Vegetation ecological feature-aware multimodal network: Fine-scale classification of the Yellow River Delta wetlands using UAV-based hyperspectral and LiDAR data

• A new method is proposed for classifying coastal wetlands with spatially mixed vegetation and heterogeneous spatial patterns. • Incorporating vegetation trait features reduces interspecific misclassification in mixed-vegetation environments of coastal wetlands. • Suaeda glauca was classified with 94.08% accuracy despite its sparse distribution and soil background interference. • A fine-scale classification map of the Yellow River Delta wetlands was generated, with an overall accuracy of 96.70%. Wetland vegetation exhibits considerable diversity, making the accurate classification of these vegetation types essential for the effective management of wetland resources and related research. However, achieving fine-scale vegetation classification mapping in wetlands with complex vegetation through remote sensing technology remains a significant challenge. A Vegetation Ecological Feature-Aware Multimodal Network (VEFAMN) was developed to achieve fine-scale classification of the Yellow River Delta (YRD) wetlands. The network utilizes low-altitude multi-source remote sensing data including hyperspectral imagery (HSI) and light detection and ranging (LiDAR) data acquired by unmanned aerial vehicles (UAVs) as inputs. It successfully integrates three parallel feature extraction modules: the Spectral Feature Extraction Module (SFEM) which enhances the concentration of UAV hyperspectral features; the Morphological Feature Extraction Module (MFEM) which captures the spatial morphological characteristics of land cover; and the Vegetation Trait Feature Extraction Module (VTFEM) which efficiently extracts vegetation physiological and ecological trait features. The study accurately produced a fine-scale classification map of the complex vegetation in the YRD wetland. Experimental results indicate that the method achieved an overall classification accuracy of 96.70%, along with strong performance across multiple standard evaluation metrics, successfully distinguishing between mixed vegetation species, including Phragmites australis, Tamarix chinensis, Shriveled spartina alterniflora , and Suaeda glauca in the wetland. This research offers a solid data foundation and technical support for advanced vegetation monitoring in complex wetland environments.