According to the Bulgarian Government Standards [2,
3], the main characteristics of grain quality are appearance,
shape, color, smell, flavour, moisture content, presence of
impurities—grain and non-grain ones. Whole grains with
appearance, shape and color inherent for the variety and
hybrid, as well as broken grains bigger than the half of the
whole grain, are considered to be standard. There are
several groups, which are considered as grain impurities:
broken grains smaller than the half of the whole grain;
heat-damaged grains, burned grains with damaged kernels;
small, shriveled and green grains; sprouted grains, grains
with moldy germs, dark brown to black; Fusarium infected
grains. The group of the non-grain impurities consists of:
damaged grains, corn-cob particles, leaf and stem fractions,
pebbles, soil, sand, smutty grains, as well as harmful ele-
ments (bunt).
Most of the grain quality characteristics mentioned
above (except moisture content, smell and flavour) are
related to the color characteristics, shape and dimensions of
the grain sample elements. The main trend in the last few
years has been to use Computer Vision Systems (CVS) in
order to evaluate such characteristics. A review of the
progress of computer vision in the agricultural and food
industry was given in [4]. To obtain a complex assessment
of the grain quality using data about color characteristics,
shape and dimensions of the grain sample elements, is a
complicated and multilevel task. This is because the color
characteristics shape and dimensions of the elements in a
sample vary within a wide range.
Many results have been published, in which color char-
acteristics analysis was used to assess some particular quality
features. A digital image analysis algorithm, developed to
facilitate classification of individual kernels of Canada
Western Red Spring (CWRS) wheat, Canada Western
Amber Durum (CWAD) wheat, barley, oats, and rye using
textural features of individual grains was reported in [5]. The
textural features of individual kernels were extracted from
different colors and color band combinations of images to
determine the color or the color band combination that gave
the highest classification accuracies in assessment of
authenticity of cereal grains. Color characteristics analysis
was used also to assess class [5–7], infections [8, 9], ger-
mination [10], weed identification [11], etc.
Morphological features, related to the grain shape and
geometrical parameters were used for assessment of the
grain variety. A set of eight morphological features
namely, area, perimeter, length of major axis, length of
minor axis, elongation, roundness, Feret diameter and
compactness to recognize five different kinds of cereal
grains was presented in [12]. A broader investigation, with
a total of 230 features (51 morphological, 123 color, and 56
textural) used for classification of barley, Canada Western
Amber Durum wheat, Canada Western Red Spring wheat,
oats, and rye was presented in [13]. Assessment of the
grain sample purity was performed by profile analysis of
corn kernels using one-dimensional digital signals based on
its binary images [14], by modeling the shape using a set of
morphological features [15] and by shape curvature anal-
ysis [16]. Computer vision methods were also used for
determination of different injuries: corn (Zea mays) kernel
mechanical damage and mold damage [17], whole and
broken kernel identification [18], etc.
It is very important to use proper classification proce-
dures in order to assess precisely the main features of grain
sample elements. Different approaches were used for this
purpose. Some effective results were obtained using clas-
sification based on multi-layer neural networks (NN) [19].
An evaluation of the classification accuracy of nine dif-
ferent neural network architectures to classify five different
kinds of cereal grains was performed in [12]. A comparison
between a four-layer backpropagation neural network and a
non-parametric statistical classifier when solving a similar
task was presented in [15]. The results show that the neural
network classifier outperformed the non-parametric clas-
sifier in almost all the instances of classification.
Neural Networks were applied for quality analysis of
grain sample elements in combination with color charac-
teristics [13, 20], shape and dimensions of the objects [12,
16, 20]. They were used for evaluation of the class [12, 15],
purity [14, 16], different kernel mechanical damages [20],
germination [10], infections [7, 20], weed seeds detection,
etc.
The references cited above present some results con-
cerning the assessment of specific grain quality features.
The main goal of this paper is to present the INTECHN
approach for a complex assessment of the corn grain
quality. This approach is based on the analysis of the object
color characteristics, shape and dimensions using CVS, as
well as the fusion of the results of the separate analyses.
The final goal of the complex assessment is the categori-
zation of the grain sample elements in three quality groups.
The INTECHN platform approaches, methods and tools for
recognition of the color characteristics and shape of grain
sample elements, as well as the approaches for fusing the
results from color and shape analysis, are presented. Some
of the results obtained at this stage of project implemen-
tation are given too.
Materials and methods
INTECHN platform hardware
The INTECHN platform CVS (Fig. 1) includes two color
CCD cameras (DFK31AU03, THE IMAGINGSOURCE,
Germany) (1) with lenses (PENTAX B2514D, Hoya
112 M. I. Mladenov et al.
123