A Hybrid Pixel–Object–Rule-Based Classification Framework with Stability Maps for Large-Scale LULC Mapping

Hybrid classification approaches, combining pixel-based and object-based classification models, are increasingly being adopted to overcome the inherent limitations of Very High Resolution (VHR) image analysis. This paper proposes a hybrid classification framework that integrates probabilistic pixel-based classification, object-based aggregation, and rule-based refinement to produce GIS-ready Land Use/Land Cover (LULC) maps specifically designed for urban and regional planning. WorldView-2 imagery is first processed using an AdaBoost classifier to derive pixel-level class memberships; these results are subsequently aggregated at the object level following segmentation. Beyond thematic labeling, a Stability Map is introduced to quantify intra-object classification reliability, enabling the spatial identification of unstable or heterogeneous objects. The novelty lies not only in the integration of pixel and object paradigms but also in the operational utility of this stability map. When combined with rule-based reasoning, it provides a decision-oriented GIS product. The results demonstrate superior classification accuracy and enhanced interpretability compared to standard pixel-based or object-based approaches, highlighting the framework's relevance for geospatial data analysis and planning-oriented applications.