Classification of Date Fruits for Quality Control Using Deep Learning

Abstract

There is a significant challenge the date fruit industrial sector has to confront since an integral pipeline of classification system is still absent and depending on expert manual work that is laborious, expensive also bias-prone. In this sense, Machine Learning (ML) has brought convenience in the implementation of strategy into agriculture and fruit cultivation. In this study we use ML as an automatic system for date fruits classification, where previously expert human judgement was the core of sorting and grading systems. This research presents a robust model for date classification based on the well-known efficacy of Convolutional Neural Networks (CNNs) and transfer learning techniques for picture classification problems. To train this model, a large dataset including nine different date fruit groups was assembled. To increase accuracy, a number of data preprocessing approaches were used, such as augmentation strategies to improve images, learning rate decay over time, model check pointing, and hybrid weight adjustment combinations. The results show that the proposed model, which is based on the MobileNetV2 architecture, achieves a remarkable 99% accuracy rate. Additionally, a comparison with well-known architectures such as Alex Net, VGG16, InceptionV3, ResNet, and MobileNetV2 architecture that have been used as models shows that the proposed model performs better in terms of accuracy.