View of The Influence of Colour Features on Seed Identification Using Machine Vision

(1)

3Azad University of Mashhad, Faculty of Agriculture, Department of Crop Science, Iran;

Abstract

Little studies have been done on morphology of medicinal plants seeds. This paper presents an automatic system for medicinal plant seed identification and evaluates the influence of colour features on seed identification. Six colour features (means of red,

colours of the seed surface, as well as means of hue, intensity and saturation) were extracted by algorithm and applied as ne Different combinations of colour features (one, two three, four, five and six colour features) were used to f

combination for seed identification. Results showed that the six colour feature was the most accurate combination for seed id (99.184% and 87.719% for training and test of neural network respectively). One colour featu

seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of Keywords: automatic system, colour grading, hue, medicinal species

Introduction

Although medicinal plants play an important role in the drug industry and health care, and thus draw much attention, few studies have been conducted on their seed identification. Knowledge of seed

in theoretical botany and could be useful within seed identification for seed testing, seed quarantine, seed dispersal and soil seed bank studies (Jensen, 1995).

Recently, machine vision has become a useful technology for quick seed

et al., 2001). Recent advances in hardware and software have enabled machine vision and imaging systems to identify, analyse and display finer details of objects from their digital images (Paliwal

In laboratories, the most common method for cultivars’

identification is to compare morphological characteristics of seeds with standard samples. Such characters include length, width, thickness, shape, weight, hilum colour and seed coat colour (Cope

Studies showed that colour is a useful characteristic to divide different varieties based on seed coloration such as red-, amber- and white

for categorizing them into classes (Zhang Lev‐Yadun and Ne'eman,

Colour is one of the most important features in seeds classification and grading. Different seeds and their varieties are identified by their colours. Thomson and Pomeranz (1991) classified the Western Canadian wheat to six groups

Received: 01 Jan 2016. Received in revised form: 09 Feb 2016. Accepted:

The Influence of Colour Features on Seed Identification

Sepideh

1Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran Azad University of Mashhad, Faculty of Agriculture, Department of Crop Science, Iran;

studies have been done on morphology of medicinal plants seeds. This paper presents an automatic system for medicinal plant seed identification and evaluates the influence of colour features on seed identification. Six colour features (means of red,

seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of automatic system, colour grading, hue, medicinal species

Although medicinal plants play an important role in the drug industry and health care, and thus draw much attention, few studies have been conducted on their seed identification. Knowledge of seed

in theoretical botany and could be useful within seed identification for seed testing, seed quarantine, seed dispersal and soil seed bank studies (Jensen, 1995).

Recently, machine vision has become a useful technology for quick seed control and identification (Ureña ., 2001). Recent advances in hardware and software have enabled machine vision and imaging systems to identify, analyse and display finer details of objects from their digital images (Paliwal et al., 2003).

tories, the most common method for cultivars’

identification is to compare morphological characteristics of seeds with standard samples. Such characters include length, width, thickness, shape, weight, hilum colour and seed coat colour (Cope et al., 2012).

Studies showed that colour is a useful characteristic to divide different varieties based on seed coloration such as and white- coloured, but could not be useful for categorizing them into classes (Zhang

‐Yadun and Ne'eman, 2013).

The Influence of Colour Features on Seed Identification

Sepideh ANVARKHAH

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran

2Research

Azad University of Mashhad, Faculty of Agriculture, Department of Crop Science, Iran;

Not Sci Biol, 2016, 8(1):93

Although medicinal plants play an important role in the drug industry and health care, and thus draw much attention, few studies have been conducted on their seed identification. Knowledge of seed morphology is important in theoretical botany and could be useful within seed identification for seed testing, seed quarantine, seed dispersal and soil seed bank studies (Jensen, 1995).

Recently, machine vision has become a useful control and identification (Ureña ., 2001). Recent advances in hardware and software have enabled machine vision and imaging systems to identify, analyse and display finer details of objects from their digital

., 2003).

tories, the most common method for cultivars’

identification is to compare morphological characteristics of seeds with standard samples. Such characters include length, width, thickness, shape, weight, hilum colour and seed coat Studies showed that colour is a useful characteristic to divide different varieties based on seed coloration such as coloured, but could not be useful for categorizing them into classes (Zhang

2013).

The Influence of Colour Features on Seed Identification Using Machine Vision

ANVARKHAH Alireza

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran Research and Development Center of Astan

Print ISSN 2067 Not Sci Biol, 2016, 8(1):93

Although medicinal plants play an important role in the drug industry and health care, and thus draw much attention, few studies have been conducted on their seed morphology is important in theoretical botany and could be useful within seed identification for seed testing, seed quarantine, seed dispersal and soil seed bank studies (Jensen, 1995).

Recently, machine vision has become a useful control and identification (Ureña ., 2001). Recent advances in hardware and software have enabled machine vision and imaging systems to identify, analyse and display finer details of objects from their digital tories, the most common method for cultivars’

identification is to compare morphological characteristics of seeds with standard samples. Such characters include length, width, thickness, shape, weight, hilum colour and seed coat Studies showed that colour is a useful characteristic to divide different varieties based on seed coloration such as coloured, but could not be useful for categorizing them into classes (Zhang et al., 2012;

The Influence of Colour Features on Seed Identification Using Machine Vision

ANVARKHAH

¹

*, Ali Davari Edalat Alireza ANVARKHAH

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran and Development Center of Astan

seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of automatic system, colour grading, hue, medicinal species, seed identification, seed morphology

Print ISSN 2067-3205; Electronic 2067 Not Sci Biol, 2016, 8(1):93-97. DOI: 10.15835/

Although medicinal plants play an important role in the drug industry and health care, and thus draw much attention, few studies have been conducted on their seed morphology is important in theoretical botany and could be useful within seed identification for seed testing, seed quarantine, seed Recently, machine vision has become a useful control and identification (Ureña ., 2001). Recent advances in hardware and software have enabled machine vision and imaging systems to identify, analyse and display finer details of objects from their digital tories, the most common method for cultivars’

identification is to compare morphological characteristics of seeds with standard samples. Such characters include length, width, thickness, shape, weight, hilum colour and seed coat Studies showed that colour is a useful characteristic to divide different varieties based on seed coloration such as coloured, but could not be useful ., 2012;

using a limited s

(G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red kernel types. Also, Luo

separation of healthy seeds of Western Canadian wheat from damaged ones using colour features.

Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well as for separating imma

(Ducournau

identification of weed seeds for one crop might be a major interest in the agricultural industry. It can also be useful for chemical control of weed growth (Gran

Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded genotypes

greater oil, h

seeds of black/brown seeded genotypes of these species (Rahman and McVetty, 2011). The yellow seeded

genotypes of these species had a thinner and more translucent seed coat, lower hull proportion wit

embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of black/brown seeded

digestibility of seed meal for livestock. However, the seeds’

The Influence of Colour Features on Seed Identification Using Machine Vision

*, Ali Davari Edalat ANVARKHAH

³

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran and Development Center of Astan-e-Qouds, Mashhad, Iran

Azad University of Mashhad, Faculty of Agriculture, Department of Crop Science, Iran; [email protected]

seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of , seed identification, seed morphology

3205; Electronic 2067-3264 10.15835/nsb.8.1.9743

using a limited set of colour features (mean red (R), green (G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red kernel types. Also, Luo

Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well as for separating imma

(Ducournau et al., 2004; Liu

Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded genotypes B. rapa

greater oil, higher protein and lower fibre contents than the seeds of black/brown seeded genotypes of these species (Rahman and McVetty, 2011). The yellow seeded

embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of black/brown seeded

digestibility of seed meal for livestock. However, the seeds’

Received: 01 Jan 2016. Received in revised form: 09 Feb 2016. Accepted: 07 Mar 2016. Published online:

The Influence of Colour Features on Seed Identification

*, Ali Davari Edalat PANAH

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran Qouds, Mashhad, Iran

[email protected] (*corresponding author)

combination for seed identification. Results showed that the six colour feature was the most accurate combination for seed id

(99.184% and 87.719% for training and test of neural network respectively). One colour feature had the lowest average accuracy values for seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of

, seed identification, seed morphology

nsb.8.1.9743

et of colour features (mean red (R), green (G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red kernel types. Also, Luo et al. (1999)

Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well as for separating immature seeds apart of mature ones

., 2004; Liu et al

Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded

B. rapa, B. juncea

igher protein and lower fibre contents than the seeds of black/brown seeded genotypes of these species (Rahman and McVetty, 2011). The yellow seeded

embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of black/brown seeded Brassica genotypes and reduce the digestibility of seed meal for livestock. However, the seeds’

2016. Published online:

The Influence of Colour Features on Seed Identification

PANAH

²

,

Department of Crop Science, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran

(*corresponding author)

colours of the seed surface, as well as means of hue, intensity and saturation) were extracted by algorithm and applied as ne

Different combinations of colour features (one, two three, four, five and six colour features) were used to find out the most accurate combination for seed identification. Results showed that the six colour feature was the most accurate combination for seed id

re had the lowest average accuracy values for seed identification (3.120% and 2.771%). In general, increasing the number of colour features increased the total average of accuracy values.

, seed identification, seed morphology

et of colour features (mean red (R), green (G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red . (1999) set an experiment for separation of healthy seeds of Western Canadian wheat from damaged ones using colour features.

Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well ture seeds apart of mature ones et al., 2005). In addition, early identification of weed seeds for one crop might be a major interest in the agricultural industry. It can also be useful for chemical control of weed growth (Granitto et al

Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded

and B. carinata

igher protein and lower fibre contents than the seeds of black/brown seeded genotypes of these species (Rahman and McVetty, 2011). The yellow seeded

embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of genotypes and reduce the digestibility of seed meal for livestock. However, the seeds’

2016. Published online: 16 Mar 2016.

(*corresponding author)

studies have been done on morphology of medicinal plants seeds. This paper presents an automatic system for medicinal plant green and blue colours of the seed surface, as well as means of hue, intensity and saturation) were extracted by algorithm and applied as network input.

ind out the most accurate combination for seed identification. Results showed that the six colour feature was the most accurate combination for seed identification re had the lowest average accuracy values for accuracy values.

et of colour features (mean red (R), green (G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red set an experiment for separation of healthy seeds of Western Canadian wheat Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well ture seeds apart of mature ones ., 2005). In addition, early identification of weed seeds for one crop might be a major interest in the agricultural industry. It can also be useful for

et al., 2002).

Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded B. carinata contained igher protein and lower fibre contents than the seeds of black/brown seeded genotypes of these species (Rahman and McVetty, 2011). The yellow seeded Brassica genotypes of these species had a thinner and more translucent seed coat, lower hull proportion with a bigger embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of genotypes and reduce the digestibility of seed meal for livestock. However, the seeds’

Mar 2016.

studies have been done on morphology of medicinal plants seeds. This paper presents an automatic system for medicinal plant and blue twork input.

ind out the most accurate entification re had the lowest average accuracy values for

et of colour features (mean red (R), green (G) and blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red set an experiment for separation of healthy seeds of Western Canadian wheat Seed colour images might be used also to describe seed quality and hardness, fungal damages, viral diseases, as well ture seeds apart of mature ones ., 2005). In addition, early identification of weed seeds for one crop might be a major interest in the agricultural industry. It can also be useful for Studies showed that there is a correlation between seed colour and seed quality. For example, it has been reported that the seeds of naturally occurring yellow seeded contained igher protein and lower fibre contents than the seeds of black/brown seeded genotypes of these species Brassica genotypes of these species had a thinner and more

h a bigger embryo and consequently greater oil and protein percentage (Rahman and McVetty, 2011). Proanthocyanidins and tannins are the major compounds involved in seed coat pigmentation. These are deposited in the seed coat of genotypes and reduce the digestibility of seed meal for livestock. However, the seeds’

(2)

Anvarkhah S et al. / Not Sci Biol, 2016, 8(1):93-97

coat of black/ brown seeded Brassica genotypes contained more fiber and less protein than those of yellow seeded genotypes. Therefore, B. napus lines have been developed from interspecific crosses with related species, namely B.

rapa, B. oleracea spp. alboglabra, B. juncea and B. carinata (Rahman and McVetty, 2011).

One of the most important attributes for introducing sesame grains in the market was seed colour (Pandey et al., 2013; Zhang et al., 2013). Although most are light coloured, there is a wide variability in sesame seed coat colour, which varies from white to black. Due to the importance of this trait for the export market, seed colour is a central target in sesame breeding programs;

however, there are few studies on the inheritance of this essential seed attribute, and determination of genetic factors affecting any trait is necessary to establish useful breeding programs (Laurentin and Benítez, 2014).

Seeds acquire primary dormancy during their development to enhance adaptation, as the capacity of wild species, to diverse environments, by distributing germination over time and space. Domestication tends to reduce dormancy by selection for rapid and uniform germination. Differentiation in seed dormancy between cereal crops and wild relatives has been associated with some factors such as seed morphologies (Guo et al., 2000). For example, the most persistent type of weedy rice is red rice, which is characterized by a red pericarp colour. Red rice has strong seed dormancy (Lim and Ha, 2013).x

In some cases, there is a correlated relationship between seed coat colour and seed quality. Some studies showed that seed lots of red clover visually inspected in terms of seed colour were separated based on a larger colour spectrum thereafter, by digital colour measurement equipment, as seed colour yellow, purple, brown and mixed. Results revealed that seed coat colour of red clover could be preferred as an indicator of seed quality and seedling growth ability. Yellow coloured seeds lots of red clover had higher vigour and seed quality than other colours. Mean germination time (MGT) and electrical conductivity (EC 4 h) test showed significant differences among the seed coat colour. Meanwhile, tests also showed a highly significant correlation in emergence and seedling percentage in salt stress conditions (Atis et al., 2011).

Some seeds are valued according to their appearance, and thus colour is the most important factor for grading (Copeland and McDonald, 2012). The purpose of the current research was to determine the influence of colour features on seed automatic identification.

Materials and Methods Grain samples

Seeds from 75 species of medicinal plants (Table 1) were used for this study.

Seeds were photographed using a Dinolite Digital Microscope model 4050 with 640 × 480 to 1024 × 768 pixel resolution at 30- to 80-times magnification, depending on their original size. A database containing 1,800 images of the 75 species was constructed.

94

Algorithm development

For algorithm development MATLAB 7.9 (Version 2009b) software and windows Vista (Service Pack 1) were used. Employed hardware was an IBM compatible laptop (model Vostro 1500 from DELL Company). In the algorithm, the seed image was segmented from the background image, and its features were extracted and used for the neural network training (Anvarkhah et al., 2012).

Six colour features were extracted by algorithm and applied as network input:

- Mean of red colour of seed surface (R) - Mean of green colour of seed surface (G) - Mean of blue colour of seed surface (B) - Hue means (H)

- Intensity (I)

- Saturation means (Sa)

Different combinations of colour features were used to find out the most accurate combination for seed identification.

Results and Discussion One colour feature

Table 2 shows the total average values of training and test parts of neural network, when using each colour feature individually. The use of hue had the highest accuracy values of training and test with values of 9.239%

and 8.771% respectively. However, employing one colour feature led to a low rate of accuracy values. For example, no accurate identification was shown when using red, blue and saturation features separately (0%). The rest of the colour features tested hereby had low accuracy percents, below 8%.

Two Colour Features

It was noted that two colour features led to a more accurate identification. However, combinations of red, green, blue and saturation with hue caused 0% of accuracy values, while by using other features paired two by two the results had higher values. The most accurate identification was shown within the combination of hue and intensity, which led to 24.184% and 19.298% for training and test parts of neural network respectively (Table 3).

Three colour features

Table 4 shows the training and test accuracy values obtained by using three colour features within the different colour combinations. Except of two combinations ([red + hue + intensity], [blue + hue + saturation], both with 0%), all others had accuracy values of 30-99% for training and 20-85% for test parts of neural network.

Four colour features

Except two combinations of [hue + saturation + red + green] and [hue + saturation + red + blue], all other combinations of four colour features caused above 60%

and 50% for training and test parts of neural network respectively (Table 5). It seems that using triple effect of

(3)

hue, saturation and red with features of green and may cause training errors.

Five colour features

Table 6 shows identification accuracy using five colour features. These combinations led to training and test accuracy values higher than 90% and 75%

respectively, for all features’ combinations.

Six colour features

The most accurate identification was shown using combination of six colour features (99.18% and 87.71%

for training and test of neural network respectively) (Table 7). In general, increasing the number of colour features increased the total

(Anvarkhah et al

Comparison among colour features combination

Fig. 1 shows the average accuracy values of different

Table 1. Scientific name and family of the 75 species of medicinal plan No. Scientific name

1 Allium sativum 2 Amaranthus albus 3 Amaranthus retroflexus 4 Anethum graveolens 5 Foeniculum vulgare 6 Dorema ammoniacum 7 Prangos ferulaceae 8 Achillea millefolium 9 Anthemis tinctoria 10 Calendula officinalis 11 Centaurea cyanus 12 Chrysanthemum superbum 13 Cynara scolymus 14 Pseudohandelia umbellifera 15 Silybum marianum 16 Echinacea 17 Rudbeckia hirta ‘ 18 Silybum marianum 19 Taraxacum sect 20 Borago officinalis 21 Lepidium perfoliatum 22 Lepidium sativum 23 Cannabis sativa 24 Saponaria 25 Erotia ceratoides 26 Kochia prostrata 27 Cucurbita pepo 28 Ricinus communis 29 Fumaria parviflora 30 Hyssopus officinalis 31 Marrubium vulgare 32 Melissa officinalis 33 Melissa axillari 34 Ocimum album 35 Ocimum basilicum 36 Origanum majorana 37 Salvia dorrii 38 Salvia sclarea 39 Satureja hortensis 40 Ziziphora clinopodioides

hue, saturation and red with features of green and may cause training errors.

Five colour features

colour features

for training and test of neural network respectively) (Table 7). In general, increasing the number of colour features increased the total

et al., 2013).

Scientific name and family of the 75 species of medicinal plan Scientific name

Allium sativum Amaranthus albus Amaranthus retroflexus Anethum graveolens Foeniculum vulgare Dorema ammoniacum Prangos ferulaceae Achillea millefolium Anthemis tinctoria Calendula officinalis Centaurea cyanus Chrysanthemum superbum Cynara scolymus Pseudohandelia umbellifera Silybum marianum Echinacea purpurea Rudbeckia hirta ‘Marmalade’

Silybum marianum Taraxacum sect Borago officinalis Lepidium perfoliatum Lepidium sativum Cannabis sativa Saponaria officinalis Erotia ceratoides Kochia prostrata Cucurbita pepo Ricinus communis Fumaria parviflora Hyssopus officinalis Marrubium vulgare Melissa officinalis L.

Melissa axillaris (Benth.) Bakh.f.

Ocimum album Ocimum basilicum Origanum majorana Salvia dorrii Salvia sclarea Satureja hortensis Ziziphora clinopodioides

hue, saturation and red with features of green and

for training and test of neural network respectively) (Table 7). In general, increasing the number of colour features increased the total average of accuracy

Scientific name and family of the 75 species of medicinal plan

Amaranthaceae Amaranthaceae

Chrysanthemum superbum Pseudohandelia umbellifera

Marmalade’

Caryophyllaceae Chenopodiaceae Chenopodiaceae

Euphorbiaceae

(Benth.) Bakh.f.

hue, saturation and red with features of green and

for training and test of neural network respectively) (Table 7). In general, increasing the number of colour average of accuracy

Comparison among colour features combination Fig. 1 shows the average accuracy values of different

Scientific name and family of the 75 species of medicinal plan Family Alliaceae Amaranthaceae Amaranthaceae

Apiaceae Apiaceae Apiaceae Apiaceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Asteraceae Boraginaceae

Brassicaceae Brassicaceae Cannabaceae Caryophyllaceae Chenopodiaceae Chenopodiaceae Cucurbitaceae Euphorbiaceae Fumariaceae

Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae Lamiaceae

hue, saturation and red with features of green and blue

for training and test of neural network respectively) (Table 7). In general, increasing the number of colour average of accuracy

colour combinations. Increasing the number of colour features led to higher accuracy values. Combination of colour features was the most accurate combination with

Scientific name and family of the 75 species of medicinal plant seeds used in the experiment No.

41 42 43 44 45 46

47 Allium ampeloprasum persicum 48

49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75

Fig. 1. Average accuracy values using different colour feature combinations

t seeds used in the experiment Scientific name Astragalus siliquosus Astragalus squarosus Hedysarum scoparium

Onobrychis radiata Onobrychis sp.

Securigera securidaca Allium ampeloprasum persicum

Allium cepa L Allium schoenoprasum L.

Linum usitatissimum Althaea officinalis Malva dendromorpha

Malva sylvestris Caryophyllus aromaticus

Oenothera biennis Sesamum indicum Digitalis purpurea Plantago major

Plantago ovata Plantago purshii Plantago psyllium Agropyron pectiniforme

Avena sativa Bromus kopetdaghensis

Hordeum bulbosum Melica persica Pennisetum orientale

Portulaca oleracea Aquilegia vulgaris Nigella sativa Ruta graveolens Physalis alkekengi Hyoscyamus niger Hyoscyamus pusillus Zygophyllum eurypterum

1. Average accuracy values using different colour feature combinations

Scientific name Astragalus siliquosus Astragalus squarosus Hedysarum scoparium

Onobrychis radiata Onobrychis sp.

Securigera securidaca Allium ampeloprasum persicum

Allium cepa L Allium schoenoprasum L.

Linum usitatissimum Althaea officinalis Malva dendromorpha

Malva sylvestris Caryophyllus aromaticus

Oenothera biennis Sesamum indicum Digitalis purpurea Plantago major

Plantago ovata Plantago purshii Plantago psyllium Agropyron pectiniforme

Avena sativa Bromus kopetdaghensis

Hordeum bulbosum persica

orientale Portulaca oleracea Aquilegia vulgaris Nigella sativa Ruta graveolens Physalis alkekengi Hyoscyamus niger Hyoscyamus pusillus Zygophyllum eurypterum

1. Average accuracy values using different colour feature

Family Leguminoseae Leguminoseae Leguminoseae Leguminoseae Leguminoseae Leguminoseae

Liliaceae Liliaceae Liliaceae Linaceae Malvaceae Malvaceae Malvaceae Myrtaceae Onagraceae

Pedaliacae Plantaginaceae Plantaginaceae Plantaginaceae Plantaginaceae Plantaginaceae

Poaceae Poaceae Poaceae Poaceae Poaceae Poaceae Portulacaceae Ranunculaceae Ranunculaceae

Rutaceae Solanaceae Solanaceae Solanaceae Zygophyllaceae

colour combinations. Increasing the number of colour features led to higher accuracy values. Combination of six colour features was the most accurate combination with

95

Family Leguminoseae Leguminoseae Leguminoseae Leguminoseae Leguminoseae Leguminoseae Liliaceae Liliaceae Liliaceae Linaceae Malvaceae Malvaceae Malvaceae Myrtaceae Onagraceae Pedaliacae Plantaginaceae Plantaginaceae Plantaginaceae Plantaginaceae Plantaginaceae

Poaceae Poaceae Poaceae Poaceae Poaceae Poaceae Portulacaceae Ranunculaceae Ranunculaceae

Rutaceae Solanaceae Solanaceae Solanaceae Zygophyllaceae

colour combinations. Increasing the number of colour six colour features was the most accurate combination with

95

s

(4)

Anvarkhah S et al. / Not Sci Biol, 2016, 8(1):93-97 96

Table 2. Accuracy values using one colour feature

Feature Training accuracy (%) Test accuracy (%)

Green 3.508 4.347

Red 0 0

Blue 0 0

Hue 9.239 8.771

Intensity 5.978 3.508

Saturation 0 0

Average 3.120 2.771

Table 3. Accuracy values using two colour features

Feature Training accuracy (%) Test accuracy (%)

Red +Green 43.070 43.859

Green+Blue 14.402 9.649

Red +Blue 18.750 14.035

Red+Hue 0 0

Red+Intensity 10.597 9.649

Red+Saturation 19.021 16.666

Green+Hue 0 0

Green+Intensity 5.163 4.386

Green+Saturation 20.380 16.666

Blue+Hue 0 0

Blue+Intensity 8.695 7.017

Blue+Saturation 18.206 14.035

Hue+Intensity 24.184 19.298

Hue+Saturation 0 0

Intensity+Saturation 18.750 15.789

Average 13.414 11.403

Table 4. Accuracy values using three colour features Feature

Training accuracy

(%)

Test accuracy

(%)

Red+Green+Blue 52.445 49.122

Red+Green+Hue 92.119 76.315

Red+Green +Intensity 32.065 28.947

Red +Green Saturation 91.032 80.701

Red+Blue+Hue 93.750 78.070

Red+Blue+Intensity 33.695 32.456

Red+Blue+Saturation 65.217 61.403

Green+Blue+Hue 78.532 68.421

Green+Blue+Intensity 40.489 40.350

Green+Blue+Saturation 59.782 52.631

Red+Hue+Intensity 0 0

Red+Hue+Saturation 98.641 84.210

Red+Intensity+Saturation 72.282 66.666

Green+Hue+Intensity 55.434 44.736

Green+Hue+Saturation 97.826 84.210

Green+Intensity+Saturation 66.847 59.649

Blue+Hue+Intensity 83.967 69.298

Blue+Hue+Saturation 0 0

Blue+Intensity+Saturation 32.608 25.438

Hue+Intensity+Saturation 97.826 84.210

Average 62.227 54.341

Table 5. Accuracy values using four colour features Feature

Training accuracy

(%)

Test accuracy

(%)

Red+Green+Blue+Hue 98.641 85.964

Red+Green+Blue+Intensity 64.673 57.894

Red+Green+Blue+Saturation 93.206 82.456

Hue+Intensity+Saturation+Red 99.184 86.842 Hue+Intensity+Saturation+Green 98.369 82.456 Hue+Intensity+Saturation+Blue 98.641 84.210

Hue+Intensity+Red+Green 97.282 83.333

Hue+Intensity+Red+Blue 98.097 85.964

Hue+Intensity+Green+Blue 88.858 78.947

Hue+Saturation+Red+Green 0 0

Hue+Saturation+Red+Blue 0 0

Hue+Saturation+Green+Blue 98.641 83.333

Intensity+Saturation+Red+Green 93.478 82.456 Intensity+Saturation+Red+Blue 80.706 72.807 Intensity+Saturation+Green+Blue 85.869 75.438

Average 79.709 69.473

Table 7. Accuracy values using six colour features Feature

Training accuracy (%)

Test accuracy

(%) Red+Green+Blue+Hue+Intensity+

Saturation 99.184 87.719

Average 99.184 87.719

accuracy values of 99.184% and 87.719% for training and test of neural network. The lowest training and test accuracy values belonged to one colour feature with 3.121% and 2.771%.

In general, increasing the number of colour features increased the total average of accuracy (Anvarkhah et al., 2013).

Colour is one of the most important features in seeds classification and grading. Different seeds and their varieties are identified by their colours. Thomson and Pomeranz

(1991) classified the Western Canadian wheat to six groups using a limited set of colour features (mean Red (R), Green (G) and Blue (B) pixel reflectance features). In general, the red, white and amber coloured wheat types were well separated, while some confusion existed between certain red kernel types. Also, Luo et al. (1999) set an experiment for separation of healthy seeds of Western Canadian wheat from damaged ones using colour features.

Conclusions

Different combinations of colour features (one, two three, four, five and six colour features) were used to find out the most accurate combination for seed identification by machine vision and algorithm determinations. Results showed that the six colour feature was the most accurate combination for seed identification for training and test of neural network respectively, while employing one colour feature led to a low rate of accuracy values, with a lack of accurate identification when using red, blue and saturation features separately.

Table 6. Accuracy values using five colour features Feature

Training accuracy (%)

Test accuracy

(%) Red+Green+Blue+Hue+Intensity 98.641 86.842 Red+Green+Blue+Hue+Saturation 99.184 87.719 Red+Green+Blue+Intensity+Saturation 93.206 78.947 Hue+Intensity+Saturation+Red+Green 99.184 86.842 Hue+Intensity+Saturation+Red+Blue 99.184 86.842 Hue+Intensity+Saturation+Green+Blue 98.913 85.964

Average 98.052 85.526

(5)

References

Atiş I, Atak M, Can E, Mavi K (2011). Seed coat color effects on seed quality and salt tolerance of red clover (Trifolium pratense).

International Journal of Agriculture and Biology 13(3):363-368.

Anvarkhah S, Hajeh-Hosseini M, Davari-Edalat-Panah A, Rashedmohassel MH (2013). Medicinal plant seed identification using machine vision. Seed Science and Technology 41(1):107- 120.

Anvarkhah S, Khajeh-Hosseini M, Davari-Edalat-Panah A (2012).

Seed identification of ten rangeland species based on machine learning using combination of RBF and Feed Forward neural networks. International Journal of Agriculture and Crop Sciences 4(14):993-1004.

Cope JS, Corney D, Clark JY, Remagnino P, Wilkin P (2012). Plant species identification using digital morphometrics: A review.

Expert Systems with Applications 39(8):7562-7573.

Copeland LO, McDonald M (2012). Principles of seed science and technology. Springer Science & Business Media.

Ducournau S, Feutry A, Plainchault P, Revollon P, Vigouroux B, Wagner MH (2004). An image acquisition system for automated monitoring of germination rate of sunflower seeds. Computer and Electronics in Agriculture 44:189-202.

Granitto PM, Navone HD, Verdes PF, Ceccatto HA (2002). Weed seeds identification by machine vision. Computer and Electronics in Agriculture 33:91-103.

Guo Q, Brown JH, Valone TJ, Kachman SD (2000). Constraints of seed size on plant distribution and abundance. Ecology 81(8):2149-2155.

Jansen PI (1995). Seed production quality in Trifolium balansae and T. resupinatum: The role of seed colour. Seed Science and Technology 23:353-364.

Laurentin H, Benítez T (2014). Inheritance of seed coat color in sesame. Pesquisa Agropecuria Brasiliera 49(4):290-295.

Lev‐Yadun S, Ne’eman G (2013). Bimodal colour pattern of individual Pinus halepensis Mill. seeds: a new type of crypsis.

Biological Journal of the Linnean Society 109(2):271-278.

Lim SH, Ha SH (2013). Marker development for the identification of rice seed color. Plant Biotechnology Reports 7(3):391-398.

Liu ZY, Cheng F, Ying YB, Rao XQ (2005). Identification of rice seed varieties using neural network. Journal of Zhejiang University- Science B 6(11):1095-1100.

Luo X, Jayas DS, Symons SJ (1999). Identification of damaged kernels in wheat using a colour machine vision system. Journal of Cereal Science 30:49-59.

Paliwal J, Visen NS, Jayas DS, White NDG (2003). Cereal grain and dockage identification using machine vision. Biosystems Engineering 85(1):51-57.

Pandey SK, Das A, Dasgupta T (2013). Genetics of seed coat color in sesame (Sesamum indicum L.). African Journal of Biotechnology 12(42):6061.

Rahman M, McVetty PBE (2011). A review of Brassica seed color.

Canadian Jounal of Plant Sciences 91(3):437-446.

Thomson WH, Pomeranz Y (1991). Classification of wheat kernels using three dimensional image analysis. Cereal Chemistry 68:357- 361.

Ureña R, Rodríguez F, Berenguel M (2001). A machine vision system for seeds quality evaluation using fuzzy logic. Computer and Electronics in Agriculture 32(1):1-20.

Zhang H, Miao H, Wei L, Li C, Zhao R, Wang C (2013). Genetic analysis and QTL mapping of seed coat color in sesame (Sesamum indicum L.). PLoS One 8(5):p.e63898.

Zhang X, Liu F, He Y, Li X (2012). Application of hyperspectral imaging and chemometric calibrations for variety discrimination of maize seeds. Sensors 12(12):17234-17246.