Data Enhancement for Date Fruit Classification Using DCGAN

Date fruits are considered essential food and the most important agricultural crop in Saudi Arabia. Where Saudi Arabia produces many types of dates per year. Collecting large data for date fruits is a di cult task and consumed time, besides some of the data types are seasonal. Wherein the convolutional neural networks (CNN) model needs large datasets to achieve high classi cation accuracy and avoid the over tting problem. In this paper, an augmented date fruits dataset was developed using deep convolutional generative adversarial networks techniques (DCGAN). The dataset contains 600 images for three varieties of dates (Sukkari, Suggai, and Ajwa). The performance of DCGAN was evaluated using Keras and MobileNet models. An extensive simulation shows the classi cation using DCGAN with the MobileNet model achieved 88% of accuracy. Whilst 44% for the Keras. Besides, MobileNet achieved better classi cation in the original dataset.