Note: Descriptions are shown in the official language in which they were submitted.
CA 02280364 1999-08-16
This invention relates to a system and process and
apparatus for the classification and sorting of wheat and
other cereal grains using machine imaging techniques.
Inspection of wheat or other cereal grains is a
subjective process. A current system of wheat and cereal
grain grading is based upon Kernel Visualization
Distinguishability or Single Kernel Characterization
(Psotka, J. 1999) as determined on site by the local grain
inspector or agent. As such, the process is subject to the
vagaries of continual human intervention, e.g. personal
biases, fatigue, poor training, etc. Furthermore, the
standards for the system are set on a yearly basis and, in
many cases, are reset during harvest if the weather
conditions become extreme, i.e. if there is too much
rainfall or if an early frost has occurred. Hence, there
may not be time to retrain all staff at all terminal
locations or grain elevators adequately to provide a
standard inspection system.
Inspection of wheat or other grains is a labour
intensive process. At each wheat and cereal grain elevator
and at each terminal an inspection of the wheat or grain
must be carried out. Such inspection may include weighing,
determination of dockage, determination of moister content,
determination of protein content, detection of the presence
and extent of fungal growth, detection of the presence and
extent of the chemical residues, etc. Furthermore,
inspection of a grain shipment performed at an elevator may
be repeated at a terminal if an extended period of time has
CA 02280364 1999-08-16
- 2 -
lapsed or if the shipment is selected for a foreign market
or specialized market. Hence, multiple and detailed
inspections of wheat and cereal grains are required in many
cases to assure their quality.
There is no inexpensive objective system available to
farmers for such inspections so that they may judge
accurately the quality of grains from different areas of the
farm. If such a system were available a farm could avoid
mixing a batch of high quality grain with a batch of poor
quality grain with the result that the quality -- and hence
the financial recompense -- may be reduced for the mixed
batches. Of course, a competent farmer can judge
subjectively but, at least in marginal cases, subject
judgement may lead to irretrievable mistakes in strategy.
Once differing grades of grain have been mixed, they can not
be separated.
Optical scanning and machine imaging of material for
quality control is a standard practice in industry.
Application of this technology to the classification of
wheat and other cereal grains has been investigated using
Fourier analysis of grain shape and relative position of the
centre of gravity of the kernel (Segerlind and Weinberg
1972, Barker, Vouri and Myers 1992), pattern classification
using recursive learning (Brogan and Edison 1974),
statistical pattern recognition based upon morphological
features (Zayas, Pomeranz and Lai 1985, Zayas, Lai and
Pomeranz 1986, Neuman et al. 1987, Symon and Fulcher 1988a,
CA 02280364 1999-08-16
- 3 -
1988b), statistical pattern recognition applied to three-
dimensional images (Thomson and Pomeranz 1992), statistical
pattern recognition applied to colour images (Neuman et al.
1989a, 1989b) and pattern recognition using neural networks
as applied to colour images (Egelberg, Mansson and Peterson
1994, Lou 1997). The usefulness of these algorithms in
solving classification problems related to wheat and other
cereal grains in real-time applications have been
demonstrated in the determination of corn kernel size
distribution (McDonald and Chen 1990), in particular, and
morphological classification problems in general (Haralick,
Sternberg, and Zhuang 1987, McCubbery and Lougheed 1985,
Kimmel et al. 1985).
Given the demands of wheat and cereal grain
classification mentioned above, it would be desirable to
couple computer algorithms and high-speed computers which
can implement these algorithms in real-time with a real-time
image processing system to the classification of wheat and
other cereal grains and also other particulate material.
Accordingly the invention provides a system for
classification of particulate material e.g. grains such as
cereal grain in comparison with a set standard, the system
comprising imaging means for providing images of grains in a
sample of said grains, computer means programmed to grade
the sample in comparison with said set standard from data
obtained from the images.
CA 02280364 1999-08-16
- 4 -
The sample of said grains or other particulate material
may be presented to the imaging means scattered on a
transparent sheet to display at least some individual
grains, which transparent sheet conveys the grains past said
imaging means. The transparent sheet may by Mylar
(trademark) .
Indeed, for any system for display of individual
grains, it is advantageous to convey the grains past a
display point on a transparent conveyor. Any stationary
display surface would accumulate dust, dirt and debris which
would eventually lead to inaccurate grading.
The transparent sheet may be conveyed from a roller
over, for example, a colour linear photo diode array of a
flatbed scanner by means of a winding powered roller which
winds used dusty or dirty Mylar.
The set standard refers to qualities of different
characteristics of the grains, for example, colour, size,
roundness, roughness -- to mention a few.
The standard may be altered from time to time. For
example, in Canada for wheat, the standard may be set by the
Canadian Grain Commission. If the quality of wheat
consistently rises for several years due possibly to
improved environmental conditions and improved strains of
wheat, it may be desirable to reset the standard upwardly.
In other countries, standards may be set differently.
CA 02280364 1999-08-16
_ 5 _
Whatever the standards, it is probably that similar
characteristics may be measured.
Any chosen standard may be made universal by the
publication on the Internet which may be downloaded by
interested parties, e.g. farmers, elevator operators, etc.
The computer program for the grading system may then be
updated using suitable simple software.
The sample should be of a size such as to be
representative of the batch of granular material to be
graded. Indeed, there is different samples from different
regions of the batch may be taken. However, the sample size
presents some difficulties in obtaining fast results. It is
preferable that the system should incorporate statistical
analysis software including an automatic learning network
(ALN). This allows for the setting up of a linear equation
so that statistical clusters may be evaluated in the grading
procedure. Of course, all final adjustments of parameters
must remain in the hands of the standard maker and these
parameters are programmed into the computer before operation
of the system.
Such a system for grading a small sample may be used as
a farm gate machine for the convenience of the farmer. It
may also be configured as intermediate size apparatus for
use , for example, at small grain elevators. It may
alternatively be utilized as a stage in large scale
apparatus for grading. Such apparatus may be utilized at
CA 02280364 1999-08-16
- 6 -
grain elevators, grain terminals and the like. In such
apparatus the volume and weight of the grain are determined,
and the grain is measured for dockage. Any presence of a
chemical residue is determined, possibly using a gas mass
spectrometer. The protein and moisture content are measured
utilizing a NIR (near infrared) scanner. Fungal growth is
detected using an ultraviolet excited fluorescent scanner.
The W excited fluorescent scanner may be a sophisticated
apparatus component capable of not only showing fluorescence
when fungal growth is present but also indicating its
extent. The grain may then pass to a singulator for
division into discreet samples each of which may be
inspected automatically utilizing the system described
above. It may be more convenient to utilize a camera, for
example a linear array digital or analogue camera, in place
of the flat bed optical scanner previously referred to.
Also, in such large scale apparatus it may be possible to
provide some separation of the differently graded grains
downstream of the imaging means, e.g. the camera. Such
separation of the grains may be, for example, by pneumatic
means such as air sorter. It is possible that other means
such as knife-edge separator may be used. If it desired to
isolate any particular group of grains such as wild oats
which are difficult to detect in conventional dockage
measurement, a proximity detector may be utilized to
indicate the region of these grains and the sorting means
may be applied in that region.
CA 02280364 1999-08-16
The invention includes a method of grading particulate
material utilizing the system of the invention.
Embodiments of the invention will now be described with
reference to the drawings in which:
Figure 1 shows a schematic view of a small scale system
for grading a manually loaded sample;
Figure 2 is a flow-chart showing the operations of the
system illustrated in Figure 1;
Figure 3 is a schematic view of a grading system for
individual grains as part of a large system for bulk
grading; and
Figure 4 is a flow chart showing the operations of the
system of Figure 3.
Figure 1 shows a simple system which may be provided
quite inexpensively and which is simple to operate. A
farmer may use it on his own premises to give information
concerning batches of grain, for example, wheat, from
different regions of his farm. In this case a small sample,
for example one cup full, of grain is utilized.
A sample of particulate material 14 is scattered in a
single layer onto the surface of a transparent Mylar film
16. Scattering may be automatically controlled from a sample
container 12 which may be utilised to ensure samples of
uniform volume. However, scattering may equally be manual.
The Mylar film 16 is conveyed above a colour linear
CA 02280364 1999-08-16
_ g _
photodiode array 18 from a free roller 20 onto a driven
roller 22. Tension may be maintained on the Mylar sheet
using a braking system 23 of any suitable type on roller 20.
The driven roller 22 may be driven by motor 24.
The sample 12 scatters indiscriminately on the Mylar
sheet carrying with it any dockage such as dust, dirt or
other debris scooped up with the sample. It should be
noted, however, that it is entirely possible to utilize the
system as part of a large scale grading operation. In this
latter case, it is probable that the grain supplied onto the
Mylar sheet will be single line or multiple line feed of
individual grains.
The Mylar sheet between rollers 20, 22 may be supported
on a transparent platen 26 between the Mylar sheet 16 and
the photodiode array 18.
A vacuum hold down device 28 is utilized to remove air
from between the Mylar sheet 16 and the platen 26.
The photodiode array 18 may be any convenient array
sensitive to coloured light and calibrated using a standard
RGB palette. Each photodiode is typically 1024 pixels
mounted transversely to the direction of motion, each pixel
being 50~m wide (e. g. an EG&G Reticon K-series wide aperture
linear photodiode array and necessary colour filters). A
source of light 30 illuminates the sample 12, the reflected
CA 02280364 1999-08-16
- 9 -
light from the sample being received by the photodiode array
18.
More than one photodiode array may be mounted parallel
to each other and transversely to the sample flow, each
photodiode array possibly having its own source of light.
Alternatively, the photodiode arrays may be arranged in
longitudinal, staggered rows each array extended
transversely to the direction of motion.
While this photodiode array system is described with
reference to a randomly scattered sample on the Mylar sheet
16 it is worth noting that when the system is part of larger
grading system and allowing ten line feeds of grain from a
singulator to cross a 1024 pixel photodiode array, then 200
grains of the sample may be analysed and classified per
second.
The photodiode array 18 sends details to computer 32
utilizing an automated learning network 33(ALN).
Individually scanned grains may be allocated a performance
number for each characteristic. The performance numbers are
allocated in accordance with the command program of the
computer which itself is conformed to the standards set by
the standards maker. Statistical analysis software,
including the ALN, sets up a linear equation so that the
computer may recognize statistical clusters of certain
characteristics. Each characteristic has a vector and
similar vectors are grouped together. The vectors are the
CA 02280364 1999-08-16
- 10 -
detected values composed of parameters like length, width,
aspect ratio, convexity, first and higher order moments of
pixel coordinates, averaged IGB colour coordinates, and
morphological parameters like roughness of perimeter ,etc.
Thus there is little possibility, in view of the statistical
grouping, that, for example, some shrivelled grain would be
confused with frost bitten grain.
The computer system itself may comprise a central
microprocessor of any suitable type for example a 586
(Pentium) microprocessor may be used. The computer system
also comprises all necessary peripherals for the storage of
data, reproduction of data and/or analysis performed on the
data, connections to the various components of the machine
and any external connections deemed desirable for the
efficacious operation of the machine.
Figure 2 shows a flow chart showing the sequence of
operations described with reference to Figure 1.
Figure 3 shows a schematic view of another embodiment
of the invention in which a system such as that described
with reference to Figure 1 is included as part of a large
scale grain grading operation. It is to be noted that the
system previously described with reference to Figure 1 is
itself a different embodiment when described with reference
to Figure 2. The photodiode array 18 has been replaced by a
camera system with resulting minor changes in mechanical
structure. The camera system is described by way of a mere
CA 02280364 1999-08-16
- 11 -
alternative imaging means but it is possible that the camera
system may be more suited to the case where the sample is
presented as a single line feed.
A sample of wheat or other cereal grains is selected
randomly or by whatever means is the usual procedure for the
inspector and passes through a standard sampler device 42
(see Figure 4) (e.g. a Winchester bushel), in order to
determine the weight and volume of the sample. The sample
is then passed to device 44 (see Figure 4) for measuring the
dockage (e. g. a Seedburo Star docket tester or a Carter-
Dockage tester). It is placed on a conveyor 46 for example
a vibratory chute, and is passed through an NIR scanner 48
(e. g., a standard ultraviolet "black-light" and NIRS Systems
Model 6500 scanning monochromator operating in reflectance
mode between 400 and 6500 nm) scanning for fungal damage,
moisture content and protein content. Next the sample may
pass onto an aspirator e.g. a Bates Laboratory Aspirator.
It also passes gas mass spectrometer 52 used to detect any
residues of insecticides, herbicides, etc., and grain
determinants, e.g. bin burn. It should be noted that in
this embodiment any of devices 42, 44, 46, 48, and 52 may
be interchanged in position or wholly removed or any
combination thereof depending upon the desired application,
cost, etc. The standard maker is likely to use a large
scale machine which allows for direct verification of
accuracy. The standard can then be applied to small and
intermediate scale machines and to other large scale
machines.
CA 02280364 1999-08-16
- 12 -
The sample 40 now passes onto device 54, a singulator
(e.g. Seedburo Count-a-Pak Seed Counter) which produces a
single line or multiple lines of wheat or other cereal
grains. Upon exiting from device 54 the singulated sample
40a falls under the force of gravity onto belt conveyor 58
and is scanned using a series of two or more line scan
cameras 60 (e. g. LC 1911 Reticon Modular Line Scan Cameras
which may operate with 256, 512, 1024 or 2048 elements and
have pixels of l3,um or 26,um and may operate up to 35000
scans per second). Cameras 60 are equipped with all of the
necessary optics and filters to be colour sensitive and are
calibrated using a standard RGB palette. More cameras and
channels from the singulator(s) may be used to increase the
speed with which a sample may be analysed. The height
through which the sample falls may also be varied as desired
to meet any space requirements, etc. After passing cameras
60 the sample is separated by air sorter 62 or some other
suitable device which allows some of the sample to pass and
fall to a suitable receptacles 64, and the remainder of the
sample to continue on to the next conveyor system.
The singulated sample 40b then passes onto another
conveyor belt 66 or other transport system, and is
transported past a plurality of air sorters 68 each
associated with a proximity detector, which allows for an
unambiguous determination of the location and sorting of the
individual wheat or other cereal grain kernels. The air
sorters 68 remove the individual wheat or cereal grain
kernels and place them into final containers 70.
CA 02280364 1999-08-16
- 13 -
The cameras 60 are linked with a computer32 and ALN 33
similar to that described with reference to Figures 1 and 2
in a similar manner.
