Analysis of Arabic letters recognition using artificial neural networks

مقدم أطروحة جامعية

al-Shaltouni, Ali Mahmud

مشرف أطروحة جامعية

al-Shalabi, Riyad

أعضاء اللجنة

al-Shaykh, Isam
Kanan, Ghassan Jaddu
al-Qarini, Shihadah

الجامعة

الأكاديمية العربية للعلوم المالية و المصرفية

الكلية

كلية نظم و تكنولوجيا المعلومات

القسم الأكاديمي

قسم نظم المعلومات الحاسوبية

دولة الجامعة

الأردن

الدرجة العلمية

دكتوراه

تاريخ الدرجة العلمية

2010

الملخص الإنجليزي

The Analysis of Arabic letter recognition using Artificial Neural Networks had been affected by many network factors; these factors had many changeable parameters in order to reach to the optimum architecture of pattern recognition for Arabic letters.

The presented work aims to enhance the main factor which affected the network architecture which was the network architecture, also the work presented the best value of other factors which gave the best solution in terms of training ratio, training time, simulation time and number of iteration to meet the network goal and the activation function.

In order for the artificial neural network to be useful, it is necessary to adjust the network factors (number of hidden layers, learning rate, number of neuron in each layer, etc) to reach the optimum network architecture appropriately.

The adjustments of network factors are accomplished by comparative results, along with examples of how the system should function.

The basic idea behind the analytical study is to find the best set of factors based on the statistics of training examples that produces correct input/output/relationship.

In this thesis, I obtained the optimal neural network architecture that can reach the goals in minimum number of training iterations and minimum time of training.

From the experimental results, we can reached the goals by adding a hidden layer to the network, and an optimized solution can be achieved if the number of neurons is equal to the power of 2 of the number of inputs (n2) where n represents the number of input layer.

Splitting the number of neurons in the hidden layer to 2 or more hidden layers make the network not optimal by mean of increasing the number of training iterations and increasing the training time.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• اللسانيات الحاسوبية
• اللغة العربية
• الشبكات العصبية(الحاسبات الإلكترونية)

عدد الصفحات

108

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-307226