Vegetation extraction and classification based on UAV hyperspectral data for the Loess Hilly and Gully region

Vegetation has the ability to regulate the runoff and sediment yield in the underlying surface, thus, watershed vegetation monitoring is of great significance to the effectiveness of soil and water conservation programs in the Loess Plateau watersheds. Taking the typical Chabagou watershed, the first sub-region of the Loess Hilly and Gully Region, as the research area, this research considers the unmanned aerial vehicle (UAV) hyperspectral data as the main data and draws the auxiliary data from the UAV orthophoto model, field surveys, and surface feature spectral measurement, in an attempt for multi-phase aerial surveying and data acquisition. Based on the multi-phase vegetation spectral features extracted, the integrated SVM-SAM method is suggested for vegetation classification. The accuracy evaluation shows that: (1) the UAV-based hyperspectral measurement is capable of generating centimeter-level hyperspectral data of the underlying surface. It is proven applicable in the Loess Plateau; (2) by virtue of a combination of UAV data acquisition, field survey, and surface feature spectral measurement, an elaborate database with high precision for vegetation features is attainable; (3) this integration possesses an overall classification accuracy of 82.82% and a Kappa coefficient of 0.785. (4) the research results can effectively enhance the accuracy of vegetation monitoring, and improve its spatiotemporal scale. It is expected that the research achievements will provide essential data for exploring the soil and water erosion mechanisms in the Loess Plateau and establishing soil erosion models, which demonstrates great significance to the practices of soil and water conservation in the region.