Classification of Dates (Phoenix Dactylifera L.) Varieties Using Texture Feature Analysis

Abstract

The Date Palm (Phoenix dactylifera L.) is an ancient evergreen fruit tree. It belongs to the palm family Phoenix (Arecaceae). Dates are Pakistan's 3rd most important fruit crop, after mangos and citrus. Almost 325 different varieties of date fruit are grown in several parts of Pakistan. Visual perception makes it difficult to classify dates by looking at their fruits. The best determinant for date classification is the external appearance of the fruits. Their varietal classification is crucial for meeting consumer and market expectations in terms of quality. By automating processes like fruit classification, machine learning (ML) has transformed the fruit-producing business. The excellent performance of CNNs in image categorization has increased productivity and efficiency overall. The goal of the research is to develop an automated framework for the classification of date fruits using the deep learning technique based on texture feature analysis. For evaluation purposes, a date fruit image dataset of 1200 images have developed. Four types of date fruits were chosen for experiments, namely Aseel, Karbala, Muzawati, and Zahedi. Their images were captured, with 80% being utilized for the training dataset and 20% being used for testing. Experiments were carried out on the chosen date fruit varieties by the defined framework. The CNN models Inception V3, VGG16, and VGG19 were used as classifiers. Other ML classifiers like KNN and SVC were also deployed. CNN, on the other hand, outperformed all others with 99.44 percent accuracy across 10 epochs.