Comparison of OMN and RGO Parameters in OBIA Segmentation for Land Cover and Land Use Classification Using Random Forest in Sumedang Regency

Muhammad Ardiansyah (Departemen Ilmu Tanah dan Sumberdaya Lahan, Fakultas Pertanian, IPB Uninersity)
Boedi Tjahjono (Departemen Ilmu Tanah dan Sumberdaya Lahan, Fakultas Pertanian, IPB University)
Niken Dwia Oktaviani (Departemen Ilmu Tanah dan Sumberdaya Lahan, Fakultas Pertanian, IPB Uninersity)

Jurnal Ilmu Tanah dan Lingkungan
Vol 28 No 1 (2026): Jurnal Ilmu Tanah dan Lingkungan

Land use/land cover (LULC) information is essential for environmental analysis, yet its accuracy depends on segmentation or classification methods. This study aims to analyze differences in segmentation results and classification accuracy using Original Multiresolution (OMN) and Region Grow on Object (RGO) integrated with the K-Nearest Neighbor (K-NN) method in Sumedang Regency. An Object-Based Image Analysis (OBIA) approach was applied with variations in scale, shape, and compactness, and accuracy was assessed using Overall Accuracy (OA), User’s Accuracy (UA), and Producer’s Accuracy (PA). The results show that OMN is sensitive to the scale parameter and tends to produce over-segmentation, while RGO generates larger, more homogeneous, and more stable objects. RGO-based classification achieved more consistent performance with slightly higher then OA, particularly for homogeneous classes such as forest and water bodies. In contrast, OMN captures finer spatial details but is prone to fragmentation. Therefore, RGO with moderate parameters (scale 0.5; shape 0.1 and compactness 0.4) is considered the optimal configuration. However, uncertainties remain in spectrally similar classes, suggesting the need for more adaptive classification methods or higher-resolution data in future studies.