Multi-level perception of forest resource information based on UAV hyperspectral remote sensing

Abstract Precise and efficient monitoring of forest resources is crucial for sustainable management. Addressing the limitations of traditional remote sensing technologies in detailed forestry applications–such as insufficient spatial resolution and inadequate spectral information-this study systematically explores the potential of UAV-based hyperspectral remote sensing as an integrated solution. A multi-level analytical framework spanning macro to micro scales was established: First, land use classification at the study area level was achieved using clustering algorithms based on spectral-spatial characteristics, generating land cover maps encompassing water bodies, bare soil, cropland, and diverse forest types. Subsequently, at the plot level, Support Vector Machine (SVM) classification algorithms were employed for detailed classification and mapping of representative tree species. Finally, a forest health assessment model was developed by integrating vegetation indices sensitive to physiological status–including greenness, chlorophyll content, and canopy water content–to generate health condition maps. The results indicate that the ISODATA-based land use classification achieved an overall accuracy of 67.44%, yet effectively extracted forest tree information. The SVM method demonstrated excellent performance in the detailed identification of tree species across different sample plots, with overall accuracies ranging from 75 to 90.85%. The tree health assessment results obtained using the multi-index combination approach were consistent with the image features, providing a reliable method for the detailed monitoring of forest resources.