Classification of Pistachio Species based on Deep Learning Models

Abstract

Pistachios are highly nutritious and have a large market presence. Since different species vary in quality and can be sold at different prices, it is crucial to classify prominent species accurately. Additionally, given the large number of species, improving the speed of species identification is also important. However, traditional methods for identifying pistachio species can be challenging in both accuracy and efficiency. This study examines the utilization of Deep Learning technologies to automatically classify the two pistachio species from Turkey: Kirmizi and Siirt. The performance of Convolutional Neural Networks (CNNs) is compared to that of Vision Transformers. Evaluated CNN models include ResNet and EfficientNet, which were trained and tested using both actual and augmented images. The experimental results show that ResNet achieves the best accuracy and fastest inference time, the peak accuracy is 99.30%. Data augmentation improves performance but increases inference time. The research findings emphasize the potential of Deep Learning for improving pistachio species classification.