Investigation the success of different classification approaches and different classification algorithms in mapping vineyards

Mapping and monitoring of agricultural products is important for reasons such as yield determination, farmer inspection, product inspection. Today, different information extraction methods such as satellite image classification are successfully applied in the mapping of agricultural products. In this study, the success of the images obtained from PlanetScope satellite systems in mapping vineyards was investigated. As the study area, a region where vineyards were intense in Saruhanlı district of Manisa province was selected. Object-based and pixel-based classification approaches were used for the mapping process. Maximum Likelihood (ML), Random Forest (RF), Support Vector Machine (SVM) algorithms are used in the pixel-based classification process. In the object-based classification approach, SVM, RF, Naive Bayes (NB), K-Nearest Neighborhood (K-NN) and Decision Tree (DT) algorithms were used. The RF algorithm was found to be the most successful algorithm in both pixel and object-based classification approaches in the mapping of vineyards. In the pixel-based classification with the RF algorithm, 87% overall accuracy, 0.83 kappa value and 85% producer and user accuracy in vineyards were obtained. In the object-based classification process with the algorithm; overall accuracy, kappa, user accuracy, and producer accuracy were determined as 91%, 0.87%, 88%, and 92%, respectively. When the generated maps were examined, the object-based classification approach has eliminated the false pixel classification errors such as the salt-pepper effect in pixel-based classification for vineyards parcels.