Hybrid CNN–MLP Deep Learning Framework for Urban Land Cover Mapping in Heterogeneous Tropical Landscapes

Abstract

Rapid urbanization in sub-Saharan Africa presents significant challenges for sustainable land management, particularly in tropical landscapes where spectral similarities between built-up areas and bare soil often lead to classification inaccuracies. While Convolutional Neural Networks (CNNs) have shown promise in remote sensing, they often struggle to resolve 'spectral confusion', a phenomenon where distinct land covers exhibit identical spectral signatures, because they prioritize spatial feature extraction over the non-linear spectral relationships required for class discrimination.  To overcome these limitations, this study proposed a Hybrid Deep Learning framework that integrates a Multi-Layer Perceptron (MLP) with a CNN, combined with expert-driven spatial feature engineering. Using multi-temporal Landsat imagery (2000–2025) for Uyo Metropolis, Nigeria, the study developed a dual-stream architecture that fuses raw spectral bands with Center-Versus-Neighbors (CVN) texture descriptors and the Normalized Difference Vegetation Index (NDVI). The empirical findings demonstrated that the hybrid framework significantly outperforms traditional baseline classifiers (Random Forest), achieving a validated Overall Accuracy of 96.17% and a Kappa Coefficient of 0.94, calculated across the independent test set. The inclusion of spatial texture successfully resolved spectral confusion, reducing misclassification of "Bare Land" as "Built-up" by 22%. Spatiotemporal analysis revealed a 158.10% increase in the urban footprint over the 25-year period, with the most aggressive expansion (3.95 km²/year) occurring between 2020 and 2025. The study concluded that hybrid architectures are essential for mapping heterogeneous tropical environments. The proposed framework offers a scalable, high-precision, and low-cost tool for monitoring urban sprawl in secondary African cities, providing a robust empirical foundation for achieving Sustainable Development Goal (SDG) 11 targets.