Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

Recognition of Brain Tumors Using Radon Transform

and Gray Level Co-Occurrence Matrix Algorithm
Nashwan Amin Al-khulaidi1 Mohammed Hashem Almourish2*
[email protected] [email protected]

Mohammed Essam Ali3 Abobakr Ali Ghalib3

[email protected] [email protected]

Omar Abdulaziz Alnaqeeb3 Wegdan Mohammed Saeed3

[email protected] [email protected]

Abstract: The field of recognizing brain tumors in recent years has received wide attention by
researchers due to the rapid spread of the disease. In this paper, brain tumors (normal, glioma and
meningioma) were recognized by the proposed networks (Support Vector Machine (SVM),
Multilayer Perceptron Neural Network (MLPNN) and Radial Basis Function neural network
(RBFNN)). The dataset is divided into two parts: 450 MRI images for the training phase and 240
for the testing phase. Pre-processing was applied to the input images by converting the images to
grayscale and removing noise on them by median filter. Then, the texture features of MRI images
were extracted using Gray Level Co-Occurrence Matrix algorithm after radon transformation. The
extracted texture features were passed to the proposed networks for classification and recognition
of tumors. The technique of extracting brain tumors from the magnetic resonance image was added
by segmentation algorithm and morphology operations on the images. The performance of (SVM,
MLPNN and RBFNN) was measured based on sensitivity, specificity, and accuracy and the best
results obtained by SVM with a quadratic kernel function were 98.45, 98.97, and 98.71 for
sensitivity, specificity, and accuracy, respectively.
Keywords: multilayer perceptron neural network, Support vector machine, radial basis function
neural network, gray level co-occurrence matrix, Radon transform, magnetic resonance imaging.

1
Dr., Department of Robotics & Mechatronics, Faculty of Engineering and Information Technology, Taiz University, Taiz,
Yemen
2
Prof. Dr, Department of Communication & Computer Engineering, Faculty of Engineering and Information Technology,
Taiz, University, Taiz, Yemen
3
Engineer, Department of Software Engineering, Faculty of Engineering and Information Technology, Taiz University,
Taiz, Yemen
* Corresponding Author
 Nashwan A. Al-khulaidi et al.

1. Introduction
Recently, cancers have become one of the complex medical problems that need
great attention by researchers and link them to the field of computer. One of these
diseases is brain tumor. Brain tumor is one of the most complex diseases,
knowing that there are many studies that have made progress in finding solutions
to this disease. The number of brain tumors in the Americas is estimated at 43,800
per year, and the number of malignant tumors is 18,500 [1]. Brain cancer can be
defined as an abnormal growth of cells inside the brain or skull [2]. Tumors can
also be classified into benign tumors and malignant tumors. Malignant tumors can
be classified into primary or secondary tumors. The primary tumor consists of the
cells of that organ in which the tumor is located. The secondary tumor consists of
cells belonging to different other parts of the body [3]. There are several types of
brain tumors as shown in figure 1, the most common being glioma and
meningioma [4]. To diagnose brain cancer, it is done with the help of a computer
by analyzing CT and MRI images, which give comprehensive information about
the nature of the tumor.

a) Glioma b) Meningioma
Figure 1. Brain tumors

One of the main problems in the research was to obtain high accuracy in the
analysis of the images and extract the important features and then classify them.
them. It is known that classical techniques take longer time to identify images of
cancerous diseases and require a large memory to save the extracted features and
also the accuracy of recognition is weak, so it was necessary to overcome this
problem by using modern algorithms and techniques to analyze and classify
images. Recently, modern technologies such as artificial intelligence have been
used to identify brain cancers, due to the high accuracy of disease recognition. So
it was necessary to have algorithms that extract important features without
complexity, as well as use artificial intelligence networks to obtain high
63
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

classification accuracy. In this paper, brain tumors (normal, glioma, meningioma)

were identified by proposing the GLCM algorithm to extract the texture features
of magnetic resonance images from the radon transform, and after obtaining the
important features, they were passed to the proposed networks (SVM, MLPNN
and RBFNN) for feature classification and tumor image recognition.
The importance of using the Radon transform is that it reconstructs the original
image of the MRI of the brain through the reverse Radon transform and from
several projections from different angles, where the reconstructed image is
characterized by high computational efficiency and high image quality. While the
importance of the GLCM algorithm is its ability to classify pixels in comparison
to neighboring pixels in terms of value, color, density, direction, and
homogeneity, so it has the ability to identify brain tumor images, because the
tumor image has a density that differs from the density of the rest of the brain
image, so there will be a change in homogeneity with the rest of the images. It
will be easier to use GLCM to extract the histological characteristics of the brain.
The results of the proposed networks showed varying accuracy, the best being
SVM. This paper is organized as follows: In the second section, the relevant
literature on tumor recognition is presented. In the third section, the methodology
for recognizing tumors was clarified. The results of the proposed methods were
listed in the fourth section. Finally, the conclusion of the paper is presented in the
fifth section.
2. Related Works
We provide an overview of the literature related to the recognition of brain
tumors. There are many studies in this field and we will present some studies
related to the algorithms used in the research. Reference [5] proposed a GA
algorithm to collect texture features from the gray level system and then pass
them to the SVM to recognize brain tumors and it has achieved good results.
Reference [6] proposed a fuzzy c-means algorithm to segment images of cancer
tumors and achieved high accuracy and quality in image recognition. while
reference [7] proposed and compared three methods of segmenting MRI images
of brain tumors and found that the modified region gradient magnitude growth
technique yielded the best results. The authors of [8] presented a method for
detecting tumors using image segmentation and histogram thresholding.
Reference [9] proposed a method for recognizing cancer cells by morphological
segmentation and achieved good results. In [10] a method classifier was used to
identify tumors using an SVM classifier, the images were optimized by changing
the threshold value using the sigmoid function, and the features were extracted
using GLCM passed to the SVM classifier. The accuracy of the proposed method
was 96.51%. Reference [11] presented a detailed study of modern methods and
techniques used in segmenting MRI images of brain cancer. In the study [12]
 Nashwan A. Al-khulaidi et al.

GLCM and Convolutional Neural Network are used to obtain a strong image
texture analysis and a high network depth in image data processing classification.
In the reference [13], a feature extraction method was proposed using GLCM for
brain tumor detection, and the features were passed to a KSVM classifier to
classify them into four different classes. In the study [14], tumors were recognized
using the tumor image segmentation algorithm, then image features were
extracted by GLCM algorithm, and finally, they were passed to PCA to classify
tumor images into five classes, and this method achieved an accuracy of 95%.
The authors of [15] suggested using neural networks to identify tumors and
classify them into benign tumor or benign tumor, and image features were
extracted by GLCM algorithm. The results showed an accuracy of 93.33%.
Reference [16] proposed a method for identifying tumors using a probabilistic
neural network, where the features of the images were extracted using discrete
wavelet shift with GLCM algorithm followed by morphological processes.
In the study [17], a program for diagnosing and identifying brain cancer using
ANN was planned and developed. The surface features of tumors were extracted
using GLCM algorithm and a mysterious neurological concept was finally
established. The authors of the reference [18] proposed a recurrent neural network
to identify brain tumors (pituitary, gliomas and meningioma) and the results
showed that the proposed recurrent neural network effectively classified brain
tumors with an accuracy of 98%. Reference [19] proposed a convolutional neural
network to recognize tumors, and the features of brain tumor images were
extracted using discrete wavelet transform technology and GLCM, and the results
showed an accuracy of 91%. This paper presents the identification of three brain
tumors (normal, glioma and meningioma) using three proposed networks (SVM,
MLPNN and RBFNN) after extracting texture features for magnetic resonance
images using GLCM algorithm after Radon transformation.

3. Methodology
In this section, we will present the method of recognizing brain tumors (normal,
glioma and meningioma). The proposed method is illustrated in the following
steps: data collection, pre-processing, Radon transformation for MRI images,
extracting features using GLCM algorithm, classification of extracted features and
recognition of tumors using three proposed networks (MLPNN, RBFNN, and
SVM), Finally, the brain tumor is cut from the MRI images of the brain by
segmentation algorithm as shown in Figure 2.

65
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

3.1 Data Collection

In this phase, 690 MRI brain images were collected and divided into 450 MRIs
for training the system (150 MRIs for normal brain and 300 images for two types
of abnormal brain (meningioma and glioma)). The testing phase was assigned 240
MRI images.

3.2 Pre-processing
Pre-processing plays a very important role in this system. Median filter is used to
remove the noise and unwanted objects from MRI brain image. Color images are
converted to grayscale images.

Figure 2. Methodology for brain tumors recognition.

3.3 Feature Extraction

As maintained in the introduction, an important task is extracting the most
important features from the images. Features extraction step was done by several
stages. The first stage is applying Radon transformation to the MRI image as
shown in figure 3.a. In order to select the best projections, we have computed the
Radon transform of the MRI images at different projections, then we reconstruct
 Nashwan A. Al-khulaidi et al.

MRI images by the inverse Radon transform figure 3.b. The best projections are
when the reconstruct MRI images (with the inverse Radon transform) is closer to
the original image. Reconstructing MRI image using 180 projections gives the
closer MRI image to the original image as can be seen in figure 3.

a) Applying Radon b) Invers Radon

transform
Figure 3. Radon Transformation
After applying Radon transformation, we remove all zeros in order to produce a
suitable image with good features. Finally, Gray Level Co-Occurrence Matrix
(GLCM) is used to extracted texture features from the Radon transformed brain
MRI image. In this paper nine important features are extracted as shown in table
1.
Table 1. GLCM texture feature.
No. Features Description Formula
The energy is uniformity
1 Energy (EN) measurement between the
pixels. Range= [0,1].
Contrast is difference
measurement in
Contrast
2 luminance for making
(CO)
object distinguishable.
Range = [0,1]
Correlation is the relation
Correlation measurement between the
3
(Cor) neighbor pixels. Range =
[-1,1].
The homogeneity
Homogeneity measures of closeness of
4
(HOM) the element distribution in
GLCM to GLCM
67
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

diagonals. Range = [0,1]

The mean of an image is

calculated by summing of
5 Mean (M) all the pixel values of an
image divided by the total
number of pixels.
The Standard describes
Standard
the possibility distribution
6 Deviation
of an observed
(SD)
population.
Entropy is calculated to
characterize the
7 Entropy (E)
randomness of the
textural image.
Skewness is a measure of
Skewness
8 symmetry or the lack of
(SK)
symmetry.
The shape of a random
variable probability
Kurtosis
9 distribution is described
(Kurt)
by the parameter called
Kurtosis.

3.4 CLASSIFICATION
As mentioned in the introduction, SVM, MLPNN and RBFNN were used for
classification of brain tumors in MRI images. Brief explanation of these
algorithms is described in the following sections.
1. Support Vector Machine: SVM is utilized which is powerful supervised
machine learning techniques for classification and regression as shown in Figure
4. Simply stated, the SVM identifies the best separating hyper plane between the
two classes of the training samples within the feature space by focusing on the
training cases placed at the edge of the class descriptors.
SVM have several kernel functions. In this work we used three different kernel
functions (linear. radial basis and quadratic kernel function).
 Nashwan A. Al-khulaidi et al.

Figure 4. Architecture of support vector machine for pattern

classification

2. Multilayer Perceptron Neural Network

A multilayer perceptron neural network is feed-forward neural network (Figure 5)
which consists of at least three layers of nodes: an input layer, at least one hidden
layer and an output layer [20,21]. MLP Networks are sensitive to the number of
neurons in their hidden layers. In this paper, the optimum number of neurons in
the hidden layer of MLPNN was investigated from 1 to 9 neurons. Increasing
number of neurons in the hidden layer more than 9 neurons the performance of the
brain tumors classification takes more time for training the dataset and
performance do not increase.

Figure 5. Architecture of multilayer perceptron neural network

69
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

3. Radial Basis Function Neural Network

Radial basis function neural network is also a feed-forward network with three
different layers: an input layer, a hidden layer and an output layer [22, 23]. Figure
6 shows the structure of the Radial basis function neural network.
The most important part in training RBFNN is determining a spread parameter.
The way that the optimum number of this parameter was found is by investigating
different RBFNN with various values of spread parameter. The performances of
these RBFNN are computed and the value of spread parameter that gives the best
performance was selected.

Figure 6. Structure of an RBF network

3.5 SEGMENTATION
In this part, we will explain the process of brain tumor recognition and diagnosis
by segmentation algorithm. First we apply median filter to the gray-scale image
and then use edge detection to define the boundaries of the brain MRI image using
the Sobel gradient filter and add the result of the Sobel filter to the median filter to
get a sharper image as shown in Figure s (7.a and 7.b). After that, various image
processing techniques were applied to the MRI brain image in order to cut the
tumor image from the original image of the brain. The images are converted into a
binary image with a threshold as shown in Figure 7.c. And then converted into a
labeled image (Figure 7.d) that returns the label matrix containing the labels of
the connected objects present in the binary image. The area and stiffness of each
label in the image is then calculated. The area and stiffness of each label in the
image is then calculated. These followed steps are repeated with morphological
functions such as dilation to obtain the segmented tumor image as shown in
Figure 7.
 Nashwan A. Al-khulaidi et al.

4. Results
In this section, we will present the results obtained during the process of
recognizing brain tumors using the proposed networks (SVM, MLPNN and
RBFNN). To evaluate the performance of the proposed networks, three measures
were used: sensitivity, specificity, and accuracy.

a) Image with median filter b) Image after Median and sobel

filter

c) Binary image d) Labelled image

e) Segmented tumor
Figure 7. Segmentation algorithm

71
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

A. Support Vector Machine

In this part, we will show the results obtained after executing the SVM
classifier where we split the dataset into two parts, 450 MRI images for the
training phase and 240 for the testing phase. Nine texture features of the GLCM
algorithm were extracted after radon transformation of images and passed to the
SVM classifier with different kernel functions (Linear, RBF and quadratic) which
gave different results and the best performance of quadratic Kernel was achieved
(sensitivity, specificity and accuracy is 98.45, 98.97 and 98.71 respectively) as
shown in table 2.
Table 2. Classification by SVM with different kernel functions.
Kernel Sensitivity Specificity Accuracy
functions

linear 95.42 97.70 96.57

RBF 96.67 98.33 97.5

Quadratic 98.45 98.97 98.71

B. Multilayer Perceptron Neural Network

For multiple neural networks, performance was measured with different number
of neurons in the hidden layer. The results obtained by changing the number of
neurons were different and the best network performance (sensitivity, specificity
and accuracy) was (97.92, 98.95 and 98.44, respectively) at 8 or 9 neurons in the
hidden layer as shown in Figure 8 and table 3.
 Nashwan A. Al-khulaidi et al.

100
95
90
Sensitivity
85
80 Specificity
75
70
65
1 2 3 4 5 6 7 8 9
Number of neurons in the hidden layer

Figure 8. Relation between number of neurons in the hidden layer and

sensitivity and specificity.

Table 3. Classification by MLPNN

Number of neurons Sensitivity Specificity Accuracy
in the hidden layer

8-9 97.92 98.95 44.99

C. Radial Basis Function Neural Network

Finally, the results obtained from the radial-based neural network, where the
performance was measured with the values of different spread parameters, and we
got the best value for performance at the spread value of 21, where it achieved
(97.5, 98.75, 97.5) for the performance measures (sensitivity, specificity, and
accuracy, respectively) as shown in Figure 9 and table 4.

73
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

100
95
90
85
80
75 Sensitivity
70
Specificity
65
60
9 10 11 12 13 14

Values of spread parameter

Figure 9. Relation between values of spread parameter and sensitivity and
specificity.

Table 4. Classification by RBFNN.

Spread Sensitivity Specificity Accuracy

12 97.5 98.75 97.5

Table 5 presents the performance measures results for the proposed networks
(SVM, MLPNN and RBFNN) to recognize brain tumors, and the best results
obtained by SVM with quadratic Kernel functions were 98.45, 98.97 and 98.71
for sensitivity, specificity, and accuracy, respectively.
Table 5. The performance of the proposed models.
Classifier Parameters Sensitivity% Specificity% Accuracy%
linear 95.42 97.70 96.57
SVM RBF 96.67 98.33 97.5
Quadratic 98.45 98.97 98.71
MLP The number of 97.92 98.95 98.44
neurons in the
hidden layers (8-
9)
RBFNN Spread 97.5 98.75 97.5
parameters (12)
 Nashwan A. Al-khulaidi et al.

5. Conclusions
In this paper, brain tumors were recognized and classified into three classes
(normal, glioma and meningioma) by the proposed networks (SVM, MLPNN and
RBFNN) after passing them nine features that were extracted from the Gray Level
Co-Occurrence Matrix algorithm after radon transformation of the images, the
brain tumor was extracted from the magnetic resonance image and the
segmentation algorithm and morphology processes were applied to the images.
The performance metrics of the proposed networks (SVM, MLPNN and RBFNN)
were compared for the best algorithm to recognized and classify brain tumors
based on accuracy, specificity, and sensitivity. The results showed that SVM
using quadratic Kernel functions has better performance than MLPNN and
RNFNN as it achieved 98.45, 98.97 and 98.71 for sensitivity, specificity, and
accuracy, respectively.

6. References
[1] Koo, Y. E. L., Reddy, G. R., Bhojani, M., Schneider, R., Ross, B.D., Rehemtulla, A.,
Philbert, M. A., and R. Kopelman. (2006). Brain Cancer Diagnosis and Therapy with
Nano-platforms. Advanced drug delivery reviews, 58(14), 1556-1577.
[2] M.A. Hayat. (2016). Brain metastases from primary tumors. Epidemiology, biology,
and therapy of melanoma and other cancers. Academic Press is an imprint of Elsevier.
[3] Jain, S. (2013). Brain cancer classification using GLCM based feature extraction in
artificial neural network. International Journal of Computer Science & Engineering
Technology, 4(7), 966-970.
[4] Meghana Chougule. (2020). Neuropathology of Brain Tumors with Radiologic
Correlates. Springer Nature Singapore Pte Ltd.
[5] Kharrat A, Gasmi K, Messaoud M B, Benamrane N, Abid M. (2010). A hybrid
approach for automatic classification of brain MRI using genetic algorithm and support
vector machine. Leonardo J Sci;17(1):71–82.
[6] R. Preetha, G. R. Suresh. (2014). “Performance Analysis of Fuzzy C Means Algorithm
in Automated Detection of Brain Tumor”,IEEE CPS, WCCCT.
[7] R. B. Dubey, M. Hanmandlu, Shantaram Vasikarla.( 2011). “Evaluation of three
methods for MRI brain tumor segmentation”, IEEE computer society, ITNG.92
[8] Manoj K Kowear and Sourabh Yadev,( 2012). “Brain tumor detection and segmentation
using histogram thresholding”, International Journal of engineering and Advanced
Technology.

75
 Al-Mansour Journal/ Issue (53) 2021 )53( ‫ العدد‬/‫مجلة المنصور‬

[9] Vinay Parmeshwarappa, Nandish S. (2014). “A segmented morphological approach to

detect tumor in brain images”, IJARCSSE, ISSN: 2277 128X , volume 4, issue 1,
January.
[10] Bahadure N, Kumar Ray A, Pal Thethi H. (2017). Image analysis for MRI based brain
tumor detection and feature extraction using biologically inspired BWT and SVM. Int J
Biomed Imaging; 2017:1–12.
[11] WADHWA, Anjali; BHARDWAJ, Anuj; VERMA, Vivek Singh. (2019). A review on
brain tumor segmentation of MRI images. Magnetic resonance imaging.
[12] Wijang Widhiarso, Yohannes, Cendy Prakarsah.( 2018). Brain Tumor Classification
Using Gray Level Co-occurrence Matrix and Convolutional Neural Network.
Indonesian Journal of Electronics and Instrumentation Systems (IJEIS) Vol.8, No.2.
[13] Megha Kadam, Prof.Avinash Dhole. (2017). Brain Tumor Detection using GLCM with
the help of KSVN. International Journal of Engineering and Technical Research
(IJETR). ISSN: 2321-0869.
[14] Megha Kadam, Prof.Avinash Dhole. (2018). Brain Tumor Detection using PCA and
NN with GLCM. International Journal for Research in Applied Science & Engineering
Technology (IJRASET) ISSN: 2321-9653. Volume 6 Issue IX.
[15] Himaja Byale, Dr Lingaraju G M and Shekar Sivasubramanian. (2018). Automatic
Segmentation and Classification of Brain Tumor using Machine Learning Techniques.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13,
Number 14.pp. 11686-11692
[16] N. Varuna Shree, T. N. R. Kumar. (2018). Identification and classification of brain
tumor MRI images with feature extraction using DWT and probabilistic neural network.
Brain Informatics.
[17] R. Manimegala, K. Priya, S. Ranjana. (2020). Brain cancer classification using artificial
neural network. European Journal of Molecular & Clinical Medicine. ISSN 2515-8260
Volume 07, Issue 09.
[18] K. Kalaiselvi, C. Karthikeyan, M. Shenbaga Devi, C. Kalpana. (2020). Improved
Classification of Brain Tumor in MR Images using RNN Classification Framework.
International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-9 Issue-3.
[19] Alpana Jijja, Dinesh Rai. (2020). Segmentation of Brain Tumor using Glcm and
Discrete Wavelet Transform. International Journal of Innovative Technology and
Exploring Engineering (IJITEE) ISSN: 2278-3075, Volume-9 Issue-6.
[20] Pijush, S. Sanjiban Sekhar and Valentina E. Balas.( 2017). Handbook of Neural
Computation. / S. Pijush, S. Sanjiban, E. Valentina.1st Edition.Elsevire. – 658 p.
[21] Anuradha Srinivasaraghavan , Vincy Joseph. (2019). Machine Learning Paperback.
Wiley –328 pages.
[22] Charu C. Aggarwal. (2019). Neural Networks and Deep Learning: A Textbook.
Springer.
[23] Seyedali Mirjalili, Hossam Faris, Ibrahim Aljarah.( 2019). Evolutionary Machine
Learning Techniques Algorithms and Application. Springer.
 ‫‪Nashwan A. Al-khulaidi et al.‬‬

‫التعرف على أورام المخ باستخدام تحويل الرادون وخوارزمية مصفوفة التواجد‬
‫المشترك ذات المستوى الرمادي‬
‫‪1‬‬ ‫‪2‬‬
‫د‪ .‬نشوان امين الخليدي‬ ‫أ‪.‬د‪ .‬محمد هاشم المريش‬
‫‪[email protected]‬‬ ‫‪[email protected]‬‬

‫‪3‬محمد عصام علي‬ ‫‪3‬ابوبكر علي غالب‬

‫‪[email protected]‬‬ ‫‪[email protected]‬‬

‫‪3‬عمر عبدالعزيز النجيب‬ ‫‪3‬وجدان محمد سعيد‬

‫‪[email protected]‬‬ ‫‪[email protected]‬‬

‫ان‬ ‫المستخلص‪ :‬حظي مجال لتعرا ع ى او ام لل لتفا لاي لت األخلر لباها ا لاعفالل مللاا ماح يناف لتنالحلهح لاي اا ل لتفجال‬
‫لالنعشل لت يا ت ف ض‪ .‬لي ا ل لتنحث‪ ،‬تم لتعرا ع ى او ام لل لتاغمل (طنهرها‪ ،،‬م ل ب ياي‪ ،‬مم ل لاحل ي) ماح اا لتشان لر‬
‫لتفيع ح‪( ،‬آت‪ ،‬لتفعجهلر لتغلىف‪ ،،‬ملتشان ‪ ،‬لتربانه‪ ،‬معراغبا لتبنيالر ‪ Perceptron‬ملتشان ‪ ،‬لتربانه‪ ،‬ت خفه ا‪ ،‬لتشارلىه‪ .)،‬تألي ام‬
‫مجفخىاا‪ ،‬لتنهلناالر جتااو ا ايح‪ 954 :‬صااخ ا االت نهح لتفيأللطه ااي تف ح اا‪ ،‬لتعااغ ي م‪ 254‬صااخ ا تف ح اا‪ ،‬لالاعناال ‪ .‬تاام تبنه ا‬
‫لتفرلتج‪ ،‬لتف ني‪ ،‬ى او لتباخ لتفغا ا‪ ،‬ىاح ط يا تحخياف لتباخ جتاو ب الر ملبيا‪ ،‬مج لتا‪ ،‬لتضخضال ى ههال خللاب‪ ،‬م ا‬
‫معخلط‪ .‬رغ ذتك‪ ،‬تم للعخ لج لفلر ن هج صخ لتعبخي الت نهح لتفيأللطه اي للاعخغلل اخل مها‪ ،‬مبا خل‪ ،‬لتعخل اغ لتفشاع‬
‫ذلر لتف عخى لت ملبي رغ تحخيف لت لبمن‪ .‬تم تف ي مه لر لتأل هج لتف عخ ج جتو لتشن لر لتفيع حا‪ ،‬تعباألها لبم لل ملتعرا ع‬
‫ى ههل‪ .‬تفت جضلل‪ ،‬تيأله‪ ،‬للعخ ص ام لل لتف مح صخ ا لت نهح لتفيأللطه ي خللب‪ ،‬اخل مها‪ ،‬لتعج ا‪ ،‬مىف هالر لتعشا ف ى او‬
‫لتبخ ‪ .‬تم يهلس ابل آت‪ ،‬لتفعجهلر لتغلىف‪ ،،‬ملتشن ‪ ،‬لتربنه‪ ،‬معرغبا لتبنيلر ‪ Perceptron‬ملتشن ‪ ،‬لتربانه‪ ،‬ت خفه ا‪ ،‬لتشارلىه‪،‬‬
‫ألاال ع ى ااو لتح للااه‪ ،‬ملتألخىهاا‪ ،‬ملتغياا‪ ،‬مالضااف لتألعاال ج لتعااي تاام لتحبااخ ى ههاال خللااب‪ SVM ،‬مااا مفه اا‪ ،‬لتألااخلا لتع هرهاا‪ ،‬لناات‬
‫‪ 44.95‬م‪ 44.49‬م ‪ 44.92‬ت ح لله‪ ،‬ملتألخىه‪ ،،‬ملتغي‪ ،،‬ى و لتعخلتي‪.‬‬

‫الكلمات المفتاحية‪ :‬لتشن ‪ ،‬لتربنه‪ ،‬معرغبا لتبنيلر لإلب ل ه‪ ،،‬آت‪ ،‬نلي ر لتغىم‪ ،‬مفه ‪ ،‬لبللس لتشرلىي ت شن ‪ ،‬لتربنه‪،،‬‬
‫مب خل‪ ،‬لتعخل غ لتفشع ذلر لتف عخى لت ملبي‪ ،‬تحخيف لت لبمن‪ ،‬لتعبخي لت نهح لتفيأللطه ي‪.‬‬

‫‪1‬‬
‫دكتور؛ قسم هندسة الميكاترونيكس والربوتات‪,‬كلية الهندسة وتقنية المعلومات‪,‬جامعة تعز‪,‬تعز‪,‬اليمن‬
‫‪2‬‬
‫استاذ دكتور؛ قسم هندسة االتصاالت والحاسوب‪ ,‬كلية الهندسة وتقنية المعلومات‪,‬جامعة تعز‪,‬تعز‪,‬اليمن‬
‫‪3‬‬
‫مهندس؛ قسم هندسة البرمجيات‪,‬كلية الهندسة وتقنية المعلومات‪,‬جامعة تعز‪,‬تعز اليمن‬

‫‪77‬‬