Object-Based LULC Classification Framework in Heterogeneous Landscapes: A Comparative Case Study of Urban vs. Wetland Areas

This dataset was collected and generated to support the study of fine-grained land use and land cover (LULC) classification in heterogeneous landscapes. The experimental efforts focused on a comparative analysis between a spatially continuous natural wetland (South Dongting Lake) and a highly fragmented urban center (Xiangjiang New Area). The repository contains the following core components: Remote Sensing Data: Processed high-spatial-resolution imagery patches derived from the Gaofen-2 (GF-2) satellite. Field Survey Samples: Ground-truth reference data comprising field survey points collected via GPS and mobile GIS, labeled with LULC categories (e.g., vegetation, water bodies, buildings, roads, bare land) Model Code: The source code and associated scripts for the proposed Object-Based Graph Convolutional Network (OB-GCN) framework, implemented in Python using TensorFlow. These materials are provided to demonstrate the efficacy of integrating multi-scale segmentation with graph-based deep learning and to ensure the full reproducibility of the classification experiments presented in the associated manuscript.