Classification of Dates Quality Using Deep Learning Technology Based on Captured Images

Abstract

Dates are a common fruit in many Middle Eastern and African nations and have religious and cultural value. one of the key elements in judging the quality of dates is sorting according to their health state. Combining rejected dates with accepted ones causes significant economic losses in both storage and exportation. Despite being a crucial stage for obtaining highquality dates and reducing losses, this sorting process is still conducted using traditional methods. Thus, this study aims to classify date fruit quality (accepted or rejected) with machine learning technology to reduce cost, time, and improve the quality of final product. In this study, several Convolutional Neural Network architectures VGG19) were used to classify three varieties of date fruit (Mejdool, Saiedi, El-Wadi). These varieties were classified into accepted and rejected samples to build the dataset im-ages. An Arduino Automatic mobile camera shutter controller captured the dataset images. In addition to the Kaggle dataset which was added to the accepted images. The total dataset consisted of 5,945 images, comprising 3,142 accepted images and 2,803 rejected images. By comparing the results of different architectures, Inception-ResNet-v2 demonstrated the best performance, achieving an accuracy of 98.99% and a loss of 0.0344. Therefore, it can be concluded that the Inception-ResNet-v2 model could be utilized to develop a suitable computer vision system, thereby enhancing the date sorting process and facilitating the packaging of high-quality dates.