Classification of Pistachio Varieties Using Pre-Trained Architectures and a Proposed Convolutional Neural Network Model

Abstract

Pistachio is a vital agricultural product native to the Middle East and Central Asia. The world's major pistachio producers, Iran, the USA, Turkey, and Syria, contribute close to 90% of the total production worldwide. In Turkey, there are eight primary domestic pistachio varieties, alongside five foreign varieties. Each produced kind has its unique market and pricing point for consumers to purchase. However, the existing method used to separate pistachio nuts is still carried out with basic knowledge, leading to a significant potential for errors in the classification process due to the virtually identical appearance of each pistachio variety. To address this challenge and enhance the efficiency of the packaging process, innovative technologies are required in the pistachio industry. This study focuses on the classification of three distinct pistachio varieties-Siirt, Tekin, and Uzun-using pre-trained VGG16 and Inception-V3 models, along with a proposed Convolutional Neural Network (CNN) model. The dataset used in the study was divided into three subsets, with 70% allocated for training, 15% for validation, and 15% for testing. Specifically, 1575 images were used for training, and 672 images were allocated for both validation and testing purposes. As a result of the performed classifications, test classification accuracies of 94.05%, 96.13%, and 97.02% were obtained from the pre-trained VGG16, Inception-V3 models, and the proposed CNN model, respectively. The pre-trained models and the proposed CNN model displayed impressive performance, but the proposed CNN model demonstrated a slight advantage with its higher test accuracy score. This suggests that it is more suitable and effective for the classification of the different varieties of pistachio compared to the pre-trained models.