Classification of Wheat Grains Using Machine Algorithms
2 likes514 views
The paper discusses the classification of wheat grains in India using machine learning algorithms, specifically support vector machine (SVM) and neural network (NN), to automate the time-consuming manual quality assessment. After capturing images of wheat grains, features are extracted and classified, achieving an accuracy of 86.8% for SVM and 94.5% for NN, indicating that NN outperforms SVM. The study suggests future improvements using different machine algorithms and feature sets for better accuracy.
![International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064
Volume 2 Issue 8, August 2013
www.ijsr.net
Classification of Wheat Grains Using Machine
Algorithms
Meesha Punn1
, Nidhi Bhalla2
1, 2
Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India
Abstract: India is the second largest producer of wheat in the world after China. Specifying the quality of wheat manually requires an
expert judgment and is time consuming. To overcome this problem, machine algorithms can be used to classify wheat according to its
quality. In this paper we have done wheat classification by using two machine learning algorithms, that is, Support Vector Machine
(OVR) and Neural Network (LM). For classification, images of wheat grain are captured using digital camera and thresholding is
performed. Following this step, features of wheat are extracted from these images and machine learning algorithms are implemented.
The accuracy of Support Vector Machine is 86.8% and of Neural network is 94.5%. Results show that Neural Network (LM)is better
than Support Vector Machine(OVR).
Keywords: Classification, Computer Vision System, Image Processing, Grading, Quality, SVM, Wheat
1. Introduction
India is the second largest producer of wheat in the world
after China. Determining the quality of wheat is critical.
Specifying the quality of wheat manually requires an expert
judgment and is time consuming. Sometimes the variety of
wheat looks so similar that differentiating them becomes a
very tedious task when carried out manually. To overcome
this problem, machine algorithms can be used to classify
wheat according to its quality.
Machine vision is widely used in the field of agriculture for
identifying the varieties of various food crops and for
identifying their quality as well. A machine vision system
(MVS) provides an alternative to the manual inspection of
biological products. It makes use of Machine algorithms and
digital images. These images are obtained with the help of
digital camera and are then stored in the computer for future
work. In a Machine Vision System, the camera acts as an
eye and the computer acts as the brain.
Digital images stored in the computer are processed by
Image processing algorithms. They extract the features from
the digital images and use them to generate pattern. These
patterns are input to the machine algorithms based on which
the objects are classified into their respective classes. Such
machine algorithms used for classifying objects are referred
to as pattern classifiers.
The present paper deals with classifying the wheat grain. For
doing this, two machine learning algorithms are used, that is,
Support Vector Machine (OVR) and Neural Network (LM).
Figure 1: Components of Machine Vision System
2. Previous Work
Studies have been done on classification of wheat using
machine algorithms and image processing. These studies
have used different machine classifiers and have performed
feature extraction for carrying out their work. Some studies
extracted two classes of features while others extracted more
than two classes of features.
In a study [1], four varieties of wheat were identified by
integrating machine vision and artificial neural network
using Matlab software. Different colour and morphological
features of wheat were extracted for carrying out the process
of identification. Given these features, testing of ANN was
performed. It was concluded that colour features or
morphological features alone could not recognize wheat so a
combination of both was used. The overall accuracy was
found to be 95.86%.
Another study [2] classified the rain fed wheat grain
cultivars using Artificial Neural Network. For this purpose,
colour features, morphological features and textural features
were extracted. These features were fed to multilayer
perceptron neural network. The classification accuracy was
concluded to be 86.58%. After using UTA algorithm for
feature extraction, the accuracy was increased to 87.22%.
A study by [3] used Discriminant Analysis and K Nearest
Neighbor for classification of wheat and barley grain
kernels. The system training was performed with only
morphological features, only colour features and
combination of morphological features, colour features as
well as textural features. It was concluded that accuracies
higher than 99% can be achieved when morphological,
colour and textural feature types are used together as
compared to using them alone.
3. Methodology of Proposed Work
The methodology comprises of different steps which are as
follows:
363](https://image.slidesharecdn.com/mtiwmtmxmtg-131002084338-phpapp01/85/Classification-of-Wheat-Grains-Using-Machine-Algorithms-1-320.jpg)
![International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064
Volume 2 Issue 8, August 2013
www.ijsr.net
Figure 2: Methodology of wheat classification
3.1 Obtain dataset of wheat grain images
The first step for classification of wheat is to acquire images
of wheat grains. In order to obtain the image of wheat
grains, the wheat grains were placed on a black background
and special precaution was taken so that no light gets
reflected on the black background. This is so in order to
obtain clear image of wheat grains. Digital camera was used
to obtain the images which were then stored in the computer
machine.
Figure1: Captured image of Wheat grains
3.2. Feature extraction and Thresholding
The next step in wheat classification is performing
thresholding of images. Thresholding [7] is the simplest
method of image segmentation. From a grayscale image,
thesholding can be used to create binary images by turning
all pixels below some threshold to zero and all pixels about
that threshold to one. Feature extraction [5] defines a set of
features which represents the information that is important
for classification of wheat. Different features were extracted
and they were fed to machine classifiers for wheat
classification. The following features were extracted from
wheat images:
Table 1: Features extracted from images of wheat Features
Formula
Features Formula
Area
Using offset method:
First measure the length of the longest axis of
the area (line AB). This is called the length
line. Next, divide the length line into equal
sections, for example 10 ft. These lines are
called offset lines.
Length line(AB) = 60 pixel, distance between
offset lines is 10 pixel apart
Length of each offset line;
C = 15 pixel, D = 10 pixel, E=15 pixel, F = 25
pixel, G =20 pixel
Total length of offset lines = C + D + E + F +
G = 15 + 10 + 15 + 25 + 20 = 85 pixel
Area to measure = Distance between offset
lines x sum of the length of the offset lines =
10 pixel x 85 pixel
= 850 pixel
Perimeter Number of horizontal and vertical
links+(Number of diagonal links*√2)
Volume
volume=1/ 2[x1(y2 - y3) + x2(y3- y1) + x3(y1-
y2_](z1+ z2 + z3) / 3
Where, (x1, y1), (x2, y2) and (x3, y3) are the
planar co-ordinates of the triangle vertices and
z1, z2 and z3 are the corresponding elevation
Roundness Ratio 4*pi*area/perimeter^2
Percentage of
broken grain
No. of broken grain/no. of whole grain*100
Table 2: AGMART STANDARDS [10] for grading wheat
grains
Grade
Designati
on
Foreign
matter
(% by wt.)
Damaged
grains
(% by
wt.)
Slightly
damaged grains
(% by wt.)
Immature
shrivelled and
Broken Grains
Weevilled
Grains
(% by wt.)
I 1 1 2 2 1
II 1 2 4 4 3
III 1 4 6 10 6
IV 1 5 10 10 10
3.3. Grading and labeling each grain feature row
The grading of wheat grain is done on the basis of
Agricultural Procedure (Grading and Marking) Act, 1937
AGMART STANDARDS. According to this act, wheat
grains are divided into four classes.
364](https://image.slidesharecdn.com/mtiwmtmxmtg-131002084338-phpapp01/85/Classification-of-Wheat-Grains-Using-Machine-Algorithms-2-320.jpg)
![International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064
Volume 2 Issue 8, August 2013
www.ijsr.net
Table 3: Criteria for finding grade of wheat grains
S.No Features Criteria Formula
1 Damaged
grain
Roundness
ratio≈ 0.98
No. of damaged grain
-----------------------------------------------
*100
total no. of grain in sample
2 Foreign
matter
Pixel value < 5
and volume >
50
No. of particles with pixel value less than
5------------------------------ *100
total no. of grain in sample
3 Slightly
damaged
grains
Roundness
Ratio ≈ 0.8
No. of slightly damaged grain-------------
------*100total no. of grain in sample
4 Broken
grains
Length < ¾ of
full length
No. of broken grain--------------------------
----------- *100 total no. of grain in
sample
5 Weevilled
grains
Volume < 15 No. of grains with volume less than ------
-------------------------------------------------
--- *100 total no. of grain in sample
3.4. Implementing Machine Algorithms [6]
The two machine learning algorithms, Neural Network (LM)
and Support Vector Machine (OVR), are implemented
separately.
Neural Network (NN) [8] has been the successfully used
classifier in numerous fields. So, it is of interest to use it for
wheat grain classification. It can be modeled on a human
brain. The basic processing unit of brain is neuron which
works identically in NN.
Support vector machine (SVM) [9] is a computer algorithm
that learns by example to assign labels to objects. It
maximizes a particular mathematical function with respect
to a given collection of data.
In order to implement the algorithms, the numeric
parameters are calculated to access quality of each grain.
These parameters are fed to each algorithm separately and
them the performance is evaluated.
3.5. Evaluate Performance
After implementing the algorithms, their performance is
measured to see whether desired results are achieved or not.
4. Results
After implementing the algorithms and the results of their
evaluated performance were obtained. The results show that
the accuracy of Support Vector Machine (OVR) comes out
to be 86.8% and that of Neural Network (LM) comes out to
be 94.5%. This means that Neural Network (LM) Performed
better than Support vector Machine (OVR).
Figure 3: Overall accuracy of NN (LM) and SVM (OVR)
5. Conclusion
Our objective was to classify wheat grains effectively using
machine algorithms. For this we used two machine
algorithms, that is, Support Vector Machine and Neural
Network.
The accuracy of SVM is 86.8% and Of Neural Network is
94.5%. Based on these results it is concluded that this
algorithm performed better than that algorithm. The
accuracy of this algorithm emerged out to be more than that
algorithm.
6. Future Scope
This work can be carried forward by using machine
algorithms, which have better accuracy and has less
computational cost, other than used in this research work.
Moreover, research can be done on different variety of
wheat and different feature set can be used.
References
[1] Arefi, A. Modarres Motlagh, and R. Farrokhi
Teimourlou, “Wheat class identification using
computer vision system and artificial neural networks”,
2011.
[2] Alireza Pazoki and Zohreh Pazoki, “Classification
system for rain fed wheat grain cultivars using
artificial neural network”, 2011.
[3] F. Guevara-Hernandez and J. Gomez-Gil, “A machine
vision system for classification of wheat and barley
grain kernels” 2011.
[4] Sanjivani Shantaiya and Mrs. Uzma Ansari,
“Identification Of Food Grains And Its Quality Using
Pattern Classification”, 2010
[5] W. Philpot, “Feature Extraction”
[6] Francisco Pereira a, Tom Mitchell b, Matthew
Botvinick a, “Machine learning classifiers and fMRI:
A tutorial overview”, 2009
[7] https://www.comp.glam.ac.uk/pages/staff/getaylor/pap
ers/gim98.PDF
[8] Vincent Cheung, Kevin Cannons, “An introduction to
Neural Networks”, 2002.
[9] William S Noble, “What is Support Vector Machine”,
2006.
[10] http://agmarknet.nic.in/Wheat_manual.html
365](https://image.slidesharecdn.com/mtiwmtmxmtg-131002084338-phpapp01/85/Classification-of-Wheat-Grains-Using-Machine-Algorithms-3-320.jpg)
Classification of Wheat Grains Using Machine Algorithms
- 1. International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064 Volume 2 Issue 8, August 2013 www.ijsr.net Classification of Wheat Grains Using Machine Algorithms Meesha Punn1 , Nidhi Bhalla2 1, 2 Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India Abstract: India is the second largest producer of wheat in the world after China. Specifying the quality of wheat manually requires an expert judgment and is time consuming. To overcome this problem, machine algorithms can be used to classify wheat according to its quality. In this paper we have done wheat classification by using two machine learning algorithms, that is, Support Vector Machine (OVR) and Neural Network (LM). For classification, images of wheat grain are captured using digital camera and thresholding is performed. Following this step, features of wheat are extracted from these images and machine learning algorithms are implemented. The accuracy of Support Vector Machine is 86.8% and of Neural network is 94.5%. Results show that Neural Network (LM)is better than Support Vector Machine(OVR). Keywords: Classification, Computer Vision System, Image Processing, Grading, Quality, SVM, Wheat 1. Introduction India is the second largest producer of wheat in the world after China. Determining the quality of wheat is critical. Specifying the quality of wheat manually requires an expert judgment and is time consuming. Sometimes the variety of wheat looks so similar that differentiating them becomes a very tedious task when carried out manually. To overcome this problem, machine algorithms can be used to classify wheat according to its quality. Machine vision is widely used in the field of agriculture for identifying the varieties of various food crops and for identifying their quality as well. A machine vision system (MVS) provides an alternative to the manual inspection of biological products. It makes use of Machine algorithms and digital images. These images are obtained with the help of digital camera and are then stored in the computer for future work. In a Machine Vision System, the camera acts as an eye and the computer acts as the brain. Digital images stored in the computer are processed by Image processing algorithms. They extract the features from the digital images and use them to generate pattern. These patterns are input to the machine algorithms based on which the objects are classified into their respective classes. Such machine algorithms used for classifying objects are referred to as pattern classifiers. The present paper deals with classifying the wheat grain. For doing this, two machine learning algorithms are used, that is, Support Vector Machine (OVR) and Neural Network (LM). Figure 1: Components of Machine Vision System 2. Previous Work Studies have been done on classification of wheat using machine algorithms and image processing. These studies have used different machine classifiers and have performed feature extraction for carrying out their work. Some studies extracted two classes of features while others extracted more than two classes of features. In a study [1], four varieties of wheat were identified by integrating machine vision and artificial neural network using Matlab software. Different colour and morphological features of wheat were extracted for carrying out the process of identification. Given these features, testing of ANN was performed. It was concluded that colour features or morphological features alone could not recognize wheat so a combination of both was used. The overall accuracy was found to be 95.86%. Another study [2] classified the rain fed wheat grain cultivars using Artificial Neural Network. For this purpose, colour features, morphological features and textural features were extracted. These features were fed to multilayer perceptron neural network. The classification accuracy was concluded to be 86.58%. After using UTA algorithm for feature extraction, the accuracy was increased to 87.22%. A study by [3] used Discriminant Analysis and K Nearest Neighbor for classification of wheat and barley grain kernels. The system training was performed with only morphological features, only colour features and combination of morphological features, colour features as well as textural features. It was concluded that accuracies higher than 99% can be achieved when morphological, colour and textural feature types are used together as compared to using them alone. 3. Methodology of Proposed Work The methodology comprises of different steps which are as follows: 363
- 2. International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064 Volume 2 Issue 8, August 2013 www.ijsr.net Figure 2: Methodology of wheat classification 3.1 Obtain dataset of wheat grain images The first step for classification of wheat is to acquire images of wheat grains. In order to obtain the image of wheat grains, the wheat grains were placed on a black background and special precaution was taken so that no light gets reflected on the black background. This is so in order to obtain clear image of wheat grains. Digital camera was used to obtain the images which were then stored in the computer machine. Figure1: Captured image of Wheat grains 3.2. Feature extraction and Thresholding The next step in wheat classification is performing thresholding of images. Thresholding [7] is the simplest method of image segmentation. From a grayscale image, thesholding can be used to create binary images by turning all pixels below some threshold to zero and all pixels about that threshold to one. Feature extraction [5] defines a set of features which represents the information that is important for classification of wheat. Different features were extracted and they were fed to machine classifiers for wheat classification. The following features were extracted from wheat images: Table 1: Features extracted from images of wheat Features Formula Features Formula Area Using offset method: First measure the length of the longest axis of the area (line AB). This is called the length line. Next, divide the length line into equal sections, for example 10 ft. These lines are called offset lines. Length line(AB) = 60 pixel, distance between offset lines is 10 pixel apart Length of each offset line; C = 15 pixel, D = 10 pixel, E=15 pixel, F = 25 pixel, G =20 pixel Total length of offset lines = C + D + E + F + G = 15 + 10 + 15 + 25 + 20 = 85 pixel Area to measure = Distance between offset lines x sum of the length of the offset lines = 10 pixel x 85 pixel = 850 pixel Perimeter Number of horizontal and vertical links+(Number of diagonal links*√2) Volume volume=1/ 2[x1(y2 - y3) + x2(y3- y1) + x3(y1- y2_](z1+ z2 + z3) / 3 Where, (x1, y1), (x2, y2) and (x3, y3) are the planar co-ordinates of the triangle vertices and z1, z2 and z3 are the corresponding elevation Roundness Ratio 4*pi*area/perimeter^2 Percentage of broken grain No. of broken grain/no. of whole grain*100 Table 2: AGMART STANDARDS [10] for grading wheat grains Grade Designati on Foreign matter (% by wt.) Damaged grains (% by wt.) Slightly damaged grains (% by wt.) Immature shrivelled and Broken Grains Weevilled Grains (% by wt.) I 1 1 2 2 1 II 1 2 4 4 3 III 1 4 6 10 6 IV 1 5 10 10 10 3.3. Grading and labeling each grain feature row The grading of wheat grain is done on the basis of Agricultural Procedure (Grading and Marking) Act, 1937 AGMART STANDARDS. According to this act, wheat grains are divided into four classes. 364
- 3. International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064 Volume 2 Issue 8, August 2013 www.ijsr.net Table 3: Criteria for finding grade of wheat grains S.No Features Criteria Formula 1 Damaged grain Roundness ratio≈ 0.98 No. of damaged grain ----------------------------------------------- *100 total no. of grain in sample 2 Foreign matter Pixel value < 5 and volume > 50 No. of particles with pixel value less than 5------------------------------ *100 total no. of grain in sample 3 Slightly damaged grains Roundness Ratio ≈ 0.8 No. of slightly damaged grain------------- ------*100total no. of grain in sample 4 Broken grains Length < ¾ of full length No. of broken grain-------------------------- ----------- *100 total no. of grain in sample 5 Weevilled grains Volume < 15 No. of grains with volume less than ------ ------------------------------------------------- --- *100 total no. of grain in sample 3.4. Implementing Machine Algorithms [6] The two machine learning algorithms, Neural Network (LM) and Support Vector Machine (OVR), are implemented separately. Neural Network (NN) [8] has been the successfully used classifier in numerous fields. So, it is of interest to use it for wheat grain classification. It can be modeled on a human brain. The basic processing unit of brain is neuron which works identically in NN. Support vector machine (SVM) [9] is a computer algorithm that learns by example to assign labels to objects. It maximizes a particular mathematical function with respect to a given collection of data. In order to implement the algorithms, the numeric parameters are calculated to access quality of each grain. These parameters are fed to each algorithm separately and them the performance is evaluated. 3.5. Evaluate Performance After implementing the algorithms, their performance is measured to see whether desired results are achieved or not. 4. Results After implementing the algorithms and the results of their evaluated performance were obtained. The results show that the accuracy of Support Vector Machine (OVR) comes out to be 86.8% and that of Neural Network (LM) comes out to be 94.5%. This means that Neural Network (LM) Performed better than Support vector Machine (OVR). Figure 3: Overall accuracy of NN (LM) and SVM (OVR) 5. Conclusion Our objective was to classify wheat grains effectively using machine algorithms. For this we used two machine algorithms, that is, Support Vector Machine and Neural Network. The accuracy of SVM is 86.8% and Of Neural Network is 94.5%. Based on these results it is concluded that this algorithm performed better than that algorithm. The accuracy of this algorithm emerged out to be more than that algorithm. 6. Future Scope This work can be carried forward by using machine algorithms, which have better accuracy and has less computational cost, other than used in this research work. Moreover, research can be done on different variety of wheat and different feature set can be used. References [1] Arefi, A. Modarres Motlagh, and R. Farrokhi Teimourlou, “Wheat class identification using computer vision system and artificial neural networks”, 2011. [2] Alireza Pazoki and Zohreh Pazoki, “Classification system for rain fed wheat grain cultivars using artificial neural network”, 2011. [3] F. Guevara-Hernandez and J. Gomez-Gil, “A machine vision system for classification of wheat and barley grain kernels” 2011. [4] Sanjivani Shantaiya and Mrs. Uzma Ansari, “Identification Of Food Grains And Its Quality Using Pattern Classification”, 2010 [5] W. Philpot, “Feature Extraction” [6] Francisco Pereira a, Tom Mitchell b, Matthew Botvinick a, “Machine learning classifiers and fMRI: A tutorial overview”, 2009 [7] https://www.comp.glam.ac.uk/pages/staff/getaylor/pap ers/gim98.PDF [8] Vincent Cheung, Kevin Cannons, “An introduction to Neural Networks”, 2002. [9] William S Noble, “What is Support Vector Machine”, 2006. [10] http://agmarknet.nic.in/Wheat_manual.html 365
- 4. International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064 Volume 2 Issue 8, August 2013 www.ijsr.net Authors Profile Meesha Punn is pursuing M.Tech (Computer Science) form Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India. She did her B.Tech (Information Technology) form Doaba Institute of Engineering and Technology, Kharar, Punjab, India. Nidhi Bhalla is working as Assistant Professor in Swami Vivekanand Institute of Engineering and Technology, Banur, Punjab, India. She did her B.Tech (Information Technology) and M.Tech (Computer Science) form Lovely Professional University, Jalandhar, Punjab, India. 366