Abstract:

Considering the critical importance of pistachio quality and its impact on export markets, the development of an intelligent system for classifying open-shell (Khandan) and closed-shell pistachios is an essential requirement in the pistachio processing industry. In this study, a deep learning–based system was developed to analyze pistachio images and perform highly accurate classification. Given that proper imaging of samples is a crucial part of the processing pipeline, a dedicated pistachio analyzer device was designed and constructed to ensure optimal image acquisition. After image collection, the data underwent preprocessing, feature extraction, and final classification using deep learning models. Three deep learning architectures were investigated and evaluated for classifying pistachio images into three categories: "open-shell," "semi-open-shell," and "closed-shell": a basic Convolutional Neural Network (CNN), a pre-trained GoogleNet model, and an EfficientNet model. Among these, the EfficientNet model achieved the highest overall accuracy of 99.68%, outperforming the other models. The results demonstrate that deep learning–based computer vision methods surpass traditional approaches in both speed and accuracy, providing a viable solution for improving efficiency and reliability in pistachio production lines.