated easily from a PPC using wireless Bluetooth® radio). VRCreceived target (weeds taller than blueberry plants) detection sig-nal from sensor and opened the valve in a specific section of boomwhere the target had been detected. The plant height raged from12 cm to 27 cm. The VRC was installed in the ATV cab and was con-nected to LMC (DICKEY-John Corporation, Auburn, IL, USA). Afterreceiving target detection information from the sensor VRC auto-matically communicated with LMC. The LMC regulated dischargeof the nozzles in specific sections of the boom where the targethad been detected based on ground speed obtained from WAAS-enabled DGPS (Garmin International Inc. Olathe, KS, USA) througha DJ servo valve and DJ flow meter (Fig. 2).

2.2. Laboratory tests

Ultrasonic sensors were calibrated to measure the distance fromthe ultrasonic sensor to the target in the metal shop at Nova Sco-tia Agricultural College (NSAC), Truro, NS, Canada. The distances(from sensor to the target; cardboard) were measured three timesat ∼12.7 cm intervals up to 152.4 cm with measuring tape. The cor-responding voltages were recorded using a digital multimeter atthe time of distance measurements for comparison. The multi-meter was connected to a sensor. The voltages were digitized by10-bit format from analog to digital to make compatible with theprogram software installed in PPC. The measured distances (fromsensor to target) and voltage were compared by linear regressionusing SAS 9.1 software (SAS Institute, Cary, NC, USA) to examine theperformance accuracy of the ultrasonic distance measurements.The calibration equation was incorporated into program softwareinstalled in PPC.

An experiment was conducted in the metal shop at NSAC to cal-culate response time (i.e., lag time between sensor detection andtarget spray) for VRC to open the valve at right target after receiv-

ing target detection information from the sensor. The time to buildup the cone after the discharge is started has been added in cal-culating response time for precise real-time spray at right target.An LED bulb was wired into switch #8 on VRC. A �Eye camera(UI-1220SE/C, IDS Imaging Development System Inc., Woburn, MA,

Q.U. Zaman et al. / Computers and Electronics in Agriculture 76 (2011) 175–182 177

Fig. 1. Schematic diagram of automated prototype variable rate sprayer for spot-application of agrochemicals.

Fig. 2. Flow chart showing the complete process of spraying on target (from weed detection to spray).

1 ctron

UrlvtswIr

tamoovpm2sL

2

tatcwthrapbbfCad

tm3tTtTtra

taw(wwpwfpGUa

78 Q.U. Zaman et al. / Computers and Ele

SA) was positioned in front of the spray nozzle (nozzle #8) toecord the video. The bulb was placed within 5.8 cm of the cameraens so that it could be seen in the centre of the video frame. Theideo was recorded with149 frames s−1 when the sensor detectedhe target and bulb was ON until controller opened the valve topray at the target. Test was repeated ten times and video imagesere analyzed with V1 HOME 2.0 software (Interactive Frontiers,

nc., Plymouth, MI, USA), allowing for a frame by frame analysis ofesponse time between sensor detection and spray the target.

The VRC received target detection signal from sensor and simul-aneously communicated with LMC. An experiment was conductedt NSAC to evaluate the performance of LMC for flow measure-ents from the nozzles mounted on the sprayer boom. The volume

f water from each nozzle and then from different combinationsf nozzles was recorded from the LMC and also at the same timeolume was measured manually with graduated cylinders for com-arison. The experiment was repeated three times and volumeeasurement readings were averaged in Excel (Microsoft Office

003). The differences between LMC volumes and manually mea-ured volumes were used to characterize the performance of theMC.

.3. Real-time field test

Two wild blueberry fields were selected in central Nova Scotiao test the performance accuracy of the VR sprayer for spot-pplication of agrochemicals in July, 2009. The selected fields werehe Londonderry site (field 1; 45.441199◦N, 63.541249◦W) and theattle market site (field 2; 45.45202◦N, 63.449626◦W). Both fieldsere in their sprout vegetative year of the biennial crop produc-

ion cycle in 2009, having been in the crop year in 2008. The fieldsad been under commercial management over the past decade andeceived biennial pruning by mowing for the past several yearslong with conventional fertilizer, weed, and disease managementractices. The main weed species, in competition with wild blue-erry were goldenrod (Solidago sp.) in Londonderry field, and blackulrush (Scirpus atrovirens Willd.), dogbane (Apocynum androsaemi-

olium L.) and bracken fern (Pteridium aquilinum L.) were more inattle market field. The average blueberry plant height was 24 cmnd weed height ranged from 50 cm to 70 cm. Meteorological con-itions were same for both UA and SA during the field experiments.

Ultrasonic sensors were calibrated to measure the distance fromhe ultrasonic sensor to the weeds in a wild blueberry field. The

aximum height (from the ground surface) of selected weeds was0 cm. The distances (from sensor to the weeds) were measuredhree times at ∼10 cm intervals up to 140 cm with measuring tape.he corresponding voltages were recorded using a digital mul-imeter at the time of distance measurements for comparison.he multimeter was connected to a sensor. The measured dis-ances (from sensor to weeds) and voltage were compared by linearegression using SAS 9.1 software to examine the performanceccuracy of the ultrasonic distance measurements.

Two tracks (100 m × 6.1 m each) were selected in each field toest the accuracy of VR sprayer for detecting weeds and sprayingt right targets. The boundaries of selected tracks were mappedith real time kinematics -global positioning system (RTK-GPS)

On GRADE Inc., Dartmouth, NS, Canada) in both fields. Twelveater sensitive papers (targets) were stapled at randomly selectedeed spots in each track in field 1. In field 2, 6 and 12 water sensitiveapers were stapled in track 1 and track 2, respectively (Fig. 3). Theeed patches are not randomly distributed in the fields. There were

ew weeds at the west part in field 1. The papers were orientatedarallel to the ground and selected targets were marked with RTK-PS for mapping in ArcView 3.2 GIS software (ESRI, Redlands, CA,SA). The water was sprayed on automated mode (spot-specific)nd water sensitive papers were collected after drying. Then papers

ics in Agriculture 76 (2011) 175–182

were stapled again and water was sprayed on uniform basis (allnozzles were opened) for comparison. The water sensitive paperswere scanned and processed to calculate percent area coverage(PAC) of the sprayed targets with both spot-application (SA) anduniform application (UA) using WinRHIZO image analysis system(Regent Instruments Inc., Canada). The student t-test was per-formed to examine whether the PAC of the sprayed targets withSA different than the PAC of the targets with UA. The heights ofselected targets (weeds) were measured manually with ruler atthe time of water application during the experiments. The linearregression method was used to test the relationship between tar-get height and PAC of the sprayed targets with both methods UAand SA. The parallel curve analysis was performed with Genstat5.0 (Lawes Agricultural Trust, Rothamsted, UK) to test whether theslope and intercept of the calibration equation for target height andPAC are significantly different between UA and SA.

The ground speed during the field operations was 6 ± 0.2 km/h.Threshold distance from the sensor to weed canopy was set at58 cm height to detect the weeds accurately in the selected fields.The buffer, before and after the target, was adjusted at 25.4 cm forprecise overlapping of agrochemical applications on targets. Thelook-ahead delay time and other constant parameters, thresholdheight from the sensor to weed canopy and buffer before and afterthe target, were stored in non-volatile flash memory on the mainmicrocontroller for accurate spray on right targets. The stored oper-ating parameters could be easily retrieved and activated by linkingthe VRC with a PPC using wireless Bluetooth® radio and editing orselecting values on the setup screen with Windows Mobile® com-patible software. The application rate was set up at 187.0 L/ha inLMC.

3. Results and discussion

The principle components of VR sprayer (sensors and con-trollers) were successfully calibrated and tested in the metal shopat NSAC for target sensing and spraying at right target in a specificsection of the sprayer boom where the target has been detected.Ultrasonics were calibrated for distance measurement and the lin-ear calibration model showed that distance measured from sensorto cardboard was correlated highly significantly with the volt-age (R2 = 0.99; P < 0.001). Ultrasonic sensor was also calibrated fordistance measurement in a wild blueberry field and the linearregression results indicated highly significant correlation betweendistance measured from sensor to weeds and voltage (R2 = 0.99;P < 0.001). Ultrasonic sensor calibration equation for distance mea-surement (Fig. 4a and b) incorporated into software installed in PPCto permit the sensor of target detection accurately.

Results indicated that VRC was operated easily from PPC usingwireless Bluetooth® radio with Windows Mobile® compatible soft-ware. VRC was fast and accurate enough to open the valve at righttarget with delay time 0.05 ± 0.003 s (Table 1) after receiving thesensor target detection information. The LMC also performed reli-ably and rapidly to regulate flow rate in each nozzle through servovalve and flow meter with ≤2.5% difference from manual flow mea-surements during calibration test (Table 2). Therefore, principlecomponents of VR sprayer could be used to develop VR sprayerto detect targets (weeds taller than plants) and spot-application ofagrochemicals at right targets reliably to reduce cost of productionand protect environment.

3.1. Real-time field test

Tall weeds remain a serious threat for the growth of wild blue-berry as well as for smooth mechanical harvesting causing fruitlosses. Results of this study indicated that automated identification

Q.U. Zaman et al. / Computers and Electronics in Agriculture 76 (2011) 175–182 179

F ons wL

ouwvictprh

ig. 3. Maps showing targets (water sensitive papers) selected for spray applicationdonderry field 1 and (b) cattle market field 2.

f taller weeds in real-time and spot-application of herbicides usingltrasonic and fast VRC would help in monitoring their growth inild blueberry fields. Non significance of the t-test for uniform

ersus SA targets PAC indicated that there was no significant biasn the SA and that the SA technique was accurate to apply chemi-

als at selected targets (Table 3). Visual observation also revealedhat VR sprayer performed reliably during the field experiments,ermitting real-time target (weed) sensing and spot-application atight targets in a specific section of sprayer boom where the weedsave been detected. Therefore, it is important to develop VR sprayer

ith prototype variable rate sprayer in each track of both wild blueberry fields (a)

using ultrasonics and VRC for in-season spot-application of agro-chemicals in wild blueberry fields in order to increase net economicreturns.

The VRC was accurate and faster enough with 0.05 s responsetime to spray water on selected targets accurately during the

experiments in selected fields. The VRC automatically compensatedchanges in response time caused by variation in ground speed dur-ing operation for accurate applications. The timing calculation forthe response time is dependent on the ground speed. The groundspeed was obtained in real-time from a WASS-enabled DGPS to

180 Q.U. Zaman et al. / Computers and Electronics in Agriculture 76 (2011) 175–182

y = 0.2594x - 1.2323R2 = 0.998

0

20

40

60

80

100

120

140

0 100 200 300 400 500 600Voltage (mV)

Dis

tanc

e (c

m)

(a) Card board

y = 0.2228x - 0.7078R2 = 0.994

0

20

40

60

80

100

120

0 100 200 300 400 500 600Voltage (mV)

Dis

tanc

e (c

m)

(b) Weeds

Fig. 4. (a and b) Relationship between voltage obtained from voltmeter connected touat

cgwsbt

Table 1Response time for 8-channel computerized variable rate controller to open valveand spray in a specific section of the boom where the target has been detected.

Trial Camera starttime (s)

Spray time (s) Difference (s) Average responsetime (s)

1 3.497 3.544 0.047 0.05± 0.003

2 7.946 7.993 0.0473 12.329 12.376 0.0474 16.765 16.819 0.0545 21.087 21.141 0.0546 25.490 25.537 0.0477 29.564 29.618 0.0548 33.564 33.618 0.0549 37.336 37.390 0.054

TC

TSw

ltrasonic sensor and actual distance measured from sensor to target (a: cardboardsnd b: weeds) with measuring tape for calibration of ultrasonic sensor to measurearget heights.

ompensate changes in look-ahead delay time to spray at right tar-ets. Based upon the results, it is suggested that VRC interfaced

ith PPC should be incorporated into commercial sprayer for spot-

pecific application of agrochemicals reliably and accurately in wildlueberry fields to improve profitability and environmental protec-ion.

able 2omparison of Land Manager controller discharge measurement with manually measure

Number of nozzles

Volume (l/min) All 8 1 1, 2Manually measured 36.9 12.0 13.3LM Controller 37.4 11.7 13.3Difference (%) +1.4 −2.5 0

able 3ummary statistics of percent area coverage of the sprayed targets for determining the prith prototype VR sprayer of selected points in each track (T) of both fields.

Londonderry

Track (n) Minimum (%) Maximum (%)

SAT1 (12) 21.10 65.20UAT1 (12) 17.59 59.03SAT2 (12) 13.08 81.22UAT2 (12) 10.01 84.60

Cattle marketSAT1 (06) 12.91 68.64UAT1 (06) 5.39 72.67SAT2 (12) 19.84 71.29UAT2 (12) 24.28 64.97

10 41.282 41.336 0.054

Results of this study indicated that LMC regulated the flow cor-rectly through DJ servo valve and flow meter in specific nozzleof the boom section where the weeds had been detected duringthe operation with SA (Fig. 5). Non significance of the t-test forUA versus SA targets PAC indicated that there was no significantbias in the SA and that the SA was accurate. It is observed thatLMC automatically compensated the changes in nozzle flow ratecaused by variation in ground speed during operation. On a conven-tional chemical broadcast application sprayer, the nozzle spacingand the boom height chosen mainly depend on the overall spraypattern uniformity requirement. For the new VRC, the sensing sys-tem spatial resolution was considered as the major factor in thenozzle spacing selection (Fig. 6). For each individual nozzle to becontrolled separately, the size of the section which one nozzle cov-ered was equal to, or slightly larger than the detection zone of thesensing system.

The PAC of the sprayed targets with both applications rangedfrom 10.01% to 81.22% and from 5.39% to 72.67% in field 1 andfield 2, respectively (Figs. 7 and 8). The reason of variation in PACmight be due to the designed boom height for the VR sprayer.

The height of the targets was observed in selected tracks rangedfrom 48.26 cm to 68.58 cm in both fields. Linear regression analy-sis results showed that weed heights were significantly correlated(R2 ranged from 0.60 to 0.83) with PAC in selected fields for both

d discharge from the nozzles in different sections of sprayer boom.

1, 2, 3 1, 2, 3, 4 1, 2, 3, 4, 5 1, 2, 3, 4, 5, 6 1, 2, 3, 4, 5, 6, 714.9 18.6 18.9 18.9 18.615.2 19.0 19.4 19.6 19.0+2.0 +2.1 +2.0 +1.5 +2.1

ecision of spot-application (SA) technique relative to the uniform application (UA)

Mean (%) SD (%) t(d.f.) F-probability

38.33 14.49 0.03 (22)38.17 13.86 0.97852.88 20.52 0.13 (22)51.70 23.46 0.897

39.08 21.64 0.24 (10)42.32 25.58 0.81747.76 17.89 0.09 (20)47.15 14.15 0.929

Q.U. Zaman et al. / Computers and Electronics in Agriculture 76 (2011) 175–182 181

F nt aret

aiic(ft(fV

ciflag

Fs

processors can be an option to differentiate weeds, bare spots andblueberry plants real-time in the field. The weed or plant detectioninformation will be sent to the controller to spray agrochemicalin the specific boom section where the weeds or plants have beendetected.

(Uniform) y = -3.21x + 231.65

R2 = 0.7580

100 UniformSA

ig. 5. Showing the targets (water sensitive papers) sprayed with water and percerack 1 of field 1 using prototype variable rate sprayer.

pplications (Figs. 7 and 8). The parallel curve analysis showed thatndividual equations (UA and SA) were not significantly differentn the intercept and slope. Therefore, only one equation for all sitesould be used to describe the relationship between height and PACFig. 9). The results indicated that the VR sprayer performed wellor <55 cm tall weeds with the existing arrangements. It is proposedhat herbicide should be applied at early stage of weed growthweed height ranged from 30 cm to 55 cm and plant height rangedrom 12 cm to 27 cm) for appropriate application with these specificR sprayer arrangements.

Growers apply herbicides during the growing season (velpar toontrol grasses in April, calisto to control goldenrod in June, kerb

n the fall to control fescue grasses and sheep sorrel), fungicides fororal blights (monilinia and botrytis) and leaf diseases (septoriand rust) on foliage only, and insecticides for fruit fly in July. Therasses and weeds are not tall enough to sense using ultrasonic

ig. 6. Schematic diagram showing threshold height from sensor to target and sen-or target zone for spray.

a coverage (PAC) of the sprayed targets for both uniform and spot-application in

sensors in April and October. The digital color cameras with customsoftware, using the advanced image processing techniques, and fast

(SA) y = -2.91x + 214.39R2 = 0.69

0

20

40

60

7065605550Weed height (cm)

Are

a co

vera

ge (%

)

Fig. 7. Relationship between percentage area coverage of sprayed targets with tar-get height for both uniform and spot-application at selected points in both tracks offield1.

182 Q.U. Zaman et al. / Computers and Electron

(SA) y = -2.1955x + 168.04

R2 = 0.59

(Uniform) y = -2.1811x + 167.98

R2 = 0.63

0

20

40

60

80

100

7065605550Weed height (cm)

Are

a co

vera

ge (%

) UniformSA

Fig. 8. Relationship between percentage area coverage of sprayed targets with tar-get height for both uniform and spot-application at selected points in both tracks offield 2.

(Uniform) y = -2.6284x + 195.61R2 = 0.66

(SA) y = -2.471x + 186.68R2 = 0.61

0

20

40

60

80

100

7065605550

Weed height (cm)

Are

a co

vera

ge (%

)

UniformSA

Fti

4

rlVhbtogia

iabne

ig. 9. Relationship between percentage area coverage of sprayed targets witharget height for both uniform and spot-application at selected points in both exper-mental fields in central Nova Scotia, Canada.

. Conclusions

The sensing and control systems of VR sprayer are efficient andeliable enough to detect tall weeds and spray correctly at an actualag time of 0.05 ± 0.003 s. Based on the results of this study, theR sprayer proved very efficient for in-season spot-application oferbicides to eradicate tall weeds <55 cm (taller than plants) in wildlueberry fields. Although the percent area coverage of the sprayedargets with spot-application and uniform application was limitedn taller weeds in both fields and chemical spraying at early weedrowth stage is needed to avoid less spray coverage on taller weeds,t should be possible to increase the VR sprayer’s performance tochieve enough coverage on weeds.

This VR sprayer could be used for a variety of precision farm-

ng applications including site-specific liquid fertilization in plantreas, fungicide/insecticide spraying on foliage only in wild blue-erry cropping systems. Further research and experimentation areeeded to determine the optimal chemical input amount for differ-nt weed coverage, control zone size, and timing combinations. The

ics in Agriculture 76 (2011) 175–182

knowledge of agricultural engineers, agricultural economists, weedscientists, and agrochemical experts must be brought together inorder to develop the high performance expert system required fora VR sprayer.

Acknowledgements

This work was supported by Oxford Frozen Foods Limited, Agri-Futures (ACAAF) Nova Scotia, Wild Blueberry Producers Associationof Nova Scotia and the Nova Scotia Department of Agriculture Tech-nology Development Program. The authors would like to thankGary Brown and Doug Wyllie (farm managers Bragg Lumber Com-pany), Robert Morgan and Kelsey Laking (research Assistants) fortheir assistance during the experiment.

References

Carrara, M., Comparetti, A., Febo, P., Orlando, S., 2004. Spatially variable herbicideapplication on durum wheat in Sicily. Biosyst. Eng. 87 (4), 387–392.

Dammer, K.-H., Wollny, J., Giebel, A., 2008. Estimation of the leaf area index in cerealcrops for variable rate fungicide. Eur. J. Agron. 28, 351–360.

Giles, D.K., Slaughter, D.C., 1997. Precision band sprayer with machine-vision guid-ance and adjustable yaw nozzles. Transact. ASAE 40 (1), 29–36.

Jensen, K.I.N., Yarborough, D.E., 2004. An overview of weed management in the wildlowbush blueberry-past and present. Small Fruits Rev. 3, 229–255.

McCully, K.V., Sampson, M.G., Watson, A.K., 1991. Weed survey of Nova Scotia low-bush blueberry (Vaccinium angustifolium) fields. Weed Sci. 39, 180–185.

Meyer, G.E., Mehta, T., Kocher, M.F., Mortensen, D.A., Samal, A., 1998. Texturalimaging and discriminant analysis for distinguishing weeds for spot spraying.Transact. ASAE 41 (4), 1189–1197.

Michaud, M.-A., Watts, K.C., Percival, D.C., Wilkie, K.I., 2008. Precision pesticidedelivery based on aerial spectral imaging. Can. J. Biosyst. Eng. 50, 2.9-2.15.

Miller, W.M., Schumann, A.W., Whitney, J.D., Buchanon, S., 2005. Variable rate appli-cation of granular fertilizer for citrus test plots. Appl. Eng. Agric. 21 (5), 795–801.

Rockwell, A.D., Ayers, P.D., 1994. Variable rate sprayer development and evaluation.Appl. Eng. Agric. 10 (3), 327–333.

Schumann, A.W., Hostler, H.K., 2009. Computerized automatic variable rate con-troller for application of agrochemicals in orchards. Submitted for U.S. patent as“Managing application of agrochemicals to plants”.

Schumann, A.W., Miller, W.M., Zaman, Q.U., Hostler, K., Buchanon, S., Cugati, S., 2006.Variable rate granular fertilization of citrus groves: spreader performance withsingle-tree prescription zones. Appl. Eng. Agric. 22 (1), 19–24.

Schumann, A.W., Zaman, Q.U., 2005. Software development for real-time ultrasonicmapping of tree canopy size. Comput. Electron. Agric. 47 (1), 25–50.

Shearer, S.A., Holmes, R.G., 1990. Plant identification using color co-occurrencematrices. Transact. ASAE 33 (6), 2037–2044.

Steward, B.L., Tian, L., 1999. Machine-vision weed density estimation for real-time,outdoor lighting conditions. Transact. ASAE 42 (6), 1897–1909.

Sui, R., Wilkerson, J.B., Wilhelm, L.R., Tompkins, F.D., 1989. A microcomputer basedmorphometer for bush-type plants. Comput. Electron. Agric. 4 (1), 43–58.

Swain, K.C., Zaman, Q.U., Schumann, A.W., Percival, D.C., Nams, D., Esau, T.J., 2009.Detecting weed and bare-spot in wild blueberry using ultrasonic sensor tech-nology. ASABE Paper No. 096879, ASABE, St. Joseph, MI, USA.

Tian, L., Slaughter, D.C., Norris, R.F., 1997. Outdoor field machine vision identifica-tion of tomato seedlings for automated weed control. Transact. ASAE 40 (6),1761–1768.

Tian, L., 2002. Development of a sensor-based precision herbicide application sys-tem. Comput. Electron. Agric. 36, 133–149.

Yarborough, D.E., 2006. Innovations in weed management in wild blueberry fields inMaine. In: Eighth International Symposium on Vaccinium Culture, Sevilla, Spainand Oeiras , Portugal, pp. 197–202.

Zaman, Q.U., Schumann, A.W., Miller, W.M., 2005. Variable rate nitrogen applicationin Florida citrus based on ultrasonically sensed tree size. Appl. Eng. Agric. 21 (3),

331–335.

Zhang, N., Chaisattapagon, C., 1995. Effective criteria for weed identification in wheatfields using machine vision. Transact. ASAE 38 (3), 965–975.

Zhang, F., Zaman, Q.U., Schumann, A.W., Percival, D.C., Nams, D., Esau, T.J., 2009.Detecting weeds in wild blueberry field based on color images. ASABE Paper No.096146, ASABE, St. Joseph, MI, USA.