Classification and Analysis of Pistachio Species Through Neural Embedding-Based Feature Extraction and Small-Scale Machine Learning Techniques

Abstract

Pistachios are a tremendous source of fiber, protein, antioxidants, healthy fats, and other minerals like thiamine and vitamin B6. They may help people lose weight, lower cholesterol, and blood sugar levels, lead to better gut, eye, and blood vessel health. The two main varieties farmed and exported in Turkey are kirmizi and siirt pistachios. Understanding how to detect the type of pistachio is essential as it plays an important role in trade. In this study, it is aimed to classify these two types of pistachios and analyze the performance of the various small-scale machine learning algorithms. 2148 sample images for these two kinds of pistachios were considered for this study which includes 1232 of Kirmizi type and 916 of Siirt type. In order to evaluate the model fairly, stratified random sampling is applied on the dataset. For feature extraction, we used deep neural network-based embeddings to acquire the vector representation of images. The classification of pistachio species is then performed using a variety of small-scale machine learning algorithms29,31 that have been trained using these feature vectors. As a result of this study, the success rate obtained from Logistic Regression through features extracted from the penultimate layer of Painters network is 97.20%. The performance of the models was evaluated through Class Accuracy, Precision, Recall, F1 Score, and values of Area under the curve (AUC). The outcomes show that the method suggested in this study may quickly and precisely identify different varieties of pistachios while also meeting agricultural production needs.