Object-Based Detection of Hazelnut Orchards Using Very High Resolution Aerial Photographs

Hazelnuts are a vital agricultural commodity, contributing significantly to global food systems and public health due to their nutritional value and economic importance as an export crop. Türkiye is a leading global producer of hazelnuts as a major source of income and a strategic agricultural product. In this study, we selected two regions named Açmabaşı and Paralı from Sakarya province which ranks third in the production of hazelnut among Turkish provinces and utilized very high-resolution (VHR) aerial photographs to classify hazelnut fields. In addition to the standard CORINE Land Cover (LC) classes, we defined a specific hazelnut class, verified through field observations. Various machine learning-based classification algorithms, including Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Bayesian classification, were employed with object-based classification with different feature values. The performance of the models was evaluated using overall accuracy and F1-score metrics and the best results are obtained with Support Vector Machines (SVM) with Radial Basis Function (rbf) and Bayes classifier. We obtain $\mathbf{9 6 . 2 0 \%}$ overall accuracy and $\mathbf{9 3 . 4 0 \%}$ F1-Score for Açmabaşı while using Bayes with feature combination as a best result. For Paralı region, the highest overall accuracy is obtained with SVM - rbf using feature combination while $\mathbf{F 1}$-score is the highest for Bayes classifier with $\mathbf{9 0 . 5 7 \%}$.