Application of Elliptical Fourier Analysis and Color Properties in Hazelnut Classification Using Machine Learning Algorithms

Abstract

The accurate classification of hazelnut cultivars is critical for ensuring product consistency, quality control, and market competitiveness in the food industry. Conventional identification methods remain manual, time-consuming, and error-prone, highlighting the need for automated alternatives. This study presents a novel, real-time machine vision system for classifying 11 hazelnut cultivars using a single side-view image. The proposed approach integrates three complementary feature extraction techniques: Elliptical Fourier Analysis (EFA) for contour and shape decomposition, circular masking for curvature quantification, and brown color gradient analysis for surface tone assessment. The extracted features—fully normalized and dimensionless to account for variations in imaging angle, distance, and nut positioning—were classified using three machine learning algorithms: Support Vector Machine with Radial Basis Function (SVM-RBF), Multilayer Perceptron (MLP), and Extreme Learning Machine (ELM-RBF). Among the classifiers, SVM-RBF achieved the highest performance with an F1-score of 0.92 for multi-view images and 0.89 for side-view only. MLP and ELM-RBF followed with competitive yet slightly lower scores. The system demonstrated high robustness, computational efficiency, and interpretability. Overall, the proposed method offers a lightweight, scalable, and non-destructive solution for hazelnut cultivar classification and demonstrates strong potential for real-time deployment in industrial sorting lines and embedded systems in precision agriculture.