20.500.12556/RUL-190008 Confidence-driven background selection for presence and background learning in remote sensing water body mapping Izbor ozadja na podlagi zaupanja za učenje prisotnosti pri kartiranju vodnih teles z daljinskimi raziskavami Metoda učenja prisotnosti in ozadja (PBL) je učinkovita pri polnadzorovanem učenju z omejenim številom označenih podatkov, zlasti pri daljinskem zaznavanju. Vendar lahko naključno izbiranje vzorcev ozadja povzroči vključitev pikslov ciljanega razreda, kar poslabša učenje modela. V prispevku predstavljamo izboljšano metodo PBL, ki naključno vzorčenje nadomešča z izbiro na podlagi zaupanja v napovedi modela. Ta se najprej nauči prepoznavati označene pozitivne vzorce, nato pa z oceno verjetnosti prisotnosti izbere neoznačene vzorce, ki z majhno verjetnostjo pripadajo ciljnemu razredu. Tako oblikovano ozadje omogoča zanesljivejše učenje. Metodo smo preizkusili pri zaznavanju vodnih teles na satelitskih posnetkih Sentinel-2 ekološko raznolikega parka El Kala v Alžiriji. Rezultati kažejo, da predlagani pristop pri vseh uporabljenih metrikah, vključno s koeficientom Kappa in mero F1, presega osnovno metodo PBL. The Presence and Background Learning (PBL) framework has demonstrated a robust performance for Semi-Supervised Learning (SSL) techniques in remote sensing, particularly in case of the scarce labelled data. However, it’s a conventional reliance on random sampling to form the background set introduces a critical vulnerability: random selection may inadvertently include pixels from the target class, thereby corrupting the background distribution and degrading the model adjustment. In our paper, we propose a novel modified PBL algorithm that replaces the random background sampling with a confidence-driven pseudo-label selection strategy. Precisely, our naïve PBL model is first trained on a small set of labelled positive samples and a preliminary set of unlabelled samples; its predicted probability scores are then used to filter the full unlabelled pool; only samples with a low presence probability, those unlikely to belong to the target class, are retained as the background. This produces a cleaner, more semantically coherent background distribution, thereby improving the model training. Applied to a water body extraction from a Sentinel-2 imagery over the ecologically complex of the El Kala National Park (Algeria), our method consistently outperforms the naïve PBL baseline across all evaluation metrics, including the Kappa coefficient and F1-score. Presence and Background Learning (PBL) semi-supervised learning self-training pseudo labelling water body extraction remote sensing images Sentinel-2 true false true Slovenski jezik Angleški jezik Članek v reviji 2026-10-20 10:47:03 2026-10-20 10:47:11 2026-10-21 04:19:40 0000-00-00 00:00:00 2026 0 0 Str. 206-215 št. 4 Letn. 93 2026 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 2026-01-01 004.8 0013-5852 301650179 RAZ_Attaf_Dalila_2026.pdf RAZ_Attaf_Dalila_2026.pdf 1 0EEEF7B0FF4E3B6EE76742A02D970BD7 bc67c83b1450efc1bee823aede42442d68a1a24632b652dcfdcf48023aad0553 aa201b8f-cc62-11f1-8bc5-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=250979 https://ev.fe.uni-lj.si/4-2026/Attaf.pdf 0 bd6e71ad-cc62-11f1-8bc5-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=250980 Fakulteta za elektrotehniko Elektrotehniški vestnik 0 0 0