Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Biosystems Engineering (2002) 83 (4), 433–440

doi:10.1006/bioe.2002.0139, available online at http://www.idealibrary.com on


PH}Postharvest Technology

Automatic Jalapen* o Chilli Grading by Width


Federico Hahn1,2
1
VIKSAL, P.O.Box 394, Zihuatanejo, Guerrero, M!exico
2
Departamento Irrigacio! n, Universidad Autonoma de Chapingo, Chapingo, 056230, Estado de M!exico, Mexico;
e-mail: bandido058@hotmail.com

(Received 9 June 2001; received in revised form 4 September 2002)

Jalapen* o chilli is a variety grown extensively in Mexico consumed by almost all the population. It has a high
processing demand and proper sorting is required before filling or canning. A sorter that classifies chilli by
three different width sizes was built. The conveyor used baby suckers to align each chilli during sensing. Chilli
width was determined by means of a photodiode scanner, which detected the incoming radiation sent by a
laser line generator. Chillies presenting necrosis were detected with a radiometer and removed to increase
product quality. The accuracy on the necrosis detection and width classification was of 963 and 87%,
respectively. # 2002 Silsoe Research Institute. Published by Elsevier Science Ltd. All rights reserved

1. Introduction Chilli is a basic nutritious product consumed by


almost all the population in Mexico. Jalapen* o chilli is
The inspection and sorting of fruits and vegetables cultivated all around the country and a total of 15 219 ha
can be very tedious work. In countries with cheap was planted in 1998 obtaining an yield of 192 590 tonnes
labour, inspectors classify fruits and vegetables manu- (INEGI, 1999). Sixty per cent of the chilli production
ally, but differences between visual standards and lack of was processed, while 20% was sold as fresh vegetable in
experience produce undesirable results (Peleg, 1985). supermarkets and grocery shops. The other 20% was
Machine vision classifying systems are efficient, consis- used to produce ‘chipotle’. The main quality features
tent and provide economic benefits. Howard and Searcy found in Jalapen* o chillies are: their size ranges from 6 to
(1989) studied carrot length, width, curvature and 8 cm, they weigh at least 10 g, they are bright green in
perimeter to detect their shape. Fourier-based algo- colour and they have a peduncle to avoid bacteria and
rithms applied to potato images to grade them accord- fungi contamination (SARH, 1982). Some processed
ing to their shape were proposed by Tao et al. (1995). chillies are refilled with cheese and prawns and are
Twenty-two different apple defects were identified using exported to different countries. The desired width of the
image analysis techniques (Throop & Aneshansley, chillies is about 40% of their length, and chillies that are
1997) and classified using multilayer back propagation only 30% wider than their length are considered thin
algorithms (Miller et al., 1998). Modern automatic sorters and not proper for refilling. Chillies with the proper
use individual bicone rollers to transport the fruit and to width–length ratio are bisected, their veins removed and
present the entire fruit surface to the sensing camera then refilled.
(Crowe & Delwiche, 1996; Throop & Aneshansley, Chilli sorting is done manually by experienced people
1999). and a machine was developed to classify chillies
Structured light has been used in fruit detection to automatically taking their sorting judgement into
help vision systems and Benady and Miles (1992) consideration. Chilli was classified into three different
detected the presence of melons during automatic width sizes and necrotic products were removed. The
harvesting. Yang (1993) used it for detecting defects in effect of the sensing band inclination, baby suckers per
apples and Campins et al. (1997) for apple stem and row and size on chilli alignment was studied. Chilli-
calyx identification. A chilli size sorter used structured sorting accuracy was tested on fruits with and without
light to determine product length (Hahn & Mota, 1997a; peduncles. Synchronized pulsing was used and chilli
Hahn et al. 1997b). misclassification owing to speed changes was evaluated.

1537-5110/02/$35.00 433 # 2002 Silsoe Research Institute. Published by


Elsevier Science Ltd. All rights reserved
434 F. HAHN

Table 1
Length and width measurements obtained for each volume category
Sorting class Feature Minimum, mm Maximum, mm Average, mm
First (big) Length 56 92 7935
Width 22 36 2925
Second (medium) Length 37 75 6616
Width 19 38 2836
Third (small) Length 22 59 5107
Width 12 36 2372

standards as volume. Misclassified chillies were shown


100
to the panel, who accepted that subjective measurements
were not always consistent.
80
Detection accuracy, %

On-line volume measurements are difficult to obtain,


so width-to-length ratio was used to determine the best
60
feature to predict refilling size. For both volume and
40
width-to-length ratio measurements, it was noted that
width was a better feature to sort chillies for refilling
20
(Fig. 1). Chillies that were going to be sorted by
width required non-overlapping thresholds. Chillies
0
under 24 mm in width were considered thin (third
Length Width quality) and were not used for filling. Chillies over
Chilli feature 29 mm in width were considered top quality and those
ranging from 24 to 29 mm in width were considered
Fig. 1. Length and width detection accuracy using as criteria: , second class.
volume; and , width/length ratio.

3. Equipment development and procedures


2. Theoretical considerations A prototype designed to measure produce volume was
used to study chilli width. Chilli width deformation and
Refilling chillies automatically is not an easy task, so
misalignment was studied after acquiring chilli images at
the sorting judgement of a panel of experts was
5, 10, 20 and 308. The prototype used an ELECTRIM
compared against real chilli measurements. Width and
1000N monochromatic charge coupled device (CCD)
length measurements from 1000 chillies were obtained
camera (Electrim Corp., Princeton, NJ) to revolve
with a vernier. The volume of each chilli was obtained
around the sampled chilli. The prototype (Fig. 2) uses
using the Archimedes principle and three groups were
a white panel to reflect white light coming from the
formed according to chilli volume. First and third class
halogen lamp. The object is back-illuminated providing
chillies presented volume values ranging between 50–
a shadow-free image. A stepper motor controlled the
65 ml and 20–35 ml, respectively. Maximum, minimum
movable arm geared by a chain and positioned the
and average measurements of length and width for each
camera at the right angle.
volume group are shown in Table 1.
A panel of experts classified 300 chillies using their
own particular refilling standards. Volume is the best- 3.1. Sorter design
filling indicator, but if the length–width ratio is as
accurate from the panel’s point of view as volume, it will A system was designed with two 20 cm wide belt
be easier to sample. The chillies selected for the study conveyors. The second (sensing) band ran at a speed of
had minimum length and width values of 6.5 and 2.5 cm, 60 m min 1 sorting up to 15 chillies per second. The
respectively. Expert subjective sorting was compared speed of the second band was twice that of the first, and
with classification based on volume and width/length separated chillies for individual sensing and sorting.
ratio measurements. The experts correctly selected only Chillies cannot roll owing to their peduncles, but when
75% of the volume-measured chilli. Width–length ratio they fall from the first band they bounce. The baby
was as effective in determining chilli size using expert suckers stop the chillies without causing impact
JALAPEÑO CHILLI GRADING 435

Fig. 2. Prototype for measuring chilli features; CCD, charge coupled device.

Radiometer

Sizer
Piston
Sorter

Acrylic

Fig. 3. Conveyor belt for measuring chilli width and necrosis.

damages. An acrylic band below the surface belt allow band and detected both vertical and horizontal chilli
proper baby sucker positioning on the surface (Fig. 3). widths.
Baby sucker rows are spaced 10 cm apart along the Conveyor belts without individual bicone rollers
entire belt surface. It is essential to position the chillies require an additional sorting device. This particular
parallel to the laser line to avoid erroneous measure- sorter was required to classify three different-sized
ments. chillies and mechanical sorting was implemented using
Two sensors were used, one detected chilli necrosis two flat metal sheets at the end of the conveyor belt
optically and the other measured chilli width (Fig. 3). (Fig. 4). Two relays controlled the position of the metal
The first detector was a radiometer, which activated a sheet that determined the trajectory of the chilli. During
piston for removing chillies showing necrosis. The the classification of the first size, the top plate almost
chilli-width sensor was located close to the end of the touched the band, letting the chilli fall to lateral band I.
436 F. HAHN

100

80

Reflectance, %
60

40

20

0
500 557 609 661 714 766 818
Fig. 4. Mechanical sorting device. Wavelength, nm

Fig. 5. Spectral signatures of chilli: , healthy tissue; ,


The medium sized chillies fell on to lateral band II by
necrosis; , baby suckers
gravity without touching any plate. The small sized
(third) chillies rolled over the bottom plate, which is
guided to band III. During the sorting of the third size,
movement was synchronized with the radiometric
both sheets were separated from the band and only the
measurement.
bottom plate relay was on.

3.2. Spectral detection of necrosis 3.3. Size detector

One hundred healthy chillies and 100 showing The width sizer consisted of a laser diode line
necrosis were measured using an OCEAN OPTICS generator (Applied Laser, Pasadena, CA), which pro-
PC1000 computerized spectrometer in the 500–850 nm vided a structured vertical stripe and a bar of photo-
range. A tungsten-halogen lamp (model LS1, Ocean detectors standing vertically on the opposite side of the
Optics, Dunedin, FL) provided the lighting on the band [Fig. 6(a)]. The photodetector bar consisted of 32
desired waveband and was regulated by a SOLA voltage silicon photodiodes (HAMAMATSU, 1226-18BK,
regulator. An optical bundle illuminated the sample, Bridgewater, NJ) split 1.25 cm between them. Chillies
and carried the reflected radiation towards a mono- were always positioned with the peduncle facing the
chromator. Spectralon diffuse reflectance material (Lab- laser.
sphere, Inc., North Sutton, NH 03260) was used as Each chilli’s horizontal Wh and vertical width Wv was
spectral reference due to its high reflectivity (98–99%) in evaluated. Horizontal width measurements were
the 250–2500 nm range. Chilli spectra were acquired and obtained after counting the number of blocked laser
compensated using the dark and reference spectra. stripes as the chilli advanced through the conveyor belt.
Relative reflectance showed spectral differences at the The photodetector bar scanned the laser stripe every
visible band (Fig. 5). Necrotic tissue absorbed all the 20 ms [Fig. 6(b)], and photodetector values were com-
visible radiation and its reflectance peak at 720 nm was pared with a threshold. The digital outputs were saved
twice that of the healthy tissue. The baby sucker in the memory of the 68HC11 microcontroller. Another
spectrum differed in magnitude and in shape from the silicon photodetector array (Hamamatsu, 1226-18BK)
chilli spectrum. positioned in front of the spot laser (Ealing, 21-6556)
A radiometer was built with two silicon photodetec- detected the presence of the baby sucker [Fig. 6(c)]. The
tors, which measured the radiation passing through two 0.9 mW spot laser was placed 35 cm ahead of the line
10 nm bandpass filters (EALING 35-3698 and 35-4217) generator laser considering the maximum chilli width
at 550 and 720 nm. Radiation was collected with a lens (Table 1).
directing it towards the photodetectors. The signal Sorting accuracy was tested with a trial group of 200
obtained from both photodetectors was converted chillies, initially with the peduncle and subsequently
to digital and sent to a 68HC11 MOTOROLA micro- without it. A bigger trial group of 1000 chillies was
controller which activated a piston whenever the tested taking into account an optimum inclination, baby
ratio between the 550 nm value and the 720 nm was suckers and speed compensation to know the final
lower than 0.1. A detector ensured that the produce machine efficiency.
JALAPEÑO CHILLI GRADING 437

Chilli
Photodetectors Wh

Baby

Size
sucker
WV

1 2 3 4 5 6 7 8
(a) Band (b) Sample

Photodetectors Photodetectors

Line Spot Line Spot


(c) laser laser (d) laser laser

Fig. 6. (a) Laser line generator (laser stripe) and photodetectors; (b) sampling measurements taken with the photodetectors; (c)
well-aligned chilli; (d) badly aligned chilli; Wh, chilli horizontal width; Wv, chilli vertical width

4. Results and discussion Table 2


Average horizontal width increments for variable length chillies
A sensorial analysis carried out by a panel of experts having different inclinations in front of the sensor
evaluated chilli width as the best feature for classifying Chilli length, mm Average horizontal width increment
chillies for refilling. Chilli inclination is crucial during based on chilli inclination, %
horizontal width sensing and the prototype (Fig. 2)
58 108 208 308
which rotated a CCD camera around the products
measured width under different inclinations. Horizontal 60 1447 2596 7447 9574
and vertical widths were measured for 200 chillies. Chilli 65 105 3952 6753 110
70 596 1954 4737 8947
back-illuminated images with a 308 inclination showed 75 722 205 5554 10622
horizontal widths twice that of real width measurement. 80 966 2033 5333 8333
Table 2 shows that chillies of 6.5 and 8.5 cm in length 85 1702 2553 7489 11276
show increments of 110 and 112.76%, respectively.
Horizontal width measurements also increased when
chilli shapes curved closer to the apex. Straight chillies suckers prevents chilli damaging, and the flexible rubber
without curvatures showed similar horizontal and structure helps to reduce chilli bouncing. Band slope is
vertical diameters, with an overall average diameter important to chilli alignment but limits band length. If
ratio of 096. the number of chillies to be sorted increases, additional
As chilli misalignment produces sorting errors after parallel bands should be added.
passing in front of the laser sensor, conveyor belt Chillies from each of the three different sizes were
performance was tested in three different experiments to placed on the first band and their alignment with respect
detect the causes of misalignment. The first of three tests to the laser sensor was measured manually. The sensing
studied how chillies pass from the first band to the band had a 58 slope and two baby suckers per row.
sensing band. The second test analysed the effect of the Misalignment problems resulted from chillies bouncing
slope of the sensor belt on chilli alignment. Finally, or turning on the sensing belt after falling from the first
misalignment caused by using rows of two or three baby band. Turning is produced when the stiff peduncle
suckers was studied. In a previous chilli-sorting proto- touches the sensing band before the body of the chilli.
type developed by Mota (1996), the peduncle was often When this happens, the chillies rotate one-quarter turn
separated from the base of the chilli, reducing its quality and can advance or retrocede. If the body touches the
and increasing postharvest losses. The use of baby band before the peduncle, the chilli only bounces.
438 F. HAHN

Bouncing and turning caused improper positioning in Table 4


more than 60% of the chillies, independent of product Third class chilli alignment according to band inclination
size. It should be noted that first and third class chillies Band slope, Proportion of chilli at
presented misalignments of more than 158 for 37.7 and deg various alignments, %
41.6% of the sampled products (Table 3). Bouncing 08 108 158
disappears in chillies that have been stored for more
than 36 h at 208C as their texture changes. 5 354 229 416
10 663 71 265
In the second experiment, chillies were placed on top 15 837 25 137
of the sensor band to study misalignment caused by
band inclination. The slope of the sensor band was
modified and chillies slipped until they stopped at the
baby suckers. Once the chilli stopped, its inclination addition of the spot laser, baby sucker size did not affect
with respect to the laser scanner was evaluated final chilli width measurements.
manually. The baby sucker acted as a trap or break- The microcontroller program executed three different
point, which limited chilli slipping. With the peduncle routines: the waiting cycle, the product passing routine
side heavier, chilli slipped turning around the centroid. and the baby sucker detection. During the waiting cycle,
Rotation depended on where the chilli hit the baby all the photodetectors were on, as no chillies interfered
sucker and less turning occurred when the contact with the laser stripe and the spot laser. The microcon-
was near the centroid. It was noted that 837% of troller did not obtain scanned data from the photo-
the chillies were properly aligned when the detector bar for storage. Once the chilli blocked part of
slope of the sensor band was increased to 158. the laser stripe, the microcontroller was interrupted,
With a 58 slope, only 35.4% were properly saving the scanned data on memory. In this product
aligned (Table 4). passing routine, the photodetector in front of the spot
The effect of the number of baby suckers per row on laser remained on, until the baby sucker appeared. In
chilli alignment was studied in 100 chillies. Chillies were the baby sucker routine, the spot laser beam gets
placed 5 cm away from the baby sucker and after the blocked and the scanned data from the photodetector
sensor band was turned on, the chillies slipped until they bar was not acquired by the microcontroller memory.
stopped at the baby suckers. As with previous experi- The program cycles back to the waiting routine, once
ments, chilli inclination was evaluated manually. When the photodetector bar detects the full laser stripe. The
two baby suckers were used, the peduncle attached itself number of memory locations stored in the microcon-
to the baby suckers on several occasions, causing troller memory provided the horizontal width of the
improper positioning. Chillies aligned orthogonally to chilli.
the laser scanner showed huge widths. Misalignment on Each scan shows a different number of photodetectors
50% of the sampled chillies was encountered on the 158 that do not receive the laser stripe, this quantity being
sensor band slope when only two baby suckers were saved in the microcontroller memory. The maximum
used in each row. Inserting a third baby sucker in each value saved in the memory corresponded to the vertical
row with the same sensor band slope increased proper width. Both horizontal and vertical chilli widths were
chilli alignment to 85%. In previous designs, baby compared and when the values differed, the minimum
sucker size was important for sensor performance, with value was chosen. The final width value was thresholded
the larger baby suckers working better. Nevertheless, in order to activate the mechanical sorter relays.
they required larger photodetector coverage. With the Considering that the chilli-feeding to the band is
constant and the weight is not considerable, speed
changes could only be caused by the line power supply.
Speed changes caused problems in chilli width detection
Table 3 when fixed time sampling was used. A decoder was
Percentage of different sized chillies aligned at 0, 10 and 15 introduced to synchronize the sampling period with the
degrees in front of the laser scanner after falling from the first band speed in an effort to increase sorting accuracy
band
(Mota, 1996).
Sorting class Proportion of chilli at various alignments, % The sorting sensor was evaluated after checking chilli
08 108 158 alignment performance. Two hundred chillies were
placed on the second band to avoid bouncing and the
First (big) 444 177 377 effect of the peduncle on the laser measurements was
Second (medium) 35 25 40
Third (small) 354 229 416 studied. Chillies without peduncles were sorted with
better accuracy, with first class chillies having the
JALAPEÑO CHILLI GRADING 439

Table 5
Sorting accuracy for each chilli class with and without peduncle
Sorting class Sorting accuracy of chilli without peduncle, % Sorting accuracy of chilli with peduncle, %
Normal scan Controlled scan Normal scan Controlled scan Both bands
First (big) 922 931 913 923 872
Second (medium) 911 922 892 902 862
Third (small) 901 917 843 873 871

highest sorting accuracy (Table 5). Third class chillies trial group. The 500 necrotic chillies were placed with
with and without peduncles achieved success rates of the damaged part facing upward in the sensor band. The
84.3 and 90.1%, respectively. Horizontal width mea- signal obtained from both photodetectors was acquired
surements increase when the peduncle blocks some of by the microcontroller, which activated a piston when
the photodiodes, while vertical width measurement will the reflectance ratio was close to zero. Detection of
provide a closer measurement. The laser system can necrotic samples was achieved with a 963% success
improve its performance if slight modifications are made rate.
to the program. It should include a routine, which could Necrosis radiometric measurements worked well, but
eliminate the peduncle, taking away thin measurements as the chilli remained static only the top view was
or using first derivatives. analysed. If necrosis appears in another area, the
Controlled scanning used a decoder, which was fixed machine will not detect it; therefore, rollers are required
to the motor that moved the conveyor belt. The sensing to rotate the product in front of the radiometer. Bicone
period varied according to band speed, and overall commercial rollers should be modified for careful chilli
sorting accuracy increased (Table 5). First class chillies handling to avoid peduncle ruptures and to take
with peduncles showed success rates of 92.3 and 91.3% different peduncle sizes into account.
in controlled and normal scanning, respectively. Chilli-width sensing was over 90% accurate, while the
A trial group of 1000 chillies was tested taking into entire system sorting efficiency was of 87%. Five of
account an optimum inclination of 158, with three baby every 100 chillies were misclassified due to improper
suckers per row and speed compensation, and the final positioning in front of the sensor. Imaging techniques
results obtained are shown in the last column of Table 5. should be used to increase sorting accuracy. Two
Both bands were used and no substantial sorting cameras viewing the top and lateral chilli sides should
accuracy difference was noted between the three groups. detect chilli horizontal and vertical widths, respectively.
The basic misalignment problem was caused when
chillies passed from the first band to the sensing band.
Heavier first class chillies bounced more and presented 5. Conclusions
more misalignment problems, but the 35 mm distance
limit imposed by the program did not allow any degree The system is reliable as it classifies chillies by width
of misalignment on first class chillies. The program with extreme efficiency. It is a very cheap and simple
limited horizontal width measurements in curved chillies system, easy to clean and to repair, and it can be
to 35 mm and the sorting value was always the vertical installed in warehouses at production sites. Although
value width. The program worked efficiently and only image technology can be used successfully to detect chilli
the minimum horizontal or vertical width measurement size, most rural places where chillies are sorted prefer
was considered. However, an error was detected simple systems, which do not depend on sophisticated
whenever a chilli peduncle faced upwards at 608 angle. technology. Lack of proper service and maintenance,
Both vertical and horizontal widths increased with and difficulties in getting replacement parts are the main
respect to their normal values. reasons for acquiring simple machinery. High moisture
On-line necrosis measurements were 85% accurate contents and dust in the countryside are some of the
when only the relative reflectance at 550 nm was used. A other problems encountered on field machinery.
reference measurement was taken every 5 min and used The following conclusions can be drawn from this
to obtain the relative reflectance. The reference mea- study.
surement is not practical for on-line sensors, so a (1) Although volume is the best-filling indicator, the
reflectance ratio between 550 and 720 nm was used. Five width/length ratio is as accurate from a panel of
hundred necrotic and 500 healthy chillies were used as a expert points of view.
440 F. HAHN

(2) The best band parameters are given at an inclina- Hahn F; Mota R (1997a). Nobel chile jalapen* o sorting
tion of 158, three baby suckers per row and speed using structured laser and neural network classifiers.
compensation. In: Image Analysis and Processing (Alberto del
Bimbo, ed), pp 517–523. 9th International Conference,
(3) The width sorter measures horizontal width using ICIAP’97, Florence, Italy, September 1997. Elsevier,
laser stripes, limiting the maximum size to 35 mm. Amsterdam
The width sorter measured the vertical width using Hahn F; Mota R; Vazquez M (1997b). Nobel automatic
a photodetector array. A spot laser detected the jalapen* o chilli classifier fits inside a trailer. ASAE Paper No.
baby sucker presence. 973054. ASAE, St. Joseph, MI 49085-9659, USA
Howard M S; Searcy S W (1989). Algorithm for grading
(4) The minimum horizontal or vertical width gave the carrots by machine vision. ASAE Paper No. 897502. ASAE,
final width. St. Joseph, MI 49085-9659, USA
(5) First class chillies were sorted with success rates of INEGI (1999). Centro de Estadistica Agropecuaria [Centre of
92.3 and 91.3% with and without speed scanning, Statistical Agriculture.] SAGAR, Mexico
respectively. Miller W M; Throop J A; Upchurch B L (1998). Pattern
recognition models for spectral reflectance evaluation of
(6) Chilli necrosis was detected with a 96.3% accuracy apple blemishes. Journal of Postharvest Biology and
using a radiometer, which measured at 550 and Technology, 14, 11–20
720 nm. Mota (1996). Seleccio! n automatica de chiles jalapen* os por
taman* o. [Automatic size sorting of jalapen* o chilli]. Tesis de
Maestria, Instituto Tecnologico de la Laguna, Torreon,
Mexico
Acknowledgements Peleg K (1985). Produce Handling Packaging and Distribu-
tion. AVI Publ. Co., Westport, CT, USA
I would like to acknowledge Ing. Francisco Rafael SARH (1982). Presente y pasado del chile en Mexico. [Present
and M.C. Rafael Mota for their help in gathering the and past of chilli in Mexico] Instituto Nacional de
Investigaciones Agricolas, Mexico City
data and Lisa Harrison who made this paper readable. Tao Y; Morrow C T; Heinemann P H; Sommer H J (1995).
Fourier-based separation technique for shape grading of
potatoes using machine vision. Transactions of the ASAE,
References 38(3), 949–957
Throop J A; Aneshansley D J (1999). Inspection station detects
Benady M; Miles G E (1992). Locating melons for robotic defects on apples in real time. ASAE Paper No. 993205.
harvesting using structured light. ASAE Paper No. 927021. ASAE, St. Joseph, MI 49085-9659, USA
ASAE, St. Joseph, MI 49085-9659, USA Throop J A; Aneshansley D J (1997). Inspection of processed
Campins J; Throop J A; Aneshansley D J (1997). Apple stem fruit before and after peeling. ASAE Paper No. 976044.
and calyx identification for automatic sorting. ASAE Paper ASAE, St. Joseph, MI 49085-9659, USA
No. 973079. ASAE, St. Joseph, MI 49085-9659, USA Yang Q (1993). Finding stalk and calyx of apples using
Crowe T G; Delwiche M J (1996). Real time defect detection in structured lightning. Computers and Electronics in Agricul-
fruit, part I: design concepts and development of prototype ture, 8, 31–42
hardware. Transactions of the ASAE, 39(6), 2299–2308

You might also like