Remotely-sensed phenology pattern regionalization for land cover classification of natural scenes:A case study in China
Selection of appropriate geographic boundaries for zoning the study area or images can effectively improve the accuracy and efficiency of land cover classification by reducing the complexity of land cover within the regions and the variability of its features in the images.At present,the regionalization data used in land cover mapping based on stratified classification strategies,such as ecological regionalization,lack targeted objectives,which limits its applicability and effectiveness in remote sensing classification.Vegetation phenology is the main cause of spectral heterogeneity within the land cover of natural scenes.To address this issue,this study proposed a remotely-sensed phenology pattern regionalization scheme for land cover classification of natural scenes.The regionalization scheme is implemented by constructing a zoning index system using vegetation indices,which reflect the greenness status of vegetation,and key phenological periods,which reflect the growth and development rhythm of vegetation.Small geomorphic regions are used as the zoning units,and a data-driven spatially constrained hierarchical clustering algorithm is employed in the regionalization.The evaluation results based on statistical tests and multi-source land cover products indicate that the remotely-sensed phenology pattern regionalization in this study effectively reduces the complexity of land cover within the region and the intra-class feature heterogeneity caused by vegetation phenology,and shows high potential in constructing representative land cover sample libraries and implementing stratified classification strategies.
万方数据
维普
