Enhancing Quality Control for Pistachio: Siirt vs. Kirmizi Classification via CNN-Based Image Analysis
Robert G. de Luna, Dexter P. Astorga, Trisha M. Celestial, Aira Mae T. Española, Brian Allen Q. Lanting, Danielle M. Mugar, Mateo Ramos, Jenjazel M. Redondo · 2024
Pistachios are highly valued for their flavor, nutrition, and health benefits, with distinct varieties cultivated worldwide. However, accurately distinguishing between visually similar types such as Kirmizi and Siirt pistachios can be challenging. Misclassification can lead to market pricing errors, financial losses, and quality control issues. This study proposes the deep learning techniques through Python Programming to overcome human perception limitations and develop a reliable method for distinguishing between Kirmizi and Siirt pistachios based on visual characteristics. There are 3046 augmented data used and evenly categorized into two classifications. These were utilized in creating the three convolutional neural network (CNN) models with varying architectures. Model 3 achieved the highest performance, achieving hold-out accuracy of 96.05%. The researchers recommend expanding the dataset size, including more pistachio varieties, and exploring other classifications to further improve the model's performance and applicability.