<p>Accurate semantic segmentation of very-high-resolution aerial imagery remains a critical challenge in urban remote sensing. This paper presents a novel, interpretable framework that integrates brightness-preserving contrast enhancement, fuzzy-based texture encoding, and bio-inspired feature optimization to achieve high classification accuracy while maintaining full methodological transparency. The proposed pipeline begins with brightness-preserving dynamic fuzzy histogram equalization (BPDFHE) to enhance local contrast without luminance distortion. Texture features are then extracted using a novel fuzzy local binary pattern (Fuzzy-LBP) operator, which employs fuzzy membership functions to encode structural patterns with improved robustness to noise and illumination variations. Finally, a wild horse optimizer (WHO)-enhanced decision tree (DT) classifier assigns adaptive weights to the 62-dimensional feature vector (comprising 3 color and 59 texture descriptors), maximizing discriminative power through evolutionary search. Evaluated on the ISPRS 2D semantic labeling benchmark (Vaihingen subset) across the six standard semantic classes—impervious surfaces, building, low vegetation, tree, car, and clutter/background—the framework achieves 92.96% overall accuracy and 85.32% mean intersection over union (mIoU), significantly outperforming classical machine learning baselines. Ablation studies confirm the contribution of each component, with Fuzzy-LBP and WHO-based feature weighting providing the most substantial performance gains. The modular and transparent design offers a robust foundation for future extensions to multi-sensor or multi-temporal urban scene analysis tasks. Furthermore, to validate the generalization capability of the framework, it was also evaluated on the Urban-drone dataset, yielding consistent and robust performance (achieving 89.51% accuracy and 73.58% mIoU), thereby confirming its adaptability across diverse urban aerial scenes.</p>