<![CDATA[Soil organic carbon is an important indicator of soil fertility and health, particularly in oil palm plantations. Conventional laboratory methods for organic carbon measurement are costly, time-consuming, and less practical for routine monitoring. This study aims to develop a digital image-based classification model for organic carbon levels using the Support Vector Machine (SVM) algorithm. A total of 96 soil images from the ITSI practice plantation were collected from three soil layers up to a depth of 60 cm. Feature extraction was performed using HSV Color Moment and Gray Level Co-occurrence Matrix (GLCM), producing 15 features per image. The SVM model with RBF kernel was optimized using GridSearchCV, while class imbalance was handled using SMOTE. Experimental results showed an overall accuracy of 60%, precision of 68.57%, recall of 60%, and F1-score of 58.53%. The highest accuracy was obtained in the 0–20 cm and 20–40 cm layers at 57.14%, while the 40–60 cm layer achieved the lowest accuracy at 42.86%. The results indicate that deeper soil layers have more similar visual characteristics, making classification more difficult. This study demonstrates the potential of combining SVM and digital image processing as a low-cost and environmentally friendly alternative for organic carbon monitoring in oil palm plantations.]]>
Copyrights © 2026
