<p>Hyperspectral remote sensing has become an essential tool in environmental monitoring and land use analysis, offering detailed spectral information that enables accurate surface material identification. However, effective classification of hyperspectral images (HSIs) remains challenging due to high dimensionality, spectral redundancy, and class imbalance. To address these issues, we propose KERN-HIC (Knowledge-Embedded Routing Network for HSI Classification), a novel classification model that integrates semantic knowledge with spatial information to enhance the accuracy and contextual understanding of land cover classification. KERN-HIC introduces two key innovations such as a kernel-enhanced relational graph framework that adaptively models spectral-spatial dependencies, and a semantic-aware routing mechanism inspired by capsule networks to hierarchically fuse spectral- and spatial-aware representations. This is the first model to jointly leverage kernel-based graph learning and dynamic semantic routing in a unified HSI classification framework. KERN-HIC is particularly well suited for handling the irregular spatial patterns found in real-world land use datasets and supports scalable environmental monitoring systems. We evaluate the model on benchmark datasets including Kennedy Space Center, Indian Pines, and Pavia University. KERN-HIC demonstrates substantial improvements over state-of-the-art methods, achieving gains of up to 7.23% in overall accuracy, 5.52% in Cohen’s Kappa, and 6.72% in per-class accuracy on the Indian Pines dataset, as well as up to 6.84% and 5.80% gains in overall accuracy on the Kennedy Space Center and Pavia University datasets, respectively. These results highlight KERN-HIC’s effectiveness in improving hyperspectral image classification for applications such as land use monitoring, environmental change detection, and geospatial analysis.</p>