Robust Deep Transfer Models for Fruit and Vegetable Classification: A Step Towards a Sustainable Dietary

Sustainable dietary plays an essential role in protecting the environment to be healthier. Moreover, it protects human life and health in its widest sense. Fruits and vegetables are basic components of sustainable dietary as it is considered one of the main sources of healthy food for humans. The classifications of fruits and vegetables are most helpful for dietary assessment and guidance which will reflect in increasing the awareness of sustainable dietary for consumers. In this chapter, a robust deep transfer model based on deep convolutional neural networks for fruits and vegetable classification is introduced. This presented model is considered a first step to build a useful mobile software application that will help in raising the awareness of sustainable dietary. Three deep transfer models were selected for experiments in this research and they are Alexnet, Squeeznet, and Googlenet. They were selected as they contained a small number of layers which will decrease the computational complexity. The dataset used in this research is FruitVeg-81 which contains 15,737 images. The number of extracted classes from the dataset is 96 class by expanding three layers of classifications from the original dataset. Augmentation technique (rotation) was adopted in this research to reduce the overfitting and increase the number of images to be 11 times larger than the original dataset. The experiment results show that the Googlenet achieves the highest testing accuracy with 99.82%. Moreover, it achieved the highest precision, recall, and F1 performance score if it is compared with other models. Finally, A comparison results were carried out at the end of the research with related work which used the same dataset FruitVeg-81. The presented work achieved a superior result than the related work in terms of testing accuracy.