Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Oman → Classification of Omani’s Dates Varieties Using Artificial Intelligence Techniques

Sultan Qaboos University (2019)

Classification of Omani’s Dates Varieties Using Artificial Intelligence Techniques

Salima Al Abri

Titre : Classification of Omani’s Dates Varieties Using Artificial Intelligence Techniques

Auteur : Salima Al Abri

Université de soutenance : Sultan Qaboos University

Grade : Master of Science (MS) 2019

Résumé
Date fruits are considered as one of the most popular fruits in the Middle East. Oman is one of the countries that has many varieties of dates and the most well-known dates are Khalas, Fardh, and Khunaizi. Nowadays, human workers do the process of classifying different varieties of dates in date’s industries manually. The manual process affects the quality of dates and takes a long time to be accomplished. The objective of this research was to classify six different varieties of dates : Khalas, Khunaizi, Fardh, Qash, Naghal, and Maan automatically from their images based on color, shape-size, and texture features. Different artificial intelligence techniques, Artificial Neural Network (ANN), Support Vector Machine (SVM), and K-Nearest Neighbor algorithm (KNN) were used for automatic classification process and qualitative comparison. This project proposed an iterative approach to optimize the number of neurons in ANN, the number of nearest neighbors in KNN, and the box constraint in SVM. In addition, these three optimized networks were utilized to classify six different varieties of dates. The Dates’ varieties were obtained from AL-Dhahira Governorate. In total, 600 date samples (100 dates /class) were used. MATLAB Toolboxes were used to process the date’s images, extract the features for the different varieties and classify them into six different classes. The accuracy results show that the combination of color and shape-size features outperforms the texture feature when used with the three classifiers. The results show that ANN with only one hidden layer containing seven tan-sigmoid neurons can perform better than both SVM and KNN in terms of accuracy, average recall, and average precision. The achieved classification accuracy reaches up to 99.2% for ANN, 97.5% for SVM, and 95.8% for KNN. For average recall, 99.12% for ANN, 97.53% for SVM and 95.7% for KNN were obtained. In terms of average precision, 99.25%, 97.72%, and 96.18% were achieved by ANN, SVM and KNN respectively

[ Présentation

Page publiée le 3 mai 2021