A Hybrid Deep Learning Spectral–Spatial Graph Neural Network with Harris Hawk and Levy-Flight Optimization for Robust Hyperspectral Image Classification

Kamala R

Hyperspectral image (HSI) classification is a crucial task in many remote sensing applications including environmental monitoring, precision agriculture, mineral exploration, and land-cover mapping. However, the high dimensionality of hyperspectral data, spectral redundancy, noise contamination and complex spatial variations often pose challenges to the classification process and can significantly reduce classification accuracy. To overcome these problems, this paper proposes a Hybrid Deep Learning Spectral- Spatial Graph Neural Network(HSS-GNN-HHLF) optimized by Harris Hawk and Levy Flight Strategy. The proposed framework integrates spectral feature reduction, intrinsic spatial reflectance extraction, spectral-spatial feature fusion, graph-based feature learning and adaptive metaheuristic optimization into a unified model. First, spectral redundancy is reduced through spectral fusion and intrinsic decomposition techniques. Then the spatial affinity modeling is adopted to preserve the texture continuity and relieve the illumination distortions. The improved spectral and spatial features are fused and represented as graph structures to capture the nonlinear neighborhood relationships. Finally, the Levy Flight search based Harris Hawk Optimization is employed to adaptively tune the graph-learning parameters, thereby enhancing the convergence and classification performance.