Accurate Land Use/Land Cover (LULC) classification is essential for environmental monitoring, resource management and urban planning. Existing LULC classification methods often fail to effectively model long-range spatial-spectral dependencies and suffer from high computational complexity or limited robustness in heterogeneous landscapes. To address these issues, this study introduces a Dynamic gates-controlled Deep Graph Attention Unfolding Network optimised with the Lyrebird Optimisation Algorithm (DDG-AUN-LOA) for precise LULC mapping using multispectral Sentinel-2 and Landsat-8 imagery. The framework begins with Adaptive Gradient Domain Guided Image Filtering (GDGIF) to enhance spatial-spectral consistency, followed by R2U++, a Multiscale Recurrent Residual U-Net, for extracting rich spectral-spatial features. The aforementioned characteristics have been categorised with the help of the Dynamic gates-controlled Deep Graph Attention Unfolding Network (DDG-AUN). This framework is based on dynamic graph attention to learn context and gates-controlled deep unfolding. The Lyrebird Optimisation Algorithm has been applied to optimise the parameters of the model for effective convergence. The suggested DDG-AUN-LOA has provided an overall accuracy rate of 97.31%, Kappa coefficient of 0.96, and Mean Absolute Error (MAE) of 0.051, which helped classify the developed, forested, desert and water land uses successfully.
Jagadeesh et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: