Akademik Veri Yönetim Sistemi
9th International Symposium on Multidisciplinary Studies and Innovative Technologies, ISMSIT 2025, Ankara, Türkiye, 14 - 16 Kasım 2025, (Tam Metin Bildiri)
Synthetic Aperture Radar (SAR) imaging systems play a critical role in the field of remote sensing due to their ability to provide data independently of weather conditions and capture the structural details of surfaces. However, the presence of speckle noise in SAR images blurs class boundaries and reduces classification accuracy. Therefore, filtering methods that suppress noise while preserving edge integrity are of great importance prior to classification. Beyond speckle suppression, these filters enhance spatial consistency by making neighboring pixels more homogeneous, thereby reducing intraclass variance and facilitating the separation of different land cover types. In recent years, edge-preserving filters such as Bilateral Filtering (BF), Domain Transform (DT), and Fast Guided Smoother (FGS) have gained prominence. Nevertheless, their effects on different machine learning algorithms have not yet been sufficiently investigated. In this study, a systematic analysis was carried out on four benchmark datasets (SFBay_C, SFBay_L, Flevo_C, Flevo_L), employing classical machine learning methods such as Support Vector Machines (SVM), Random Forests (RF), and k-Nearest Neighbors (KNN) as baseline classifiers. Performances before and after filtering were comprehensively compared. The experimental results demonstrate that edge-preserving filters, particularly in SVM-based classifications, provide substantial improvements in overall accuracy (OA). Furthermore, comparison of classification maps with ground truth data revealed that filtering enhances both numerical performance and visual quality. In conclusion, integrating edge-preserving filters with machine learning models in SAR imagery not only improves classification accuracy but also supports the reliable mapping of complex environments.