Multiple Classification of Date Fruit using Transfer Learning

Date fruits are considered one of the healthy and favorite fruits in the MENA region, where millions of tons are collected every year, and more than 76,000 tons are produced in the UAE. There are more than 200 types of date fruit produced around the world, where we can find 150 types produced in the UAE. Each type has 5 different maturity levels; Al-Talae, Al-Khilal, Al-Bisr, Al-Rutab, Al-Tamr. The harvesting of these fruits mostly depends on its maturity level and type, so, the technique used to sort them will significantly affect profit. In this work, we propose a deep learning approach for sorting date fruits depending on their types. The system employs three Convolutional Neural Network (CNN) architectures; ResNet-50, VGG-16, and AlexNet, to estimate the type of dates according to specific features. The data is collected for the most four popular types of dates in UAE; Al-Mejdool, Al-Sukariu, Al-Khalas, and Al-Barahi. The best validation accuracy achieved was 100% for AlexNet, while the best testing accuracy was for VGG16 with 98.33%. This intelligent system can help the farmers in harvesting the date fruits, as it will make the harvesting process faster, easier, and more accurate. Based on our research, the classification of UAE date fruits using CNN have not been proposed before and we propose it in this paper for the first time.