Semi-automated object-based image analysis for peri-urban land use/cover classification using WorldView-3 imagery of Mankweng, South Africa.

Peri-urban landscapes in semi-arid regions exhibit high spatial heterogeneity and fragmented land use/cover (LUC) patterns, challenging traditional classification methods. This study introduces a semi-automated, hybrid Object-Based Image Analysis (OBIA) framework for detailed LUC classification using WorldView-3 imagery of Mankweng, South Africa. The novelty of the approach lies in the sequential, order-dependent integration of statistical feature selection methods, specifically Coefficient of Variation (CV), Feature Space Optimization (FSO), Linear Discriminant Analysis (LDA), and Otsu's Multi-Thresholding, where each method resolves a limitation left by the previous step rather than being applied independently. These methods reduce the trial-and-error process in feature selection and rule-set development, enabling finer class separation. The Soil Adjusted Vegetation Index (SAVI) and New Built-up Extraction Index (NBEI), used as spectral indices, and Canny edge detection, used as an image derivative layer, were embedded as supplementary input layers to improve object segmentation across hierarchical levels. A four-level Hierarchical Classification Scheme (HCS) guided segmentation and classification, enabling the differentiation of 38 LUC subclasses, including roof types and diverse vegetation types. Applied to four peri-urban test sites within the study area, the method indirectly revealed settlement structures aligned with varying socio-economic and development stages. Compared to standard OBIA, the workflow improves classification efficiency while maintaining thematic detail without relying on extensive training data. The framework provides transferable rulesets adaptable to other fragmented, data-scarce peri-urban environments. It demonstrates that combining statistical measures with known OBIA-features, reduces user input, supporting a semi-automated approach that is helpful for spatial image analysis and ultimately policy-relevant urban planning.